Where does JQ1 bind this protein, and does it stay?

1. The structure, predicted

From the sequence alone, five models from the complex predictor, best first. All five are very confident: pLDDT 95.9, 95.9, 95.8, 95.7 and 95.6; confidence 0.92 to 0.93; pTM 0.79. The best model lies 0.46 Å (Cα, residues 44–168) from BRD4's crystal structure, PDB 3MXF, which the prediction never saw. 9.6 credits, 9 minutes 43 seconds.

2. The pocket, found

The free pocket search on the best model ranks first the acetyl-lysine pocket where BRD4's inhibitors bind: druggability 0.92 out of 1, 872 ų, lined by the WPF shelf (Trp81, Pro82) with Leu92, Asp145, Ile146 and Met149. The next pocket scores 0.08. Free.

3. JQ1, docked

JQ1, from its SMILES, docked into that pocket of the predicted model. The best pose scores −6.5 kcal/mol, against Asn140, Leu92, Pro82, Asp144, Asp145 and Ile146. Docked into the crystal structure instead, the same molecule scores −8.3 and lands 0.51 Å from where the crystal holds it: a predicted pocket, with its side chains as the prediction placed them, is a looser fit. 3 credits, 2 minutes 30 seconds.

4. The complex, simulated

The best pose, solvated (23,106 atoms in OPC water with 150 mM NaCl), ff19SB for the protein and JQ1 parameterized for the simulation, two independent replicas of 20 ns at 300 K after 4 ns of equilibration each. The protein holds its fold in both: backbone RMSD 1.30 and 1.57 Å at the end. Quoted at 88.8 credits, charged 86.46 for the compute it used.

The answer

JQ1 stays in the pocket, but not in the pose it was docked in. It ends 2.5 and 3.9 Å from its docked pose in the two replicas. In the second it reached 6.9 Å, past the 5 Å at which it would count as having left, before coming back. Over the second half of each replica it kept 77% and 73% of the residues it started against.

What holds it, as a share of frames across both replicas: hydrophobic contacts with Ile146 (92%), Leu92 (85%), Tyr97 (81%), Pro82 (67%), Leu94 (65%) and Trp81 (62%), and a water bridge to Asp145 (78%). The hydrogen bond to Asn140 that holds JQ1 in its crystal structures is not among them.

That is the useful result. A pose docked into a predicted pocket moved and found contacts of its own rather than the crystal's, and the simulation is what showed it. The next steps a chemist would take are here too: triage the top poses, run longer and with more replicas, or start from the crystal structure where one exists.

What it cost

Those credits come to about $2.48 at today's pay-as-you-go price, or $1.98 with the largest pack. Planning it (within the AI planner's free monthly credits), the quotes and the report were free.

The report and the files

• The report (PDF): the answer, the charts, the 2D map of what holds JQ1, the table of interactions that last, the quality checks and the methods, as the run made it (but for the name it was prepared for, below).
• The methods (PDF): the protocol and every program and parameter set, with their citations.
• The predicted structure (PDB): the best of the five models.
• The docked complex (PDB): the protein with JQ1 in its best pose, the structure the simulation started from.

The trajectories (28 MB each) and the system to rerun the simulation yourself (about 10 MB) stay with the run, as they do for every study.

Read this first

• Residue numbers. A predicted model is numbered from 1, so the report and the files call Asn140 Asn97 and Trp81 Trp38: add 43 for BRD4's own numbering, which this page uses.
• A short study. Two replicas of 20 ns say what the pose does over tens of nanoseconds, not how tightly JQ1 binds: stability is not affinity.
• Where it ran. On our development service, which runs the same code as this site. The report's "Prepared for" line names a placeholder, John, in place of the account it ran under; nothing else in it was changed.
• The demo on the front page tells the same story from the crystal structure, to keep the page light; this is the study run end to end from the sequence.

Ask your own question