How to use ConformerLab

What ConformerLab does

ConformerLab runs structure-based studies for you. You ask a question in your own words and give it a structure, a protein's name or just a sequence. The AI planner works out the steps with you: predicting a structure from its sequence, finding a pocket and docking a molecule, a peptide or a partner protein into it, screening a library of compounds, molecular dynamics (a simulation of how the molecules move over time), and QM/MM where the chemistry matters. Each step is checked against a fixed set of scientific rules and priced in credits, and nothing runs until you approve the price. Each result can feed the next step.

A simulation comes back as a one-line answer, charts, a playable movie of the molecules, a 2D map of what holds them together, a methods paragraph, a PDF report, and the files to run the same simulation yourself. You do not need to know how to set a simulation up: the planner chooses the settings and says why, and each run's methods name every program and setting for a specialist reader.

What you need

To make an account, press Create account on the front page, or open /signup, and give a Name, an Email and a Password of at least 10 characters. Or press Continue with Google to sign up, and later sign in, with your Google account; if an account already uses that address, Google signs you in to it. Sign in at /login.

A new account gets a 6-digit code by email. Type it into the banner at the top of the app and press Confirm. Until you do, you can plan studies and see their prices, but the plans come from ConformerLab's rules rather than the AI planner, the assistants wait, and you cannot approve a run. A code expires after a while: Send a new code sends another, and only the newest works. Signing up with Google needs no code.

If you forget your password, on the sign-in page press Forgot your password?, enter your email and press Send code. Enter the code from the email with a new password and press Set new password. You are signed in, and every other place you were signed in is signed out. An account made with Google can get a password this way too.

To change your password or end your sessions, press Account in the sidebar, then the Profile tab (see Your account). Under Change password, enter your Current password and a New password, then press Change password: you stay signed in here, and everywhere else is signed out. Sign out everywhere ends every session of the account at once, this one too; use it after signing in on a computer that is not yours.

On the same tab, Download my data saves one file with everything held about the account: your profile, your API keys (never the keys themselves), every credit entry, payments, runs, notices, and your ratings of the assistant's replies. A run's files and report are downloaded from the run's own page. To delete the account, type your email in Your email, to confirm under Delete account, and your Current password if the account has one, then press Delete my account. Everything is erased at once: your profile, studies, runs and their files, notices and conversations with the assistants. It cannot be undone, and credits left are lost with it, so write to us first about unused credits you paid for. Records of payments are kept, without your email address, as tax law requires. While a study is running the account cannot be deleted: stop it or let it finish first.

Compute time is paid for in credits. A new account starts with none. Buy them with Buy credits on the sidebar's credits card (see Credits and costs), or ask the ConformerLab team to add them, by the email you signed up with. Where the site does not sell credits yet, the button is not shown. The sidebar shows what you have.

Planning, checking and pricing a study never charge for compute time. The AI planner's work and the assistants' replies come out of a free allowance of 40 credits a month, then from your balance; see Credits and costs.

Structure files you open are read in your browser. A PDB entry you name is fetched by our server and then read in your browser too; the privacy policy explains how lookups are handled.

Describe your study

The start screen

Signed in, you land on the start screen, headed What do you want to find out?. New study in the sidebar brings you back to it. Type your question in the box labelled Your question and press Start planning, or Enter. Shift+Enter adds a line.

Under the box, Start from what you have offers four ways to bring a structure (see Give it a structure), and Or try one of these offers example questions: clicking one copies it into the box for you to edit, without sending it. At the bottom, Recent studies lists your latest runs with their status and credits.

The menu button beside the logo, at the top of the sidebar, folds the sidebar to a narrow strip of icons and opens it again: + for a new study, a folder for your studies, ? for this guide, a key for API access, your credits and your account. On a phone the sidebar starts hidden, and the menu button at the top left opens it.

Writing the question

Say what you want to learn, in your own words, and anything you have already decided: the structure, if you know its PDB ID, how long to simulate, how many repeats, and what to keep the cost under. Whatever you leave out, the planner chooses and tells you. The examples on the start screen are good models:

• Does the drug in the crystal structure stay in its pocket over 100 ns?
• Is this protein stable over 100 ns?
• Run 50 ns as three repeats, so I can see how much the results vary.
• Quick 10 ns check, I'm in a hurry.
• Is the protein stable without the ligand bound?

A few terms help. A PDB ID is the four-character code of an entry in the Protein Data Bank, such as 1M17. A simulation's length is the time it simulates, in nanoseconds (100 ns; 2 µs is 2,000 ns): a longer run can show slower motions, and costs more. A repeat (a replica) is an independent run of the same simulation: three repeats (three repeats, three replicas, or triplicate) show how much a result varies, one does not. You can also state the salt (150 mM NaCl) or ask for a quick check (in a hurry). If you have a preference for the force field, the physical model of the molecules the simulation uses, and its water model, name them (ff14SB with TIP3P, CHARMM36m); otherwise the planner chooses.

Put everything in one message. Before a plan exists, a new message replaces the request rather than adding to it, except while the AI planner is still working the question out with you (see When your question names no structure).

A question typed on the front page under Try your own before you had an account is kept and appears in the box after you sign up.

Give it a structure

Every study starts from a 3D structure of the molecule. There are six ways to give one, on the start screen or later in the chat.

A PDB ID

Type the ID in the A PDB ID box and press Fetch, or name it in your question (“chain A of 1AKE”). A question that names its entry is planned at once: the chat says back in one sentence what you want to find out, then confirms the entry, for example You named 1AKE, so I'm fetching it from the Protein Data Bank. If the fetch fails, a red notice gives the reason, such as no entry with that ID, and the chat asks where the structure should come from.

Your own file

Press Your own file, or the Structure chip under the question box, and pick a .pdb, .cif, .mmcif, .ent or .pdbx file, including models from AlphaFold Server. In the chat, the upload icon beside the message box (Attach a structure) does the same. A file that cannot be read shows a red chip beginning could not read this file — with what to check, and the chat says the same, with what went wrong under Technical detail.

Drag and drop

Drop a file anywhere on the page: Drop it anywhere — a structure file becomes the protein to study, a file of compounds becomes the molecules. A file of compounds (.sdf, .smi and the like) becomes the study's molecules (see Ligands); anything else is read as the structure.

A protein name

Type a name in the A protein name box and press Look up, choose Look it up by name when the chat asks, or just name the protein in your question (“Does remdesivir bind Dengue virus RdRp?”). The chat asks Do you have a structure in mind? with two buttons: Explore options, which offers I have a PDB ID, Show me the list, Upload a file and Look at the sequences; and one that says how many structures the PDB holds and lists them. Typing an option's words works as pressing it. The lists:

• Structures in the PDB, sharp enough to simulate (3.5 Å or better): experimental structures, best first, each with its title, how it was measured (X-ray, cryo-EM or NMR), resolution (the smaller the number, the sharper the structure), year, the matching chain and the molecules bound in it, each by its name with its code in brackets, such as heme (HEM). Press Use this to take one. If nothing matches: None found sharp enough to simulate (3.5 Å or better).
• Sequences in UniProt: up to six protein entries, each with Structures of this, Similar sequences, Copy and, where it applies, Predicted model.

Once a structure is chosen, the lists fold away; open a heading again to choose another.

A sequence

Paste a protein sequence of at least 10 amino acids, as one-letter codes or FASTA, in the Just a sequence box and press Find structures, or choose Start from a sequence in the chat. The chat lists PDB entries with a chain at least 30% the same as it, with the same Use this buttons, and matching UniProt entries. Similar sequences looks for related proteins among UniProt's curated entries (Swiss-Prot), each shown with how much of it is the same as yours and offered with Structures of this; it usually takes one to five minutes, and you can carry on meanwhile.

An AlphaFold model

Predicted model on a UniProt entry loads that protein's model from AlphaFold DB, marked as predicted, with its confidence. Where there is none, the button reads No predicted model. Where no structure exists anywhere, Predict its structure predicts one from the sequence (see Predicting a structure); where that button is not shown, prediction is not offered on this site yet.

When your question names no structure

A question that names no PDB entry and comes with no file, such as “What does erlotinib do to EGFR?”, is answered first, in words, while a line reads Thinking about your question. The AI planner says what you are asking, gives the context that matters, says what a simulation can and cannot show here, and names up to three proteins or complexes the question points to, each with PDB entries to Use this, or Find structures of this.

The answer ends with a card headed The question to plan: the study it proposes, as one question a simulation can answer, the steps it would take in order (a docking, then a simulation, for example), and the structure it would start from. Nothing is planned until you press Plan this question or type yes. To change it, say so (Or say what to change.); each answer puts the question again as it now stands. Once you confirm, the study goes on as if your question had named its structure. A study whose first step is a structure prediction opens the prediction card instead (see Predicting a structure).

Other ways to start skips the conversation. The AI planner answers in words only once your email address is confirmed; until then the chat asks straight away where the structure should come from. Each answer costs what any reply does (see Credits and costs).

When the chat asks

If planning starts without a structure, the chat asks Where should the structure come from? with four options: Fetch from the PDB, Upload a file, Look it up by name and Start from a sequence. To step back from a choice, press × on its pill (Not that — choose another way) or type never mind.

What is read from the file

When a structure arrives in a study, the chat confirms it, gives a one-line summary such as “2 protein chains · 1 peptide chain · ATP bound · 2 magnesium ions”, and opens How I read this file. Each bound molecule is named in words with its code from the file in brackets, such as heme (HEM): by the name the file gives it, else by the name ConformerLab finds for the code; a code nothing names is “the molecule” and its code.

• A bar for each chain with its kind and length: a chain of 25 or more amino acids is a protein, a shorter one a peptide; DNA and RNA are recognised.
• Chips for the bound molecules and the metal ions. Waters, crystallisation additives, detergents and bulk salt are left out. A structure with nothing bound says Nothing bound — the protein is on its own (an apo structure).
• Rows for Residues, Chains, Bound molecules, Unresolved gaps (stretches missing from the structure), Alternate positions, Metal ions and Predicted model, each with a dot for where the value came from: green, read from the file; indigo, you told me; amber, assumed; a hollow red ring, unknown.

Where a file gives several models, or two positions for some atoms, the first is used. Short gaps are rebuilt; long ones are asked about (see The plan in the chat).

A study is built on a protein, DNA or RNA chain. A file with none (a ligand saved on its own, for example) is refused as it is read: the attachment turns red and says No protein, DNA or RNA chain in it, with what the file holds instead, and nothing is planned or priced from it.

