Virtual Screening Protocol
In this document, the current Rizzo group protocol will be described in detail. This protocol has be through successive iterations and has be used to select compounds for Flu and HIVgp41 (n=112).
ZINC is a free, online database of commercially-available compounds. In June 2009, the "big" vendor libraries at a pH of 7 were downloaded and processed into chunks of approximately 100,000 molecules each. The ZINC molecules come with AMSOL charges already determined.
These chunks are then processed through MOE and sorted by rotatable bonds in ascending order. These processed chunks are available on BG, path=~pholden/RCR/projects_ZINC8/ZINC8, or on ringo, path=/media/sdb1/pholden/ZINC8.descending.rot.bonds. The lab is currently using the ChemDiv library for its virtual screen.
Downloading and Preparation of Receptor
Preparation of Reference Molecule for Footprint Rescoring
The reference molecule is the molecule whose footprint guides the selection of molecules during footprint rescoring. The reference molecule will most likely be a native substrate or known inhibitor. The molecule should be prepared with charges unless it already has charges. These charges can be applied using MOE or antechamber (AmberTools). The charge model selected is not of great importance, as per Sudipto's testset paper.
Once the molecule is prepared, it should be minimized in the receptor using a tethered minimization.
An example input file:
ligand_atom_file ligand.mol2 limit_max_ligands no skip_molecule no read_mol_solvation no calculate_rmsd yes use_rmsd_reference_mol no use_database_filter no orient_ligand no use_internal_energy yes internal_energy_rep_exp 12 flexible_ligand no bump_filter no score_molecules yes contact_score_primary no contact_score_secondary no grid_score_primary no grid_score_secondary no dock3.5_score_primary no dock3.5_score_secondary no continuous_score_primary yes continuous_score_secondary no cont_score_rec_filename receptor.mol2 cont_score_att_exp 6 cont_score_rep_exp 12 cont_score_rep_rad_scale 1 cont_score_use_dist_dep_dielectric yes cont_score_dielectric 4.0 cont_score_vdw_scale 1 cont_score_es_scale 1 gbsa_zou_score_secondary no gbsa_hawkins_score_secondary no amber_score_secondary no minimize_ligand yes simplex_max_iterations 1000 simplex_tors_premin_iterations 0 simplex_max_cycles 1 simplex_score_converge 0.1 simplex_cycle_converge 1.0 simplex_trans_step 1.0 simplex_rot_step 0.1 simplex_tors_step 10.0 simplex_random_seed 0 simplex_restraint_min yes simplex_restraint_coefficient 10.0 atom_model all vdw_defn_file /sbhome0/sudipto/RCR/projects_BNL/parameters/vdw_AMBER_parm99.defn flex_defn_file /sbhome0/sudipto/RCR/projects_BNL/parameters/flex.defn flex_drive_file /sbhome0/sudipto/RCR/projects_BNL/parameters/flex_drive.tbl ligand_outfile_prefix xtalmin write_orientations no num_scored_conformers 1 rank_ligands no
The version of DOCK currently being used for virtual screening is dock6_09-09-08.footprint. The most up to date script is Trent's modifications, which include the database filter. The database filter removes molecules with charges greater than +2 and less than -2, and also molecules with greater than 15 rotatable bonds.
The script also takes the processed chunks from ZINC and splits them into two subsections: the first 60,000 and the remainder. Because the sets are sorted by rotatable bonds, the first 60,000 should dock in fairly quickly, and be completed within the allowable wallclock limit. The second set, or molecule 60,001 and beyond, will have higher numbers of rotatable bonds, and take exponentially more time for each molecule. As per Sudipto's testset paper, we also have less confidence in the molecules with more rotatable bonds. If this second job does not finish, DO NOT restart it. The molecules that did not dock in will have higher numbers of rotatable bonds (should this be revised in light of the database filtering?)
Note that the values for the virtual screen are the default values.
Path for script: ~balius/RCR/projects_BG/screening/run.dock2grid.max_lig_tebmod.csh
Minimization of Poses
In this step, the two multimol2s generated from the virtual screen step are recombined and minimized on the continuous receptor. This step removes artifacts from the lower resolution grid, generally resolving clashes with sidechains. The minimization will be tethered with a 10kcal/mol restraint, such that with each step of minimization, the molecule cannot move too much. The original DOCK pose is more likely to be kept with this method.
Path of script: ~pholden/RCR/projects_BG/screening/HIV/run.minoffgrid.csh
Pose Rescoring using Molecular Footprints
In this step, the minimized molecules are rescoring using the molecular footprints. The footprint is a per-residue decomposition of the interaction energies between the ligand and the receptor. These energies will then be compared to a reference molecule's footprint, prepared earlier, and a numerical comparison between the two will be determined. The possible comparison techniques include a Pearson Correlation, Euclidean, Normalized Euclidean, and a threshold based method, where only values exceeding a certain energy will be compared. The energies compared are van der waal's energy (vdw_fp), electrostatic energy (es_fp), and the number of hydrogen bonds (hb_fp). Combinations of the footprints can also be determined.
The script simply reads in a multimol2 and a reference and reports the per-residue decomposed energies and the actual footprint value.
For Pearson, values close to 1 are best. For Euclidean, values close to 0 are best.