PharMD is a tool to retrieve pharmacophore models from MD trajectories of protein-ligand complexes, identification of redundant pharmacophores and virtual screening with multiple pharmacophore models using different scoring schemes.

mdtraj >= 1.9.3
plip >= 1.4.2
pmapper >= 0.3.1
psearch >= 0.0.2

pip install pharmd

To retrieve individual snapshots of MD trajectory mdtraj package is used. Therefore the md2pharm utility takes the same arguments as mdconvert utility from mdtraj. Thus you may extract only specified frames not all of them. You have to specify ligand code as it is given in PDB topology file. Individual frames will be stored in a single PDB file without solvent molecules. Pharmacophore models for each frame in xyz-format will be stored in the same directory as output pdb-file.

md2pharm -i md.xtc -t md.pdb -s 10 -g LIG -o pharmacophores/frames.pdb

Similar pharmacophores are recognized by identical 3D pharmacophore hashes. It is expected that pharmacophores with identical hashes would have RMSD less than the specified binning step. By default binning step equals to 1A. Pharmacophores with distinct hashes are stored in a specified directory. Optionally one may provide a path where to store hashes for al pharmacophores.

get_distinct -i pharmacophores/ -o distinct_pharmacophores/

screen_db utility from psearch package is used for this purpose. Therefore you have to generate database of compound conformers and their pharmacophore representations using utilities from psearch package. At this step you may specify a desired binning step value which will be used further in screening (default is 1).

prepare_db -i input.smi -o compounds.db -c 2 -v 

If you would like to calculate scoring based on Conformer Coverage Approach you have to specify --conf argument for screen_db. Then all conformers of a compound matching pharmacophore models will be retrieved as hits (may be slower). Otherwise only the first matching conformer will be returned.

It is recommended to restrict screening to complex pharmacophores having at least four features, because less complex models would retrieve many irrelevant compounds.

screen_db -i compounds.db -q distinct_pharmacophores/ -o screen/ --conf -c 2 -f 4

Multiple txt-files will be created in the output directory containing hit lists retrieved by individual pharmacophore models.

The advantage of ensemble scoring is that you do not need validate individual models and select best performing ones. Ensemble scoring is calculated by:

1. Conformer Coverage Approach (CCA) - the score is equal to the percentage of conformers matching at least one of supplied pharmacophore models.
2. Common HIts Approach (CHA) - the score is equal to the percentage of models matched at least one conformer of a compound.

In the case of CCA scoring you have to supply the database of screened compounds as an additional parameter.

get_scores -i screen/ -o cca_scores.txt -s cca -d compounds.db

All utilities have -h option to get help pages with descriptions of all available arguments.

Virtual Screening Using Pharmacophore Models Retrieved from Molecular Dynamic Simulations
Pavel Polishchuk, Alina Kutlushina, Dayana Bashirova, Olena Mokshyna, Timur Madzhidov
Int. J. Mol. Sci. 2019, 20(23), 5834
https://doi.org/10.3390/ijms20235834

Currently there is an issue with installation of dependencies. plip requires openbabel which causes an error during installation via pip. Therefore it is recommended to solve dependencies manually and use pip install -U --no-deps pharmd to install pharmd ignoring dependencies.

BSD-3 clause