Once you start planning, the study is a workspace. The structure fills most of the window in 3D (drag to turn it, scroll to zoom), with its name, a tag reading PDB entry, Your file or Predicted, and its chains and bound molecules listed as layers you can hide and show. Down its right edge are Reset the view, Surface, Spin, Save a picture and Enlarge (Close or Esc shrinks it again). Under it, Activity lists what was done, in order. The conversation is the panel on the right; drag its left edge to make it wider or narrower. The chat can turn the 3D view to what it talks about: see The plan in the chat.

A peptide that is not in the structure

If your question names a peptide the structure does not contain, it has to be placed on the protein before the complex can be simulated, and the chat asks What should I do with the peptide?: Dock it onto the protein, which opens the docking card with the sequence filled in; Upload a structure with the peptide already bound; simulate the protein on its own for now; or Stop here, which charges nothing.

Changing the structure

Once a structure is loaded, naming another PDB ID in the chat does not switch to it. To change it, upload or drop another file, press Use this on an earlier lookup card, choose the “another PDB entry” option when a decision offers it, or start a New study.

Ligands

Small molecules come into a study in two ways, and they are treated differently.

Molecules already in the structure

Ligands, cofactors and ions in the structure file are kept and simulated with the protein. Common cofactors and metabolites (ATP, NAD, FMN, SAM and others) and single ions are known already and need nothing from you. Metals such as zinc and iron are asked about (see The decisions you may meet), and bulk sodium, potassium and chloride are left out.

Any other molecule, including most drug-like ligands, is looked up by its code in the PDB's dictionary of molecules, which defines every molecule a PDB entry uses. ConformerLab takes its chemistry from there, gives it the charges it carries at the study's pH, and names it on the plan with where its chemistry came from: heme (HEM), from the PDB's definition, charged for pH 7.4. Only a molecule the dictionary does not have (a code a file's authors made up, such as LIG), or one whose atoms in the file do not match its definition, is asked about. Then give its chemical structure as SMILES, a one-line text notation of the molecule (PubChem lists one for most compounds), written in the charge state you want simulated, or run without it, which removes it from the file. What you type is the chemistry that gets simulated; one that cannot be read is asked for again.

Some molecules cannot be simulated: one with a metal bonded into it (heme, chlorophyll, cobalamin, cisplatin), or with boron, silicon or selenium. The chat says so and offers to run without it or to stop. A metal ion the chosen water model cannot represent is refused before you are quoted, with a choice that can (see The decisions you may meet).

One protein force field, CHARMM36m, cannot be combined with a ligand or cofactor kept from the structure (simple ions are fine). A plan that keeps one with it blocks before anything is quoted, and offers to switch to the other standard models, which can keep it (Keep the molecule L26: switch models (ff19SB with OPC water)), or running without the molecule.

Ligand libraries you attach

The Ligands chip under the question box, and drag and drop, take a file of compounds: SDF (.sdf, .sd), or SMILES one per line with an optional name (.smi, .smiles). The first 250 compounds are kept. The chip shows how many, and whether their charges were stated (charges read). In the chat, add a file by dropping it on the page.

A file of compounds says what they are but not where they sit, so the plan cannot place them, and it stops to ask. With several compounds it offers to dock them all (Dock all 12 as a library), which opens the docking card with the file as a library; to dock one into this structure (Dock aspirin into this structure); or Leave the compounds out. Once you pick a docked pose, the attached file is cleared: its compound is now in the structure.

MOL2 and CSV files are not read here (MOL2 files are not read): save them as SDF, or as lines of a SMILES and a name. A library chosen in the docking card can also be a CSV with a SMILES column.

Docking

Docking places something the structure does not contain (a small molecule, a peptide, or a second protein) so that the complex can be simulated. It is a short run of its own, before the simulation. It hands back poses, the ways the molecule can sit in the site, each with a score; you pick one, and that complex becomes the structure the simulation starts from. Each docking's report names the methods it used.

Asking for it

Any of these opens the docking card in the chat:

• a request to dock a molecule the structure does not contain, such as dock aspirin into 5KIR and run 50 ns;
• a request that names a peptide the structure does not contain (the chat offers Dock it onto the protein);
• asking the plan assistant to dock or place a molecule, once a plan exists;
• a compound file attached with the Ligands chip (see Ligands);
• Dock a molecule first, under a plan whose request asked to dock a molecule the file already holds.

The docking card

The card first asks what to dock: A small molecule, A library, A peptide or A protein.

• A small molecule is given by its name or as its chemical structure (a SMILES). A name is looked up in PubChem, and you confirm the compound with Use this.
• A library is a set of molecules, from Choose the file, Paste a link or Use a built-in library (see Where a library comes from). Up to 200 molecules are docked one by one; more are screened (see Screening).
• A peptide is given by its sequence, up to 20 of the 20 standard amino acids: linear, cyclic head to tail, or closed by a disulfide between two cysteines.
• A protein is given as its own structure file, with Choose the partner's file. You can name the partner's amino acids you know touch the protein.

Then choose where it goes. The card opens with the likeliest place already chosen and the others behind Show N other places, so often there is nothing to do. The places are where a molecule in the file already sits; where a bound peptide sits (the docked one replaces it); Where related structures bind; Find pockets in it; Around residues I name, such as Leu718 Thr790; or, for a protein, Anywhere on its surface. You can also describe the site in words under Or say where, such as the NTP entry channel, and press Find it: the AI planner works out which residues line it in your structure and says how sure it is, and when it is not sure it opens the list rather than choosing for you. This costs nothing. Whatever you choose, the 3D view turns to the site at once, so you see it before anything is docked. Docking uses the pH your request names, or 7.4.

Check it is free: it reads what you gave, finds the site and the search area around it, decides what stays in the protein around it (each molecule by its name), and says what it found and the price, with one-click answers to anything it asks. Dock it holds the price and starts the docking; a docking that fails is refunded in full. The card's other buttons plan the structure without the partner, or Stop here.

Finding a site

When nothing in the file marks a site and your request names none, the card looks for pockets in the protein's shape at once, for free; Find pockets in it does the same at any time. Up to six pockets are offered, best first, each with the residues that line it, its volume, and how likely a drug-like molecule is to bind there (its druggability), from 0 to 1. They are candidates, never chosen for you: the known site is not always ranked first, and a site known from a molecule bound in a related structure, or from mutations, is a better guide than shape alone.

Search more thoroughly offers a second, deeper search, suited to peptides as well, for 3 credits, refunded in full if it fails: Run the deeper search or Not now. It looks for places where small test atoms stick to the protein well. Its best six pockets join the list, and its own page lists each with the residues lining it, a score that weighs its size, how enclosed it is and how compact (higher is better), How enclosed (from 0, open, to 1, closed in) and its volume. A pocket both searches find is a stronger candidate than one only one finds. A protein of more than 2,000 residues is too large for the deeper search: use the first, or name the residues.

Sites from related structures

A predicted model has no ligand, metal or cofactor in it: an AlphaFold model of carbonic anhydrase has no zinc, a P450's no heme. Where this site offers it, Where related structures bind asks a public research service that copies ligands, metals and cofactors onto the model from crystal structures of related proteins. ConformerLab keeps only the copies that fit this chain well, offers up to six sites where ligands were seen, each with how many structures put a ligand there and the residues lining it, and places a metal only where the model's own residues can hold it. Choosing a site makes the model, with those molecules in place, the structure to dock into.

For a model from AlphaFold DB, the service's existing entry for that model is used. Any other file has to be sent to the service, which keeps the file and its result and states no retention period, so the card asks first whether to send it, or Don't send it. Nothing is sent until you choose. The card gives the credit the service and AlphaFold ask for.

What is docked of a small molecule

A molecule in the body can carry a different charge from the way it is usually written: at pH 7.4 aspirin has lost a proton and imatinib has gained one. And where a SMILES leaves the 3D arrangement at some atoms open, the molecule can take several shapes (stereoisomers). The check works out each likely form at the pH and docks each one on its own, leaving out arrangements that cannot exist in 3D. Every form is listed with its charge before you pay, each pose says which form it is, and the pose you pick is simulated in its own form. Dock it exactly as written docks your SMILES as you gave it instead. Up to eight forms are docked; a molecule with more is refused until you define its 3D arrangement in the SMILES (with @ and @@).

How the protein is prepared

The protein is prepared as the simulation will start from it: missing atoms are rebuilt, and each residue's charge is set for the chosen pH. Metal ions held by the protein stay. Cofactors whose chemistry is known stay, so poses are scored against them; one that holds a metal, such as a heme or an iron–sulfur cluster, is left out, and the check says so. The molecule being replaced goes, as do crystallisation additives, sugars and waters, except waters that bridge the bound ligand to the protein, which the check names, with Keep these waters one click away.

At a zinc with an open place for a fourth partner, poses are scored with terms made for zinc, at the same price: re-docking acetazolamide into carbonic anhydrase II this way put it back on the zinc, close to its crystal pose. Other metals in the site are kept but scored roughly, and the check says so.

A peptide is docked from its sequence, folding as it docks. The right pose is often not the first, so look further down the list. A protein partner is docked over the whole surface, or around the residues you name. Protein docking often ranks a wrong interface first, so residues known from mutations, cross-links or related complexes help a great deal. The receptor is kept rigid.

The price

A small molecule costs 6 credits, and 1 more for each further form docked; a form whose 3D shape cannot be built is passed over and not charged. A zinc site costs the same. A library costs 6 credits and 0.25 for each search, one per form of each molecule: 100 molecules docked in 150 forms cost 43.5 credits. A thorough screen costs 6 credits, 1 for each part of up to 500 molecules, and 0.10 for each molecule docked: 5,000 molecules in 10 parts cost at most 516 credits. A fast screen costs 6 credits, 5 for each part of up to 25,000 molecules, and 2 for every 1,000 molecules docked: 100,000 molecules in 4 parts cost at most 226 credits. Molecules left out, and parts that do not finish, are not charged. A peptide costs 12 credits, and a protein 30. The check shows the price before anything is held.

Picking a pose

The poses come back best score first, listed beside a 3D view of the one you click. Each has its docking score, which is lower the better for a molecule or a peptide and higher the better for a protein: a rough guide to how well it fits, not how strongly it binds. Each also has its form, its Distance from pose 1 (how far, on average, its atoms sit from the first pose's; under 2 Å is much the same pose), and the residues, cofactors and metals it touches. A peptide's poses also say How often found: how many of the separate searches ended there, since a pose many searches reach is more likely to be right. The up and down arrow keys, or the arrows above the view, step through them. Above the poses of a single molecule are its 2D drawing and its numbers (see Docking a library for what they mean). The best score is often not the right pose: when we docked 16 molecules back into structures where their real pose is known, the right pose was in the top five 13 times but first only 6. Check each pose against what you know about the site before you pick. Under the table, Technical detail names the program and settings that made the scores. Simulate pose makes that complex the structure, and the chat plans it like any other.

