Skip to content

Latest commit

 

History

History
101 lines (80 loc) · 4.65 KB

README.md

File metadata and controls

101 lines (80 loc) · 4.65 KB
Raw

Binder recovery benchmark

This directory contains the code required to re-run our computational benchmark in which we aim to recover known binding partners from drug-induced protein complexes. To this end, we split 14 ligand-induced protein complexes into two subunits each and search against a database of these 28 proteins plus 200 decoys. Since the ligand can be considered as a part of each subunit, we have 28 benchmarking cases. The search can be highly parallelised and should only take a couple of minutes.

Summary of targets

  • 6QTL : Caffeine recognizing nanobody
  • 7DC8 : Switch Ab Fab and hIL6R in complex with ATP
  • 4DRI : FKBP51, Rapamycin and the FRB fragment of mTOR
  • 6OB5 : Computationally-designed, modular sense/response system (S3-2D)
  • 1A7X : FKBP12-FK1012 COMPLEX
  • 6ENG : 43K ATPase domain of Escherichia coli gyrase B in complex with an aminocoumarin
  • 6H0F : DDB1-CRBN-pomalidomide complex bound to IKZF1(ZF2)
  • 6SJ7 : human DDB1-DDA1-DCAF15 E3 ubiquitin ligase bound to RBM39 and Indisulam
  • 7TE8 : CA14-CBD-DB21 ternary complex
  • 1TCO : CALCINEURIN A FRAGMENT, CALCINEURIN B, FKBP12 AND THE IMMUNOSUPPRESSANT DRUG FK506 (TACROLIMUS) (Use chain C vs AB)
  • 1S9D: ARF1 IN COMPLEX WITH BREFELDIN A AND A SEC7 DOMAIN (Ignore GDP)
  • 4MDK : Cdc34-ubiquitin-CC0651 complex
  • 3QEL : NMDA receptor subunit GluN1 and GluN2B in complex with ifenprodil
  • 6N4N: Crystal structure of the designed protein DNCR2/danoprevir/NS3a complex

benchmark_pdbs.txt contains a summary of all benchmark complexes, a definition of the two subunits, and the relevant small molecule. The format is PDB ID, chain 1, chain 2, drug, SDF name. The drug is specified as <3_letter_code>_<chain>.

Each PDB can be fetched from RCSB together with the "Instance coordinates" file in SDF format (using the online interface or a command like this:

https://models.rcsb.org/v1/6qtl/ligand?auth_seq_id=201&label_asym_id=J&encoding=sdf&filename=6qtl_J_CFF.sdf

Decoys

The decoys are taken from a list of known PPIs without small molecules at the interface. The full list is provided in masif_ppi_search_benchmark_list.txt.

Preparing the structures

Prior to the search, we need to preprocess all target complexes as well as the database proteins. This includes surface triangulation, feature calculation, MaSIF-site interface prediction and computation of MaSIF-search descriptors. Every subunit of the target complexes is processed individually and together with the drug. An already processed version of the dataset is provided on Zenodo.

Targets

A helper script generates all the required processing commands for the target complexes (to process each subunit with and without the small molecule).

python generate_preprocessing_commands.py benchmark_pdbs.txt <structure_dir> <processed_dir> preprocessing_commands.sh

<structure_dir> should contain all the PDBs and SDFs of the benchmark complexes.

All commands can be executed like this:

source preprocessing_commands.sh

The outputs will be written to <processed_dir>.

NOTE: Two benchmark complexes (7TE8/P0T_C and 6ENG/BHW_B) unfortunately required manual intervention because the automatic protonation and mol2 conversion produced inconsistent results. We will provide further instructions or the processed files upon request.

Decoys

We provide a script that parallelises the downloading and preprocessing of the decoy proteins using slurm:

sbatch decoys/prepare_decoys_parallel.slurm decoys/masif_ppi_search_benchmark_list.txt <decoy_dir>

All outputs will be saved in <decoy_dir>.

Running the search

To finally run a search, we first create a parameter file for each target (with ligand) in benchmark_pdbs.txt using:

python make_param_files.py <structure_dir> <processed_dir> <decoy_dir> <search_output_dir>

This script creates the output folder structure including parameter files in <search_output_dir>. The parameters can be changed by modifying the template in make_param_files.py and running it again.

The search & docking step can then be launched in parallel using the newly created parameter files, for example:

sbatch run_search.slurm <search_output_dir>/masif-neosurf-benchmark/with_ligand/search_params_targets

This step has to be repeated for all search_params_* folders (4 folders should have been created in the previous step). When all searches are completed, the hits and docked PDB structures can be found in <search_output_dir>. Some helper functions for further analysis are included in analysis_utils.py.