Test set of 140 complexes for AutoDock-GPU

Set of 140 protein-ligand complexes
===================================

# Overview

The ligands herein vary in the number of atoms and number of rotatable bonds.
This is the full data set used in the following study:
Accelerating AutoDock4 with GPUs and Gradient-Based Local Search
https://dx.doi.org/10.26434/chemrxiv.9702389.v1

# Warning

These structures, both proteins and ligands, were prepared in an automated way
without manual inspection. The following is a non-comprehensive list of
issues that may exist:

* missing water molecules that bridge ligand-receptor interactions,
* missing atoms in the proteins,
* non-integer sum of partial charges,
* incorrect protonation state,
* incorrect protein conformations.

This intended use of this data is to evaluate the performance of docking with
regard to computational and algorithmic efficiency, but not the accuracy of
the scoring function.

# Details

In ligand\_properties.csv
* `pdb` Protein Data Bank accession code
* `n_atom` number of atoms in the ligand
* `n_tors` number of bonds in the ligand that can rotate during docking.
* `score_of_probable_global_minimum` lowest (best) score (score is the sum of intra- and inter-mo
leculer energy components). The best score did not improve with an increase in the search effort
* `RMSD_of_probable_global_minimum` RMSD from X-ray pose for the solution with the lowst score  
* `best_score_so_far` lowest score ever found but even lower scores are likely to exist