The buttons on the view: Reset the view; The residues around it, which shows the residues close to the molecule and names the nearest; Spin; Save a picture; Explore in Mol*, which opens the same complex in Mol*, a full molecular viewer with tools of its own (Back to the simple view returns); and Full screen.

Under the view is a command line: type what you want and press Enter. colour by chain, colour start to end, colour by atom, colour by water-loving, or one colour such as colour teal or colour #4E8FD0; colour molecule orange; draw ribbon, draw backbone or draw atoms; surface protein, surface both 50% or surface off; background white; show interactions, hide hydrophobic (or h-bonds, salt bridges, pi, metal), show residues, hide names, show outlines; look binding site; zoom 140 (or zoom Asn140, zoom A:140) to centre on a residue and reset to come back; spin, picture, full screen, mol* and back. It suggests commands as you type: Tab takes the one highlighted, the up arrow brings back what you ran, help lists them all, and the / key reaches the command line from anywhere on the page.

Style, under the command line, shows the same choices as pictures. Looks are ready-made: Whole protein; Binding site, the residues around the molecule named and what holds it drawn; Surface, the protein's surface, blue-green where it is water-loving and amber where it avoids water, which shows the pocket's shape and what lines it; and For a paper, white and outlined, with names written without boxes. Colour shows each scheme as a strip of its colours: By chain, Start to end, By atom, Water-loving or not and One colour, with wells for the protein's one colour and the molecule's. Draw pictures the protein as a Ribbon, its Backbone or all its Atoms. Surface puts one On the protein or On the molecule, from see-through to solid; Background is the page's, white or black; and Show switches the Interactions, their Distances, Residue names and Outlines. While the interactions are drawn, a legend on the view lists each kind, drawn in its own dash, with how many the pose has; press a kind to hide or show it. Your choices are remembered in this browser for the next pose.

Testing the top poses

For a docked small molecule, Test the top poses lets a short simulation choose instead of the score. Each of the top two to five poses is simulated on its own for 10 ns, and the full simulation then runs on the one that stays in place: a wrong pose tends to drift out of the site, a right one stays. A pose counts as held if it stayed within 2.5 Å of where it was docked and kept at least half of its contacts with the protein. On the plan page you can change how many poses (Poses tested), for how long (Each pose, Settling first) and what counts as held (A pose held if it, Near means within).

• It is approved and paid for as one run. The quote shows the short simulations and the full one as separate lines, and each is charged for what ran.
• When no pose holds, the full simulation is not run and only the short tests are charged: the answer says how it binds is not established, and how close the nearest pose came.
• A pose whose simulation fails drops out; if every pose fails, the run is refunded.

The run page shows each pose as it goes and how each held; the results begin with which pose the full simulation ran on, and why.

Docking a library

• Check it reads every molecule and checks it as a single molecule is checked. One that cannot be docked or simulated is left out with its reason, rather than stopping the rest: an element the simulation does not cover, a peptide, a molecule too large or too flexible, a repeat, a line that is no molecule, or too many forms at the pH. The card lists every molecule and what became of it.
• Each molecule is docked in each of its likely forms at the study's pH; Dock them as written docks each as the file writes it.
• The hits come back ranked by each molecule's best docking score, lower the better, with its Score per atom beside it, higher the better: a bigger molecule scores better for its size alone, and the score per atom evens that out (it is the ligand efficiency). A docking score is a rough guide to how well a molecule fits, not how strongly it binds: use the ranking to choose which molecules to look at, and check each pose before you trust its rank. The best 25 are kept as complexes, ready to see and simulate; Every pose (SDF) has each molecule's best poses, and The ranking (CSV) the table. Under the table, Technical detail names the program and settings that made the scores, and how each number is calculated.
• Each hit is drawn in 2D beside its name, with the numbers a medicinal chemist reads first, each named in plain words with the chemist's own name under it: Size (molecular weight, MW), How oily (cLogP; above 5 it dissolves poorly in water), How polar (polar surface area, TPSA: above 140 Ų a molecule rarely crosses a cell membrane), Drug-likeness (how much it looks like an approved oral drug, 0 to 1, QED) and How hard to make (1 easy to 10 hard, SA). Hover over a column's name for what it means. They describe each molecule uncharged, as a catalogue lists it; the drawing is the form that was docked.
• The Flags column marks Not drug-like (two or more of the four usual limits for a drug taken by mouth broken, Lipinski's rule of five), False-positive risk (a group that often looks active in assays for the wrong reasons, from the PAINS list) and Group to avoid (a reactive, unstable or toxic part, from Brenk's list). Hovering over a flag says what it means; the panel under the 3D view says it in words, with each pattern's own name under Technical detail.
• Click a column's name to sort by it, and again to reverse it. Find a molecule by name or SMILES searches the list; Within drug-like limits, No warning flags and Kept as complexes keep only those hits. The hit you click is shown in 3D beside the list, with what touches it.
• Every hit at a glance plots all the hits: their score (better is up) against their size in atoms, not counting hydrogens, with a dashed line at a score of 0.3 for each atom. A hit above the line scores well for its size, not just by being large. Choose what goes Across and Up; click a dot to see that hit, or drag a box over some dots to keep only those in the list. Clear the box shows them all again.
• Simulate on a hit plans it as a docked pose is planned. Tick several and press Simulate the chosen hits to plan each as its own study with its own price; Run all approves them together. Test the chosen hits (two to five) tests them the same way (testing the top poses) instead.

Where a library comes from

• Choose the file reads a file from your computer, up to 5 MB: SDF, SMILES lines with a name after each, or a CSV with a SMILES column.
• Paste a link fetches a file from one of the public sources ConformerLab reads: ZINC-22, PubChem, ChEMBL, DrugCentral and COCONUT, and, for sets you publish yourself, Zenodo, figshare and GitHub. Each source's terms are listed under the link, and the results credit where the molecules came from. A link to anywhere else, or one that needs a sign-in, is refused. The file may be up to 1 GB, compressed or not; Take at most sets how many molecules to take, and Import it starts the download, which the card follows.
• Use a built-in library offers the sets kept here: Approved drugs (DrugCentral), 3,916 active ingredients of approved drugs, for repurposing screens, ready at once; and Natural products (COCONUT 2.0), 738,794 natural products, fetched from COCONUT when you choose it. Many natural products are too large or too flexible to dock and are left out with the reason; a screen of them is a fast screen.

However a library arrives, exact repeats are dropped, and records that are no molecule are counted, listed with the results and not charged.

Screening

A library of more than 200 molecules is screened, in one of two ways; the check offers the other, with its price, where both fit.

A thorough screen takes up to 20,000 molecules and is the default when it fits: each molecule is searched carefully, and a zinc in the site is scored with its own terms.

A fast screen takes up to 1,000,000 molecules and is far quicker and cheaper, and coarser. Each molecule's search is shorter, and a zinc is scored roughly, which the check says.

• Check it reads the first 50 molecules, to show what the file holds and how many are likely to be left out; the rest are checked as the screen runs. Each molecule is docked in up to two of its likely forms.
• The screen runs in parts, and the run page counts the molecules docked as they go. A part that fails leaves its molecules undocked and uncharged, and the rest carries on. Stopping a screen keeps the parts that have finished, ranked and charged; the parts still going are dropped and not charged.
• The best 500 molecules come back ranked, with the best 25 kept as complexes to see and simulate. The ranked poses (SDF) has the best pose of each, The ranking (CSV) the table, and Every molecule's score (TSV) every molecule in the library: its score, or why it was left out.
• Re-dock them carefully docks the best 25, 50, 100 or 200 again, as a library, in all their forms and at the thorough setting. It is checked and priced on the spot, and Dock them starts it.

A screen searches each molecule once, more coarsely than a docking of one: use its ranking to choose molecules to look at, re-dock the best carefully, and simulate the ones whose poses make sense.

What it will not dock

Small molecules over 80 heavy atoms or 25 rotatable bonds, which docking cannot search reliably; elements other than H, C, N, O, S, P, F, Cl, Br and I, which the simulation does not cover; peptides longer than 20 residues, or with non-standard residues; two proteins of more than 3,000 residues between them; and structures of more than 99,999 atoms. Each is refused before anything is charged.

Contacts in a structure

What holds a bound molecule in place, its hydrogen bonds, salt bridges and other contacts with the protein, can be read from a single structure, free and at once: a PDB entry with its ligand, a model you upload with its ligand, or a docked pose. It is the same reading a simulation makes of every frame (see The 2D interaction map), made of the one structure there is.

Where to find it

• Beside the structure's name over the 3D view, while something is bound to it: Show contacts draws the contacts on the structure and puts the map and the table in the chat. Hide contacts takes them off. Typing show contacts does the same.
• On a docking's poses, Show contacts under each pose.
• On the plan page, the Starting structure card's Show contacts.

What you see

• In 3D, each interaction as a dashed line between the atoms that make it, with its distance, and the residues holding the molecule drawn and named.
• The map: the molecule drawn flat with the residues holding it around it, coloured by kind. Click a residue or a line to see the two residues atom by atom, and ← Back to the map to return. With more than one molecule bound, buttons above the map switch between them.
• The table: each interaction with its residue, its atoms and its distance, specific interactions first and the nearest first; a contact through a water molecule says via water. SVG and PNG download the map with its key.

How each is measured gives the distances and angles each kind of interaction is read by.

With and without hydrogens

Most PDB entries, and every predicted model, carry no hydrogen atoms. Their hydrogen bonds are then read from distance alone, and charges from what each residue usually carries, so a pair close enough is counted even where the geometry would not allow a bond. A line under the map says so: This structure has no hydrogens, so its hydrogen bonds are read from donor–acceptor distances alone. A docked pose is read with its hydrogens, as its simulation would have it.

Exact contacts

A structure read by distance can be read exactly instead. Under its map, Exact contacts adds the hydrogens the way a simulation's preparation adds them, at pH 7.4, and reads the contacts again. It costs a flat 5 credits, and the button says so.

