Molecular Docking in the AI Era

Companion code for Molecular Docking in the AI Era: A Hands-On Handbook for Structure-Based Drug Discovery, by Suryaprakash Tripathy.

Twenty-five chapters ship code, one folder each, in one repository. Clone it once and you have everything in the book. Each chapter stands alone: open its folder, run run.sh, and the numbers printed in that chapter come out.

Every number in the book is asserted by a test

The book prints measured values rather than illustrative ones. Those values live in the repository as expected results, and the test suite asserts them, so a divergence between page and code shows up as a failing build rather than as a reader's afternoon.

Two hundred and fifty-three tests run from a fresh clone and the suite passes, on Windows and on Ubuntu. A few are marked expected-to-fail where different RDKit and AutoDock Vina builds produce slightly different coordinates — which few depends on the platform, and the reason is attached to each. Nothing is tuned to match.

Three things that silently ruin a result

--minimize on an RMSD call superimposes before measuring, so every redock passes. Vina's default seed of 0 means random, not zero, so an unseeded run cannot be repeated. And PDBQT drops formal charge and reorders atoms, so keep an SDF as the reference copy. None of the three produces an error.

One system, all the way through

Every chapter from 5 onward uses AmpC β-lactamase from Escherichia coli (UniProt P00811): four public crystal structures, a congeneric series of inhibitors with measured affinities, and five published prospective screening campaigns with experimental follow-up. Everything that can be reproduced from public files, can be.

The chapters

• 2 · Choosing a method
  Cost comparison across docking, co-folding and free energy methods. dock.nexagenlabs.com/ch02
• 3 · Databases
  Provenance recorder, and a check for whether your target is in PDBbind. dock.nexagenlabs.com/ch03
• 4 · File formats
  Format round-trips, measured. What PDBQT loses and PDB does not. dock.nexagenlabs.com/ch04
• 5 · Receptor preparation
  Receptor QC: gaps, altlocs, ligand copies, bridging waters, buffer. dock.nexagenlabs.com/ch05
• 6 · Predicted structures
  Per-residue confidence for a predicted structure's binding site. dock.nexagenlabs.com/ch06
• 7 · Defining the pocket
  Box definition from a reference ligand, with the derivation recorded. dock.nexagenlabs.com/ch07
• 8 · Ligand preparation
  Protonation, stereochemistry round-trip, conformer generation. dock.nexagenlabs.com/ch08
• 9 · Your first docking run
  A reproducible run: the seed, the box, and what silence looks like. dock.nexagenlabs.com/ch09
• 10 · Flexible residues
  Side-chain torsions across structures: which residues actually move. dock.nexagenlabs.com/ch10
• 11 · Web servers
  Turning a server result into something a protocol record can hold. dock.nexagenlabs.com/ch11
• 12 · Screening
  Parallel screening with checkpointing, and the filter cascade. dock.nexagenlabs.com/ch12
• 13 · Co-folding
  Co-folding submission and confidence parsing for the binding site. dock.nexagenlabs.com/ch13
• 14 · Boltz-2
  Affinity prediction, and the squared-correlation diagnostic. dock.nexagenlabs.com/ch14
• 15 · The co-folding field
  Running two co-folding models and reporting where they disagree. dock.nexagenlabs.com/ch15
• 16 · Rescoring
  Rescoring, and a check for training-set overlap with your target. dock.nexagenlabs.com/ch16
• 17 · Validation
  Redocking and cross-docking, with symmetry-corrected RMSD. dock.nexagenlabs.com/ch17
• 18 · Enrichment
  Two screens with identical AUC and opposite early enrichment. dock.nexagenlabs.com/ch18
• 20 · The protocol record
  The protocol record, blank and filled, populated from a run log. dock.nexagenlabs.com/ch20
• 21 · Molecular dynamics
  Why a flat RMSD trace is not a converged one. dock.nexagenlabs.com/ch21
• 22 · Free energy
  Free energy with replicates, and standard errors that mean something. dock.nexagenlabs.com/ch22
• 23 · Interaction fingerprints
  Interaction geometry alongside interaction names. dock.nexagenlabs.com/ch23
• 24 · Network pharmacology
  Enrichment that refuses to run without an explicit background. dock.nexagenlabs.com/ch24
• 25 · Hit to bench
  Dose-response fitting that reports the Hill slope beside the IC₅₀. dock.nexagenlabs.com/ch25
• 26 · Case study
  Three reproduction tests with published answers. dock.nexagenlabs.com/ch26
• 27 · The methods section
  Methods paragraphs drafted from a protocol record. dock.nexagenlabs.com/ch27

Chapters 1 and 19 ship no code, so there is no /ch01 or /ch19. If you typed one, you are in the right place already.

Versions

Parts of this book date faster than the rest. Model releases, licences and tool versions are listed with a dated changelog at dock.nexagenlabs.com/versions. Chapters 13 to 16 point here for a reason.

References

The reference list as machine-readable data, with a DOI on every entry that has one: dock.nexagenlabs.com/references

Errata

Corrections to the book and to the code are listed on the errata page. If you have found an error in either, please report it: the page says how.