• Pressing it opens a confirmation that says what happens. Nothing is moved, and the structure you gave is not changed. It takes a few minutes, and you can leave the page and come back.
• A bound molecule's chemistry is looked up in the PDB's dictionary of molecules. When it is not there, the confirmation asks for its chemical structure (as SMILES), in the charge state you want, and checks it against the file before anything is held.
• Only Confirm holds the credits; Cancel holds nothing. They are charged once the contacts are read, and released in full if the preparation fails. Asking again with the same chemistry charges nothing more.
• The exact reading then takes the place of the one by distance; Show the reading by distance goes back to compare. It is kept for your account, and opening the structure's contacts again offers Show exact contacts, free.
• It is offered on the study's structure and on the plan page's Starting structure card, not on docked poses, which have their hydrogens already. A Read-only account sees the price but cannot confirm, and an email address that is not confirmed must be confirmed first.

Limits

Only the first model of a file is read, and structures of up to 30,000 heavy atoms (water aside). One structure shows how a molecule is held, not whether it stays: that is what testing the top poses and a simulation's interactions, with the share of the time each holds, are for.

Predicting a structure

When there is no structure to start from (a protein the PDB and AlphaFold DB do not have, a designed variant, a mutant, an aptamer and its target), one can be predicted from its sequence, usually in minutes. Like a docking, a prediction is a short run of its own: checked and priced free, then approved. It hands back a few models, each with the predictor's confidence in it; you pick one, and it becomes the study's structure, marked Predicted. Where a site does not offer prediction yet, its buttons are not shown.

The two predictors

The complex predictor folds proteins, DNA, RNA and small molecules together: a protein alone, a complex of several chains, a protein with the molecule it should bind (co-folding), or an aptamer with its target. For one small molecule on a protein it also estimates how strongly it binds.

The protein predictor folds a protein or a complex of proteins.

Both work from the sequence alone for now, without first gathering related sequences to compare it with (a multiple sequence alignment, or MSA). Their models are therefore less accurate than a structure the PDB or AlphaFold DB already has, the protein predictor's most of all: from the sequence alone it is reliable mainly for designed proteins, so choose the complex predictor for a natural one. Look at each model's confidence before you pick, and prefer an existing structure when there is one.

Asking for it

• On a sequence the chat found (yours, or a UniProt entry's), press Predict its structure (see A sequence).
• When the AI planner works your question out with you and there is no structure to start from, the study begins with a prediction; once you press Plan this question, the prediction card opens with the sequences you gave, and the question is planned once you pick a model.

The prediction card

Under Chains, each chain is a card: Protein, DNA or RNA, an optional name, how many Copies (up to four), and the sequence as one-letter codes or FASTA. Add a chain adds another, up to six. With a small molecule takes a SMILES and a name, to fold the protein with it (complex predictor only); Estimate how strongly it binds adds the binding estimate. Choose the Predictor and how many Models (1, 3, 5 or 10; each adds compute time), then press Check the price. Stop here leaves it, and nothing is charged.

What the check refuses

Letters outside each kind's alphabet (the 20 standard amino acids; A, C, G and T for DNA; A, C, G and U for RNA); a protein shorter than 10 residues (dock a short peptide instead) or a DNA or RNA shorter than 6; and more than 1,200 residues, nucleotides and small-molecule atoms in all: predict the domain the study is about. The protein predictor takes proteins only. The binding estimate is for one small molecule on a protein, and is less reliable for large molecules, which the check warns about. A molecule with elements the simulation does not cover can be folded, with a warning that the model could not then be simulated.

The price

A prediction is priced by its compute time, at a credit a minute: about 8 minutes to start, then each model, longer for a larger complex, and half a minute for the binding estimate. Ubiquitin's 76 residues with five models from the complex predictor come to about 9 credits. Press Predict it to hold the price and start; Change something goes back to the card. A prediction that fails is refunded in full, and one you stop costs nothing.

The models

The models come back best first, each with how confident the predictor is in it, in words with each score's own name in brackets: its confidence in each residue's place (0 to 100, pLDDT: above 90 very high, 70 to 90 confident, 50 to 70 low, below 50 very low); in the overall shape (0 to 1, pTM); for a complex, in how the chains meet (0 to 1, ipTM: above 0.8 is confident, below 0.6 probably wrong) and in where a small molecule sits; and the overall score the models are ranked by. The binding estimate, when asked for, says how likely the molecule is to bind, as a percentage, and roughly how strongly, as the concentration at which it has half its effect (its IC50, in nM or µM; lower is stronger): an estimate for ranking molecules against each other, not a measurement. Technical detail, under each, names the predictor and its settings and gives the raw numbers. On the prediction's page the model you click is shown in 3D, coloured by its confidence in each residue in AlphaFold's colours: blue very high, light blue confident, yellow low, orange very low. Beside it, Expected position error maps how sure the predictor is of where each part sits relative to every other: dark green where it is sure, pale where it is not. A dark square along the diagonal is a part it is sure of; a dark block between two chains means it is sure how they sit together. Hovering over the map reads it. A prediction made before the maps were kept says so. Under the model are the same command line and Style panel as a docking's (see Picking a pose), with Confidence, in the colours above, as the first look and colour: colour by confidence brings it back, and zoom 40 centres on residue 40. The download button saves a model as a PDB file; Use this model makes it the study's structure, and a molecule folded with it keeps its chemistry, ready to simulate. The report names the predictor that made each model, with the credit it asks for; it comes with the completion email and is on the prediction's page under Download the report (PDF).

Aptamers

An aptamer's complex with its target is co-folded with the complex predictor, then simulated: the predictor's confidence in the interface says how sure it is of the fit, and the simulation tests whether the fit holds.

QM/MM

QM/MM treats a small part of the system with quantum mechanics, which follows its electrons, and the rest with the ordinary simulation around it. The part can be a ligand on its metal, an enzyme's active site, or a bond that forms or breaks. It shows what an ordinary simulation cannot: charge flowing between a molecule and the metal it binds, a proton changing hands, the energy of a reaction along its path. It runs after an ordinary simulation, from where each replica ended, and it is far slower, so it covers picoseconds (thousandths of a nanosecond) rather than nanoseconds.

Asking for it

Say so in the request, for example 100 ns of MD, then 15 ps of QM/MM of AZM on the zinc. Name the molecule by its code in the structure, a metal by name, and side chains as for docking (A:His94). Name the method if you have one in mind (DFT or B3LYP; otherwise the faster GFN2-xTB is used). What the request leaves open, the checker asks, with each choice priced.

The QM region

On the plan page the quantum chemistry step lists its Atoms computed with quantum chemistry (QM region): Molecules, Metals and Side chains, each edited as text separated by commas, with the region's Spin state (multiplicity), said as its unpaired electrons, and Overall charge (both Worked out from its parts unless you set them). The plan assistant changes them too: add Thr199 to the atoms computed with quantum chemistry comes back as a priced button. A few rules apply, and each is explained when you meet it:

• A metal goes in with what holds it. A region that leaves out one of the metal's side chains, or a molecule bound to it, is refused, with the offer to add them.
• Side chains are cut from the backbone, so glycine (no side chain) and proline (a ring through the backbone) cannot be added, and half of a disulfide only with the other half.
• Iron, copper, manganese, cobalt and nickel usually have unpaired electrons, so their spin is asked, never assumed, with the usual states offered (Iron(II), high spin (quintet) and the others).
• The few waters nearest the region join it.

The methods

GFN2-xTB, a fast, approximate (semi-empirical) method, is the default; it treats regions of up to 300 atoms for tens of picoseconds.

B3LYP-D3(BJ)/def2-SVP, a density functional theory (DFT) method, is far closer to experiment for bonds that change and far costlier, so it treats regions of up to 100 atoms for a few picoseconds.

Pick one with Quantum method: Faster (GFN2-xTB) or the more accurate, far slower one. A region too large for its method is refused, offering the faster method where that fits. A molecule's size is read from its chemistry, which ConformerLab looks up in the PDB's dictionary of molecules; only a molecule it cannot find is asked for.

How it runs

After the ordinary simulation, from the end of each repeat: a short settling of the region, QM/MM settling, then the QM/MM length that is measured. Runs sets how many to run, each from a repeat of its own, so more runs need as many repeats. An interrupted run resumes from where it last saved. These are under Specialist settings; the QM method is shown first.

Free-energy profiles

For a reaction, the plan's Energy along a reaction (QM/MM) step computes how the free energy changes along The distance followed (reaction coordinate): a distance between two atoms, or the difference of two distances (a proton passing from one atom to another), each atom named as residue@atom, such as A:His94@NE2, and each in the QM region. The system is moved along the coordinate from From to To and sampled in Windows, each held near its point by a Restraint for Each window; the windows are then joined into one profile, with its uncertainty. The checker warns when the windows are too far apart, and offers more.

QM/MM on docked poses

When the top poses are tested, setting Starts from to each tested pose's end runs QM/MM on every pose after its short simulation: the pose's charges and its bonds to a metal, beside how it held. The ordinary test still chooses the pose.

The price

The quote prices each trajectory, and each window of a profile, at the method's measured speed for the size of the region. The run is charged for what ran, never more than the hold. A QM/MM part that fails is not charged, and the simulation is delivered without it.

Results

The run page lists each QM/MM run as it goes. The results show, for each trajectory, the charge each part of the region carries in the ordinary simulation and the charge quantum mechanics gave it, and each metal's distance to what holds it; for a profile, the barrier and the reaction free energy, each with its spread, and the profile with its band. The files include the trajectories and the scripts to run it yourself.

The plan in the chat

Pressing Start planning opens the chat. One line says what is happening as it happens: Reading your request, then, once the structure is in, Drafting the plan and Pricing it. Each draft from the AI planner is checked against every rule, and a question only you can answer, such as what a molecule in the file is, is asked rather than guessed. If the AI planner's drafts do not pass, a plan is built from ConformerLab's rules and the settings read from your sentence, and the line says Planning from ConformerLab's rules. Once done, the line folds to Planned; click it to see each step, what was read from your request, and why a draft was redone. The plan opens with a line such as I've read 1AKE and set up 100 ns of ff19SB in OPC, 1 replica. A checker has gone over every setting; here it is, ready to run., or ends One thing to decide first: when something blocks.

Decisions, asked one at a time

Each thing that blocks the plan is posted as a message followed by a question with numbered options and the footer Or type below — a number, or what you want instead. When there are several, the chat says This is 1 of N things to decide. and offers See all N, which lists every finding, tagged blocking, warning or note. Answer by clicking an option, pressing its number key (when the cursor is not in a text box), typing the number, or typing words from the option. Typing stop, cancel, never mind or not now on a question that offers Stop here stops there, and the chat posts Stopped — nothing charged.

Anything else you type is taken as a new instruction. After an answer the plan is rebuilt and priced again; earlier questions grey out, and earlier plan cards read Replaced by the plan below.

The decisions you may meet

An answer that removes a molecule takes it out of the file and reads the structure again; what you had told the chat about everything else is kept. Every option that changes a setting is priced before you see it, so the credits it shows are what that plan costs.

The plan card

The card is headed Ready to run in green, or Almost ready in amber while something blocks. Its rows are Structure, Runs (for example “100 ns, run once, at 27 °C (300 K)”), Takes (an estimate of how long) and Costs (credits, dollars, and held now, charged for the compute time used, never more).

Under them, What I chose for you lists the choices that matter most to the result and the price, each tagged you said or chosen, with an alternative worth weighing as a priced chip (for example Repeat it 3 times, for confidence). Pressing one makes it the plan, priced again. Worth confirming lists warnings that have a one-click change. What runs, step by step shows the protocol in order: prepare the structure, surround it with water and salt, simulate, and analyse (how much the protein and any ligand move, which parts are flexible, the contacts that hold, and a playable trajectory).

The buttons are Run it · N credits, See every setting, which opens the plan page, and Just the setup files, which gives you a prepared system to run on your own computers instead (see The price).

If your credits are short

When the balance is below the price, Run it is disabled, and the card says how many credits it needs and how many you have. It offers the same study shorter, priced to fit; fewer replicas, when there are several; and Get more credits, which opens the billing page in a new tab. The plan stays where it is, and Run it comes back on by itself when the credits arrive.

The plan assistant

Once a plan is on screen, messages in the chat go to the plan assistant instead of starting over. It answers in under 120 words and may add up to three suggestion cards. Each is a complete plan, checked and priced, showing its label, a one-line reason, its credits and the difference from the current plan, a still needs a decision tag if it is blocked, and Use this, which becomes In use. An instruction such as make it 200 ns is applied at once as the first suggestion. The assistant can change only settings the plan already has; a different structure or molecule is planned again as a new study, and the chat says so. Asking it to dock or place a molecule opens the docking card instead.

The guide at the top of the study page shows the steps of the work in hand, and where you are in them: Describe, Structure, Plan, Approve, Simulate, Results for a simulation; Describe, Structure, Molecule, Approve, Dock, Poses for a docking (Molecules and Hits for a library, Library, Screen and Hits for a screen, Peptide or Partner for a peptide or a protein); and Describe, Sequence, Approve, Predict, Models for a prediction. Once a pose or a model is picked to go on with, the structure step reads Docked pose or Predicted model. A run's own page shows its kind's steps the same way.

The chips above the message box are the next steps for the work in hand, in plain words. Before anything is asked they are sentences to start from, which fill the box without sending: Will my drug stay bound to its target?, Dock a molecule into my protein, Screen a library of compounds against my protein and, where structure prediction is offered, Predict a protein's structure from its sequence. With a structure and nothing chosen yet, they are Show what holds the bound molecule (when something is bound), Dock a molecule into it and Simulate it as it is. While a docking, a screen or a prediction runs, they are How is it going?, Stop the docking (Stop the screen, Stop the prediction) and Show me the poses (Show me the hits, Show me the models), answered in place and free. When it is back, they are Show the best pose, Simulate the best pose and, for a molecule with several poses, Check which pose holds best; for a library or a screen, Show the top hit and Choose hits to simulate; for a prediction, Use the most confident model. Once a simulation is planned, they are the alternatives from What I chose for you that a researcher would weigh, each with what the whole plan would cost and the difference — Repeat it 3 times, for confidence or Run it once, for less, and Faster and cheaper — which one press makes the plan, re-quoted, without asking the model; and What will this tell me?, which asks the plan assistant. After an answer that suggests changes, they are those changes, priced the same way. While the plan on screen is running, they are How is it going?, Stop the run and Show me the results (see Approving and the live run); once it completes, Show me the results. The box reads Ask anything, or change the plan — e.g. “make it 200 ns”. Every reply shows what it cost.

Under each answer and each plan the chat drafts are Copy, Ask again (answers only: the same question once more, at the price of a reply), Helpful and Not helpful. Not helpful asks, if you will say, What was wrong? (optional), sent with Send. Pressing the same one again takes the rating back. A rating is kept with the question and the reply it was about, so the team can see where the assistant helps and where it misleads; it is in your data download and goes with your account.

The chat moves the 3D view. A residue named in an answer (Lys745 or A:LYS745) is a link: it shows that residue and its neighbours and turns the view onto it. Typed on their own, these are answered at once and cost nothing: show LYS 745, show chain B, show the ligand, show the binding site, show contacts (see Contacts in a structure) and reset the view. For a study planned from a docked pose, show pose 3, next pose and previous pose show the docking's other poses for a look, with Back to the plan's structure; the plan keeps the pose you picked.

Changing your mind

• Before a plan exists, a new message replaces the request. Restate everything you want, length and replicas included.
• After a plan exists, ask the assistant, pick a chip under Worth confirming, or change any value on the plan page. Every change makes a new plan with a new price.
• Attaching ligands, or answering a decision, plans again at once.

The plan page

See every setting on the plan card opens the plan page, headed Review the plan, then approve. The top bar shows the study's title and a tag, Awaiting approval or Needs a decision, with Back to the chat. The Plan reference is a short code that identifies this exact plan, and the run page later shows part of it: An edit creates a new plan and a new quote.

The Starting structure card comes first: the structure, where it came from and what it holds, with Show contacts when something is bound to it (see Contacts in a structure).

The checker

The checker's card is headed Checked by 51 scientific rules and subtitled Fixed scientific checks, not AI. A blocking issue stops the run. It tags what it finds by how much it matters:

• blocking: the run cannot be approved until it is resolved. Each asks How do you want to go on? with priced options and a last one, Stop here, marked Nothing is charged.
• warning: shown, but you may go on. Those with a one-click change ask Change it?
• note: for your information, and written into the setup files' README among its caveats.

Each finding is one plain sentence saying what it means for your result and what to do. Under it, Technical detail opens the same finding in a specialist's words, with the settings and the numbers the check used. Answering a question makes and prices a new plan. With nothing to report the card says Nothing to flag. Every rule passed. The rules you are most likely to meet:

The protein and water models must belong together: each model of the protein was built with one model of water (ff19SB with OPC, ff14SB with TIP3P, CHARMM36m with TIP3P-CHARMM, under Specialist settings). A few other pairs pass with a warning; the rest block, with the matching water offered.

One protein model (CHARMM36m) cannot keep molecules from your structure, and takes chloride as its only negative ion; either blocks, with a fix offered.

A longer time step (above 2 fs) needs the faster time step setting on, which lets the simulation take longer steps safely; without it the plan blocks. The longest step is 4 fs.

For the water around the protein, a layer thinner than 1.0 nm blocks, and salt over 500 mM blocks.

Temperature outside 250–400 K (−23 to 127 °C) warns.

A single run of 100 ns or more warns that one run gives no error bar: repeating it three times does. A run under 20 ns warns that it is a quick check, not a result.

For the structure itself, a predicted model warns. Short gaps warn and are filled in; long ones block. Metals, unknown molecules, sugars, unknown residues and attached compound files block until you answer.

The protocol

Under The protocol, each stage shows its credits, its values and why they were chosen. Most studies need no change here: the planner chose each value for your question. The settings most people decide (length, repeats, temperature, pH, salt) are shown first, each with a line saying what it does; the rest, which the platform sets for you, are under Specialist settings, still there to change if you know them. To change one, click it: it becomes a list or a number box with an Apply button. Press Apply or Enter to send it; Escape cancels. The plan is checked and priced again, and the page redraws with the new fingerprint and price. Stay within the ranges below: a value outside them is refused, with an error under the approve button. Settings marked * are under Specialist settings.

The table scrolls sideways.

In plain words: the padding is the thickness of water around the protein; equilibration is the settling time before anything is measured; the length is the part that is measured; and the frames saved are the snapshots kept for the movie and the analysis. The analysis has no settings: every run measures how much the protein and any ligand move, which parts are flexible, the hydrogen bonds, how compact the protein stays, and how much of the time each contact holds.

Depth presets

When the plan has a simulation, the quote column opens with Choose depth: As planned (the plan the planner returned), Quick look (half the length, one replica) and Thorough (double the length, and three replicas or more). Each shows replicas × length, hours and credits, and each is a real plan, checked and priced, so the number shown is the number you approve. After you edit anything, none is highlighted, and clicking one replaces your edited plan with that preset.

Ask about this setup

The Ask about this setup card is the plan assistant again (Why a value was chosen, what to change, what it costs. Every suggestion is checked and priced before you take it.). Its starter questions send at once: Is this long enough for my question?, How can I make it cheaper?, What would you change?, and How do I get past the blocking issues? when the plan is blocked. Each reply shows what it cost.

The price

How a quote is built

Under Quote there is one line per stage: Preparing the structure (0.6 credits), Adding water and salt (0.18), the simulation itself with its length and how many times it runs (Simulation · 100 ns, run once), Analysis and report (0.48), and, for setup files, Setup files to run yourself (a flat 100). Then Total, in credits and dollars, and an estimate of the hours. Under it, how many times it runs and about how much compute that takes, tagged Measured: timed on real runs, not a guess; Speed on this system opens the simulated time per day of compute for a specialist.

The simulation line is compute time, and one credit is one minute of it, so an hour is 60 credits: $1.50 paying as you go, at $0.025 a credit, and $1.20 with credits from the $200 pack. Two things set it: how much simulated time you ask for (length × replicas), and how large the system is, which is your structure plus the box of water built around it. The box follows the structure's shape, not only its number of residues, so an elongated protein costs more than a compact one of the same size. On the default plan (100 ns, one replica), for example, 3BLH, 547 residues in a compact shape, comes to 589.4 credits (about $14.74 paying as you go), or 326.2 with hydrogen mass repartitioning on; 7A3R, 762 residues stretched across 15 nm, comes to 2,302.3 credits.

Worth knowing about the estimate:

• The hours assume each replica starts at once. When the site is busy, replicas wait, and the run takes longer; the price does not change, since it is set by compute time, not by how long you wait.
• The box shape and the padding change the price: a cube holds 30% more water than the default shape, a truncated octahedron, and a rhombic dodecahedron 8% less. Salt, pH and the number of frames do not change it.
• Rebuilding the disordered ends of the chains makes the box much larger, and the price with it; leaving them off, the default, costs nothing.

Held, then charged for what ran

When you approve, the whole price is held from your balance; the run page's credits card shows the hold. When the run ends, you are charged the small fixed lines as quoted plus the minutes of compute the replicas took, never more than the hold, and the rest goes back to your balance. Minutes spent starting up, and minutes lost when a replica is interrupted, are not charged.

A run that fails on the platform's side is refunded in full, automatically. The hold is released and the run reads Failed · refunded; when the fault was in our systems, its page says Something went wrong on our side, so the run stopped; every credit it held was refunded.

A run you stop is charged the hold times the share already simulated, and never more than the hold.

A run that finishes is charged what it used. Unused credits return to your balance, and the results page's What you paid card shows the hold, the charge and what came back.

Just the setup files

Just the setup files on the plan card, or Setup files instead under Run it on your own computers? on the plan page, replaces the run with a prepared system, already in its box of water and salt, that you run on your own computers. It costs a flat 100 credits plus preparation and solvation, with no compute time, and is checked by a short test run before you download it. After approval, the results page reads Your setup files are ready. with the command to run it. It is built in your plan's protein force field (shown, and changeable, under Setup files on the plan page), with your plan's preparation and water box. Its script uses the default simulation settings (100 ns, 4 ns of equilibration, 2 fs, 300 K, no HMR, 100 frames), not other changes you made on the plan page.

Approving and the live run

Approving

Approve from the plan card with Run it · N credits, or from the plan page with Approve & run · N credits. Under the plan page's button, three lines say what will happen: the hold and the cap, that an interrupted replica resumes without charging twice, and your balance before and after. The button is disabled, and says why, while a decision is open (Decide the blocking issue first), when credits are short, after several stopped runs in a row (see Stopping a run, below), or on a site that cannot run simulations.

Read-only accounts can plan, ask the assistants and read results, but not run: their button reads Your role cannot run jobs. Your role is shown under your email in the sidebar when it is not Customer.

Approval is bound to the plan fingerprint, which covers the structure, the ligands and every setting, and to the price you saw, so what runs is exactly what was priced. When you press the button:

1. Uploading structure…: your browser sends the structure to the server.
2. Approving…: the server checks the plan, the price and the structure, then holds the price.
3. The chat records the approval and follows the run in a card, step by step (Preparing the system, each run's progress and time left, Reading the results), with Open the run page and Stop the run. Type status, stop or results at any time: the chat says where the run stands, asks before stopping it (Stop it or Keep it running), or opens the results. When the run ends, the chat says so, with what was charged and See the results. The plan card's button then reads Approved · followed below, so pressing it again cannot start a second run by accident. A change asked for after approval makes a new plan, which runs only when you approve it.

If approval fails, the reason appears in bold under the button, such as This plan has changed since it was priced. Open it again to see its price, then approve., The price has changed since you saw it., This run costs 120.0 credits, and 40.0 are available. Buy credits on the billing page, then approve again., no structure uploaded for this plan or structure hash mismatch, each saying what to do next. For any of these, plan again from the chat, or from the file. Plans are kept for 30 days; an older one answers unknown plan. Each approval starts its own run with its own hold.

The run page

The top bar shows the run's title, a status tag (Queued, Preparing, Running, Analysing, Stopping, Complete, Failed · refunded or Stopped), when it was approved, and its plan fingerprint. You can close the tab; the run carries on. Your runs are listed in the sidebar under Studies, and on the start screen under Recent studies.

A run goes through these phases:

1. Queued — waiting to start. Runs start in turn, shared fairly between accounts, so a run can wait, and the page says so.
2. Preparing your system: the structure is prepared, put in its box of water and salt, and checked by a short test run before any compute time is charged. Each stage ticks off in the Timeline.
3. Each run then goes through Warming up (relieving any clashes and bringing the system to the set temperature), Settling, then the simulation that is measured, with the share done and the time left. Each run's card says which step it is on; its pace and last save are under Technical detail.
4. Analysing: once every replica is done, the trajectories are read, and the page switches to the results by itself. If the analysis cannot be made, the run still completes with its trajectories, and the analysis is not charged.

Replicas and interruptions

Each replica has a card with the nanoseconds done out of the total, a progress bar, its phase and its speed. The nanoseconds count the measured part only, while the bar includes heating and settling, so early on it can read 0.000 ns with the bar moving. A replica that is interrupted starts again from where it last saved: its card reads interrupted: starting again from the last checkpoint, and the lost minutes are not charged. If a replica cannot go on, the whole run stops and the hold is released in full; partial results are not delivered. The time left counts only replicas already running, so it is optimistic while others wait to start.

Live charts, structure and credits

With the first update, four charts appear with one line per replica: Temperature, Potential energy, Box volume and Speed. Once the system has settled, the temperature, energy and volume level off. A structure with clashes can start with a very large energy, which falls as the clashes are relieved. Point at a chart to read every replica's value at that time, or press Show as table. On the right, Starting structure · click to enlarge shows the structure, and the Credits card shows what a stop would cost now against what is held, not the final bill.

Stopping a run

Stop run is available while the run is queued, preparing or running. It asks Stop this run? You are charged for the part already simulated, and it cannot be resumed. On OK the button reads Stopping…, and the run stops within seconds, so a little more may be charged after the click. The charge is the hold times the share of the simulation done; stopping while queued or preparing costs nothing. A stopped run keeps no files, analysis or trajectory, and its page reads Stopped — charged x of y credits. Near the end of a run, stopping can cost more than finishing, which pays only the compute time used.

After three stopped runs in a row the chat warns you once; a fourth starts a 10-minute wait before new runs can be approved, and the approve button counts it down (Too many stopped runs). Finishing a run clears the count. Planning and reading results are never blocked by it.

Notifications

The bell in the top bar shows how many notices are unread. Its Notifications panel lists each: what happened, the run it is about, and when. Details unfolds the whole notice, clicking a notice opens its run, and Mark all read clears the count. Notices come when a run completes (with its main finding and a link to the report), when it could not finish, a week and a day before its files are deleted, when they are deleted, and when a storage renewal fails; none comes when you stop a run. Email me when a run finishes and before its files are deleted is on by default; where the site has no mail service, the panel says Email is not set up on this site yet; notices appear here meanwhile. Also tell me in this browser turns on browser notifications.

The completion email carries the report as a PDF (a very large one comes as a link), so it can arrive a little after the results page shows the run complete. The report follows the kind of job: a simulation's has its findings, figures, interactions and quality checks; a docking's, its poses with their scores and contacts; a library's or a screen's, its molecules ranked and what was left out; a pocket search's, the pockets it found; a test of the top poses, how each pose held; a prediction's, its models with the predictor's confidence in each. A docking that takes less than ten minutes is followed on screen, so it sends no email. A docking's or a prediction's report is also on its page under Download the report (PDF). The email's link to the report works only in a browser where you are signed in.

Results

The answer

A completed run opens on its results. The answer gives one sentence. With a bound ligand, it says whether the ligand stayed in its starting pose (it moved less than 2 Å, a little more than the length of a chemical bond, in every replica), left its starting position (5 Å or more in every replica), or moved within the site in some replicas. Without a ligand, it says whether the protein's fold held. These distances are measured as RMSD: how far, on average, the atoms are from where they started.

Under the sentence are the first notes from the analysis, and tags: whether the runs agree, or The runs disagree; One run — no error bar; and, with a ligand, Stability, not binding affinity, because a ligand that stays put has not been shown to bind tightly. On the right are Download the report (PDF), Download everything and Your files. “Starting pose” means where the molecule was once the system had settled, not the file you uploaded. A docked molecule goes by the name it was docked under, not its code in the file.

Tiles and charts

Up to four tiles follow: How far it moved (how far the ligand had moved by the end, averaged over the runs), Time in pose (the share of the time it stayed within 2 Å of the start), Protein (how far the protein's backbone had moved), Temperature (in °C, with the kelvins beside it) and Simulated (runs × nanoseconds). The main chart follows the ligand's distance from its starting pose over time, or the protein's, one line per run. With a ligand, Is the protein holding its shape? charts the protein too. Which parts of the protein move shows each residue's flexibility along the sequence and names the most mobile residues; click a name to see it. Every chart has Show as table.

The trajectory player

The player shows the simulation as a movie: the protein as a ribbon coloured by chain, bound molecules as ball and stick. It first shows Preparing the movie…; the playable version is made once and kept.

• ▶ plays and loops, ❚❚ pauses, and the slider moves through the frames, with the time in ns.
• Playback speed offers 0.25× to 4×, remembered in this browser.
• With more than one replica, a menu switches the movie, the contacts and the map to another replica.
• Protein / With water adds the water and ions close to the molecules.
• Full screen enlarges the player; Whole molecule clears any residue focus and re-centres.
• Download video records the playback as a video file. It records in real time, so the tab must stay visible.
• Clicking an atom focuses its residue.

A residue and its contacts

In Residues and their contacts, type a residue as A:201, GLY201 or 201 and press Show, or click a residue named anywhere on the page, on the 2D map or in the 3D view. The view zooms to it, with its closest neighbours and the interactions in the frame shown. The panel lists the residues it touches, each with the share of the time they are in contact and the kind of contact (such as “H-bond 45%”). Add its map to the report keeps that residue's map for the PDF.

The 2D interaction map

Interactions in 2D draws what is bound flat, with the residues that hold it around it, coloured by kind: a ligand as its molecule, a peptide as its chain, a second protein as its interface. With nothing bound, it opens on the residue with the most hydrogen bonds and salt bridges. Buttons above it switch between the bound molecules and the residue you last selected. In at least 10%, 30%, 50% or 70% of frames sets how often an interaction must be present to be drawn (30% by default). Click a residue to show it in 3D, or a line to see the two residues atom by atom (← Back to the map returns).

The key names the kinds: Salt bridge, Hydrogen bond, Cation–π, π-stacking, Metal coordination, Water bridge and Hydrophobic contact; How each is measured gives the distances behind each. SVG and PNG download the figure with its key and a caption. The figure takes the page's colours, so switch the app to its light theme first for a printed figure. Add to report adds a residue's map; the bound molecule's map is in the standard report already.

Asking about the results

The Ask about these results card answers from what this run measured. Two questions are always offered, Is the structure stable over this run? and Which residues move the most, and where are they?, or type your own. Every residue the answer names is a link to it in 3D. Under each answer are Add this answer to the report and buttons to add the maps of residues it mentions. Each reply shows its cost (see Credits and costs); a declined reply is free. The conversation lives only in the page: a reload clears it, and only answers added to the report are kept.

Methods and limits

The Methods card holds a methods paragraph written from the settings that actually ran, the software and data the system was made from with the credit each asks for, and a line to reproduce the run. This is where a specialist reader or a reviewer finds every program, force field and parameter; Copy puts it on the clipboard. What the numbers say, and their limits lists every note the analysis wrote: always that this kind of simulation cannot break or form bonds or show electrons shifting, and, with a ligand, that staying put is not the same as binding tightly.

Report and downloads

The PDF report

The Report card is on a completed run's results page. The standard report is made automatically when the run finishes, free whatever its length: a cover with the headline finding, The answer with its tiles, findings and limits, how the structure moved, what holds it together, quality checks, methods, and the files, with how long they are kept and how to cite the run.

You can choose what a remade report holds. The standard report lists its sections as checkboxes; under Extras · each adds to the pages are further figures, then the residue maps, 3D views and answers you have added, each with Remove. Interactions shown: at least 10 to 70% of frames sets the map's threshold.

The report is priced by its pages: the first 10 are free, then 2 credits a page, paid once for the run. A report made again, or shorter, costs nothing; a longer one, only its new pages. Before anything is charged, a dialog gives the exact count, for example “This report is 13 pages: the first 10 are free, so the 3 pages past those cost 6 credits, charged once. Make it?”. The button reads Make the report, Make it again once your choices have changed, or Download the report; it reads Preparing… while the interactions are worked out.

Adding your own figures and answers

• In the player, Add this view to the report captures the current 3D view on a white background and asks for a caption.
• On a residue's 2D map, Add to report, or Add its map to the report in the residue panel. A report holds at most 12 residue maps.
• Under an answer from the assistant, Add this answer to the report keeps it for a “Questions you asked” section, up to 20 answers.

Adding something only ticks it in the card: make the report again to include it. The results page does not refresh itself, so if the automatic report finished after you opened the page, reload to see it.

Downloads

Your files, tagged Checksummed bundle, starts with Download everything, one zip named after the run with the report at the top, then lists the files in groups: report, run, bundle, one group per replica, and setup. The methods and readme are offered as METHODS.docx, METHODS.pdf and README.txt. Links expire after an hour; reload the page for fresh ones. If no report exists yet, it is made first, which can take minutes.

The bundle holds everything needed to run the same simulation yourself, and the paid run used this same script: the prepared system in its water box, the simulation script, every setting it ran with, the methods, a readme saying how to run it, and checksums to verify the files. The structure you uploaded is not among the downloads; the prepared one is.

Some of the parameters and models in these files are shared under the CC BY 4.0 licence, which asks anyone who shares them to give credit. The methods carry that credit; if you share the files, share it with them.

How long files are kept

The Storage card on a completed run's results page shows the size of its files and the date they are kept until: free for 30 days after the run finishes. If the period is not extended, the files are deleted 7 days after it ends. The summary on the page stays, so the charts, methods and assistant still work, but the trajectory can no longer be played and the report can no longer be remade.

Longer storage costs $0.045 per GB a month, which is 1.8 credits per GB-month at the pay-as-you-go price, with a minimum of one month. Keep it 1 more month charges one month after a confirmation. Keep it, renewing monthly renews a month at a time until you press Stop renewing, which keeps what is paid for to its end, or until the credits run out, with a notice. Extra months are added from the end of the current period. The size counts everything kept for the run, so it is larger than the download. The account's Storage tab lists every run whose files are kept, soonest to go first (see Your account).

What you paid

The What you paid card shows the credits charged, a tag with what was returned, and every entry for the run: Hold at approval, Charged, Hold released, Refunded. Storage and report charges for the run appear as further entries but are not in the large number.

Your account

Press Account in the sidebar. The page is headed with your name and email, and its tabs hold everything about the account: Overview, Usage, Billing, Storage, Profile and API keys. Buy credits and See usage on the sidebar's credits card open the Billing and Usage tabs. Months and dates on these tabs are UTC, as the monthly free allowance for planning is.

Overview

Four figures: what is Available, with what is held for running studies; what was Spent this month, less runs refunded, beside last month's; the Free planning left this month, and the day it renews; and the Runs with files kept, with the next renewal or the first files to go. Under them, this month's spending by kind, the latest lines of Recent activity, your last payment, and this month's statement as a PDF or a CSV.

Usage

Spent each month draws the last twelve months as bars split by what the credits went on: simulations, docking and screening, storage, planning and the assistants, report pages, and anything else. Every credit in and out lists each line, newest first: the date, what it was, the run it was for, the credits and the balance it left. The buttons above the list show one kind at a time, and Show more reads further back. A hold, the credits set aside while a study runs, is not a line; the run's charge is, when it ends.

Billing

Buying credits, your payments with their receipts and invoices (see Buying credits), and Statements: choose a Period, from This month, Last month, The last 12 months, This year and Last year, or Dates I choose with From and To, then press Download PDF or Download CSV. A statement gives the balance the period opened and closed with, what was held when it ended, the payments made in it, and every credit bought, added and spent, with the balance after each line. A statement is not a tax invoice; the receipts and invoices for payments come from the company that took the payment.

Storage

Every finished run whose files are kept, soonest to go first: the date they are kept until, and what happens then. A run in its last week is tagged Going soon; a run set to renew is tagged Renewing, with what a month costs. Runs whose files went in the last 30 days are listed as deleted; their summary stays on their page. Manage opens the run, whose Storage card keeps the files longer (see How long files are kept).

Profile and API keys

Profile holds the password and sessions, Email notices, your data and deleting the account (see What you need). API keys is where keys for scripts are made and revoked (see Use it from code); API access in the sidebar opens it too.

Credits and costs

Everything is priced in credits. Paying as you go, one credit is $0.025, and the app shows that dollar figure beside every price; a pack makes credits up to 20% cheaper (see Buying credits, below).

Where to see your balance

The sidebar's Credits card shows what is available and, under it, what is held for running studies, or Nothing held right now. Available is your balance less the credits held. The card updates as your runs change; if it looks out of date, reload the page. The start screen's Recent studies table shows, for each run, what was charged or what is held.

What costs credits

Simulations cost 1 credit a minute of compute, plus the small fixed lines (0.6 for preparation, 0.18 for solvation, 0.48 for analysis). The price is held at approval and charged for what ran, never more than the price. See The price.

Docking, screening and prediction are priced by the check before anything is held: see the prices under Docking and Predicting a structure.

Setup files cost a flat 100 credits plus preparation and solvation, with no compute time.

Exact contacts cost a flat 5 credits to give one structure its hydrogens and read its contacts exactly (see Contacts in a structure): held when you confirm, charged once the contacts are read, released in full if it fails. Seeing it again is free, and reading contacts by distance is always free.

The AI planner's work (reading your request, answering a question that names no structure, and every draft behind a plan) and the replies of the plan and results assistants are metered. Each account with a confirmed email address has 40 free credits of this a calendar month (UTC); after that, it is charged from your balance at 3 times what it costs us, at the pay-as-you-go price of a credit. Under each reply a line reads This reply: with its credits, whether they came from the free allowance or your balance, and how much free use is left this month. A reply the AI planner or the assistant declines costs nothing, and says so. When the month's free use is spent and your balance is empty, the AI planner and the assistants wait until credits are added. The start screen's Planning is free refers to compute time.

The first 10 pages of a run's report are free, then 2 credits a page, each page paid once for that run. The report made when the run finishes is free.

Storage is free for 30 days after a run, then 1.8 credits per GB-month if you extend it.

Buying credits

Press Buy credits on the sidebar's credits card. The page lists three packs, each with a Buy button: $20 for 900 credits, $100 for 4,700 and $200 for 10,000. That is about $0.022, $0.021 and $0.02 a credit, 11%, 14% and 20% less than paying as you go, as each pack's badge says. Under Pay as you go a box takes any whole number of dollars from $20 to $5,000 in Amount in US dollars, at $0.025 a credit (40 credits a dollar). Buying opens the checkout page of Lemon Squeezy, ConformerLab's authorised reseller, which the billing page names. It sells the credits on ConformerLab's behalf: it is the seller on your receipt, asks for your billing address and adds any sales tax or VAT where you live. Your card details never reach ConformerLab. Only an account whose email has been confirmed can buy, and an account buys up to $200 in its first week.

When the payment goes through, you are sent back to the same page, which reads Payment received. with the credits added; a notice says so too. The page may first say The payment is not finished yet. until the payment is confirmed, usually within a minute: reload it then. A bank payment can take a few days, and the credits arrive when it clears. On the account's Billing tab, Your payments lists every payment with what it bought, what was paid and its status, with a Receipt button for each that went through, and an Invoice button where the payment company made one; both open on the payment company's site in a new tab. A refund goes back to the card, and the same share of the payment's credits leaves your balance. The reseller, as the seller, settles any dispute about a payment with your bank; if a payment is charged back, its credits may be taken back, and a balance below zero must be paid back before anything can be approved. For a purchase order, or an invoice made out to your organisation, ask the ConformerLab team. Unused credits you paid for can be refunded within 30 days of buying them: the refund policy says how, and the terms of service cover the rest.

Getting credits otherwise

The ConformerLab team can add credits to your account, by the email you signed up with. When a plan costs more than you have, the plan card offers the same study shorter or with fewer replicas, priced to fit, and the plan stays where it is until credits arrive.

Use it from code (API)

Everything the app does, a script can do: a pipeline can upload structures, plan and price studies, approve them, follow the runs and download the results without anyone at the screen. A script carries an API key instead of your sign-in, and the key acts as your account: it spends the same credits, meets the same limits and sees the same runs. There is no separate API price. The API is in beta: this page and the reference change before the API does.

Get a key

In the app, press API access in the sidebar. Under Create a key, say what the key is for and press Create key. The key, which starts clab_sk_, is shown once: press Copy and keep it somewhere safe. Your keys lists your keys by name and when each was last used, and Revoke stops one at once. An account can hold 10 keys. A key cannot make or revoke keys or change the account: that is done in the app, signed in.

Use the key

Send the key with every request as Authorization: Bearer clab_sk_…, and keep it out of code you share by putting it in an environment variable. This request asks who the key acts as, so a script can check its key before it spends anything:

export CONFORMERLAB_API_KEY=clab_sk_…
curl -H "Authorization: Bearer $CONFORMERLAB_API_KEY" https://conformerlab.com/auth/me

A study, start to finish

A script takes the same steps as the app, and what runs is exactly what was priced:

1. Upload the structure (POST /structures), which answers the file's fingerprint.
2. Plan (POST /plan) with your question, as you would type it, and the structure's details. The answer is the plan, its findings (anything that blocks it, with a priced fix where there is one) and its price. Planning never charges for compute time.
3. Change the plan, if you want to (POST /plan/{plan_hash}/edit), which answers a new plan with a new price.
4. Approve (POST /approve) with the plan's hash, the price you accept and an idempotency_key (see Retrying safely, below). The price is held, and the answer names the run.
5. Follow the run (GET /runs/{run_id}) until it is complete, asking every minute or so.
6. Read the results: the analysis, the files, the report and what was charged, from the run's own addresses.

The same steps in Python, with requests:

import os, time, requests

API = "https://conformerlab.com"
HEADERS = {"Authorization": f"Bearer {os.environ['CONFORMERLAB_API_KEY']}"}

def call(method, path, **kw):
    r = requests.request(method, API + path, headers=HEADERS, timeout=60, **kw)
    r.raise_for_status()
    return r.json()

text = open("1ake.pdb").read()
sha = call("POST", "/structures", json={"text": text, "filename": "1ake.pdb"})["sha256"]

plan = call("POST", "/plan", json={
    "prompt": "Does the LID of adenylate kinase swing open without its inhibitor? 100 ns, three replicas.",
    "receptor": {"source": "rcsb", "identifier": "1AKE", "residues": 214,
                 "chains": [{"id": "A", "residues": 214}, {"id": "B", "residues": 214, "keep": False}],
                 "structure_sha256": sha},
})
if plan["blocked"]:
    raise SystemExit([f["message"] for f in plan["findings"] if f["severity"] == "error"])

quote = plan["quote"]
print(f"{quote['total_credits']} credits, about {quote['wall_clock_hours']} hours")
run = call("POST", "/approve", json={
    "plan_hash": quote["plan_hash"], "accepted_credits": quote["total_credits"],
    "title": "AdK LID opening", "idempotency_key": "adk-lid-2026-10-02",
})

while True:
    view = call("GET", f"/runs/{run['run_id']}")
    if view["state"] in ("complete", "failed_refunded", "cancelled"):
        break
    time.sleep(60)

results = call("GET", f"/runs/{run['run_id']}/results")
for f in call("GET", f"/runs/{run['run_id']}/files")["files"]:
    print(f["name"], f["url"])

Docking, screening, the deeper pocket search and structure prediction follow the same pattern: a free check that says what it read and the price, an approval that holds the price, then the run to follow and its results. A structure's contacts are a free request, and exact contacts a priced one.

Retrying safely

A request that times out may still have succeeded, and an approval that succeeds holds credits and starts a run. So give each approval an idempotency_key of your choosing, such as the study's name in your own system: sending the same key for the same plan again answers with the run that already started, and starts nothing new. Without a key, every approval starts a run.

Errors

A refusal comes as JSON with a detail field: a sentence saying what was refused, or an object whose error is that sentence, often with a fix and sometimes a technical line (the raw reason, for whoever debugs it). A value the API does not accept answers 422 the same way, naming the value in words. The commonest are 401, a missing or wrong key; 402, not enough credits; 404, a plan or run that is not yours or does not exist; 409, a plan or price that changed since you saw it, or a structure that was not uploaded; 422, a blocking finding or a value out of range; and 429, too many requests, with a Retry-After header saying when to try again.

Limits that matter to a script

• Limits are per account, not per key: 30 planning calls an hour and 200 a day, and 10 runs going at once. Runs start in turn, shared fairly between accounts.
• Plans and uploaded structures are kept for 30 days; after that, upload and plan again. Approved runs are kept.
• Download links last an hour.
• Call the API from a script or a server, not from another website's JavaScript.
• There are no webhooks yet, so a script asks for a run's state.

The app's API page (API access in the sidebar) links to The reference (OpenAPI), which lists every address with every field it takes and answers, at /openapi-docs; the schema itself is at /openapi.json.

Limits and what is not built yet

The app says what it cannot do rather than pricing it. These are the current limits.

Not yet available

Docking beyond one site at a time is not yet available. Docking places one small molecule, peptide or protein, a library of up to 200 small molecules, or a screen of up to 20,000 (1,000,000 in a fast screen), into one site at a time. Testing the top poses is for small molecules only.

A file of compounds cannot yet be simulated directly. Compounds attached as a file reach a simulation only by docking, on their own or as a library whose hits you send on.

CHARMM36m cannot yet be used with molecules from the structure: the plan blocks and offers ff19SB with OPC water, or running without them (see Ligands).

MM-GBSA, an estimate of binding energy from a simulation, is not yet available: it is refused rather than priced, and the plan offers Continue without the step.

Free energy perturbation is in validation, and not enabled on accounts.

Predictions from aligned sequences are not yet available. A structure is predicted from its own sequence alone, so a protein with few relatives is predicted less surely (see Predicting a structure). Docking an aptamer to its target waits too; meanwhile the two are predicted together.

Metals bonded to the protein or in a molecule are not yet simulated as such. A metal is simulated as a charged ion only, and molybdenum and tungsten not at all. A molecule with a metal bonded in it (heme, chlorophyll, cobalamin, cisplatin) can only be removed, from a simulation and from a docking.

Sugars are not yet simulated: glycans can only be removed, and residues the file marks as unknown must be.

Capped chain ends are not yet available: the ends of each chain are simulated charged, never capped; a plan that asks for capping is warned.

Other simulation programs are not yet available: every run uses one program, and the setup files are written for it.

PCA and clustering are not computed; asked for in the request, they are named under Not in this plan:.

Other methods and analyses are named back as not in the plan: enhanced sampling, replica exchange, constant-pH simulation, coarse-grained models, membranes, micelles and nanodiscs, mutagenesis, solvent-accessible surface, water-site analysis, nucleic-acid geometry, Markov state models, and analysing a trajectory you already have. Every run does measure hydrogen bonds and every other interaction, with the share of the time each holds (Results).

To change your name or email address while signed in, write to info@conformerlab.com from the address the account uses.

Rate limits and caps

• Reading a request, planning and both assistants share a budget of 30 requests an hour and 200 a day per account. Over it, the app says so; your runs are unaffected.
• Each account can have up to 10 runs going at once; approving an eleventh is refused with too many runs at once until one finishes or is stopped. Runs start in turn, the next free place going to the account with the fewest running, so one heavy user cannot hold the others up; a waiting run shows its place in the line (Queued — you are next, or how many are ahead).
• After several failed sign-ins, sign-in to that address pauses for a while, even with the right password. Codes by email are limited in the same way.
• One draft study at a time, held in the browser tab. Approved runs are kept on the server.
• Reports, the trajectory player, a run's 2D map, the results assistant and paid storage exist only for completed runs. A structure's or a docked pose's own contacts can be read at any time (Contacts in a structure).
• Some sites only plan and quote. The start screen then says This site can plan and price studies, but it can't run them yet. Nothing you approve here will be charged. New runs can also be paused: New runs are paused for a moment by ConformerLab. Planning and quoting still work.

Troubleshooting

If the chat asks where the structure should come from although your sentence named a PDB ID, the request could not be read (too many requests, no credits, or a network error). Type the ID in the box it offers.

If the plan says it came from ConformerLab's rules, either your email address is not confirmed yet, or the AI planner's drafts kept a setting the checks refuse; the plan was built from what your sentence states. Check what it read, then change a setting on the plan page or ask again in other words.

If you asked for something not offered, such as MM-GBSA, it is named on the line beginning Not in this plan:, and nothing is charged for it. What runs, step by step shows what will run.

If your sentence was read differently from what you meant, look after I read:, where a plan made from your sentence says what it read. Fix any value on the plan page, or say it again in the chat.

If “apo” or “without ATP” did not remove a molecule that is in the structure file, choose to run without it when the chat asks about it, or upload a file with it taken out.

If the plan card's button stays on Approving… after an error, approve again from the plan page (See every setting).

If approving answers “unknown plan”, the plan is older than 30 days or was lost. Plan again from the chat.

If a value you typed on the plan page is refused, it is outside the allowed range in the table above. Clearing a box sends 0.

If none of the depth presets is highlighted, you have edited the plan since it was made; clicking a preset now discards your edits.

If an mmCIF file shows 0 unresolved gaps, gaps are not counted from mmCIF files; the PDB-format file of the same entry shows them. A modified residue, such as phosphoserine, may be listed as a bound molecule and asked about.

If your own file is tagged Predicted, the file does not say how the structure was determined, so it is taken as a computed model. The plan carries a warning and otherwise runs as planned.

If a shorter plan offered for your balance still does not fit, the settling time before measurement is still priced: take one of the choices offered again, or shorten the equilibration on the plan page.

If the credits or the report card did not update after a run finished, reload the page.

If sha256sum -c reports two missing files, they are the methods and readme, which the download carries as Word, PDF and text versions instead; add --ignore-missing.

If the 2D map keeps working out the interactions, the playable trajectory could not be prepared: its files were deleted, or the 3D viewer could not load. The contacts list beside it still works.

If a downloaded 2D figure has a dark background, switch the app to its light theme before downloading.

If the “time left” on a run keeps growing, replicas are waiting to start; the estimate counts only those running.

If approving answers “too many runs at once”, this account already has 10 runs going. Wait for one to finish, or stop one, and approve again; the plan is kept.

If sign-in is refused with the right password, too many failed attempts paused sign-in for a while. Wait, then try again.

If the code does not arrive, or has expired, look in the spam folder for mail from runs@conformerlab.com, then press Send a new code. Only the newest code works.

If approving says to confirm your email address first, enter the code from the email in the banner at the top of the app; the plan is kept.

If you are sent to sign in in the middle of a study, your session ended. The draft in that tab is lost; approved runs are not.

Privacy

The privacy policy says in full what we collect, why, with whom it may be shared, where it is kept and for how long. In short:

Your studies are yours. Plans, runs, files, results and credits answer only to the account that made them; to any other account they read as unknown, the same as a run that does not exist. Your password is stored only as a salted hash, and a code we email you only as a keyed hash, so neither can be read back. Signing in with Google gives us your name, your email address and Google's identifier for your account, never your Google password. Signing out ends that session on the server, so a copy of its cookie anywhere else stops working too. An API key is shown once and stored only as a hash; revoking it stops it at once, and the only thing recorded about its use is when it was last used. Notices and usage records are stored under a hash of your email rather than the address itself, and your email is used to send notices only while the email checkbox in the notifications panel is on.

Structure files are read in your browser; a structure is uploaded to the server when you approve a run, and the run can reach only its own files. A file goes to an outside scientific service only when you choose to send it. Your question, the plan and your questions to the assistants are sent to the AI model behind the AI planner and the assistants, run by a third-party provider; a request the model declines is not charged. When you rate a reply, the rating is kept with the question and the reply, and only the ConformerLab team reads it.