Ligand-based 3D Virtual Screening

Lig3DLens

Bill Tatsis, Matt Seddon, Dan Mason, Dan O'Donovan, Gio Cincilla, Azedine Zoufir and Nath Brown

Lig3DLens performs the following tasks:

Prepares a commercial compound library to be used for a VS campaign. This task involves: i) compound standardisation and ii) filtering out compounds outside a predefined range of physicochemical properties.
Generates conformers for all of the compounds in the commercial library and calculates their 3D similarity (shape & electrostatics) to a reference compound.
Finally, it can cluster the highest scoring hits and select a set number of representative compounds that can be ordered and tested.

Installation

python -m pip install -r requirements.txt .

Running a ligand-based 3D VS campaign

Prepare a chemical library for a 3D VS campaign

Note In order to keep track of the library cmpds the input file should have a column containing the text "ID"

lig3lens-prepare --in input_SD_file --filter physchem_yaml_file --out output_SD_file

Generates 3D conformers for both the library and reference compounds and scores the library compounds using a 3D shape & electrostatics similarity function to the reference molecule

Note In order to keep track of the library cmpds the input file should have a column containing the text "ID"

lig3dlens-align --ref input_reference_molecule_file --lib input_library_file_name --conf num_conformers --out output_SD_file

Clusters the highest scoring molecules and selects a representative (diverse) set of compounds. The user can input the number of clusters (num_clusters), the fingerprint type (fingerprint_type) and its dimension (fingerprint_dimension) used for the clustering.

lig3dlens-cluster –-in input_SD_file –-clusters num_clusters –-out output_file -–dim fingerprint_dimension -–fp_type fingerprint_type

Development

Use the Makefile commands to help tidy the codebase periodically. The following will reformat the code according to PEP8, and logically sort the imported modules:

make tidy

Tests

Run pytest in lig3dlens directory

pytest tests

Requirements

Note: The whole VS workflow was tested in a linux (ubuntu) environment and this environment variable had to be set: Tell MKL (used by NumPy) to use the GNU OpenMP runtime instead of the Intel OpenMP runtime by setting the following environment variable:

export MKL_THREADING_LAYER=GNU

Future improvements

Compound library preparation:
- Apply a set of structural filters (for example REOS or PAINS) - either remove or flag compounds.
- Provide more autonomy to the drug designer when setting the physicochemical properties filters.
Compound selection:
- Multi-parameter selection of compounds using a score function that includes the 3D score, 2D similarity to the reference compound, and the physchem properties. The aim is to get an even distribution between highly scored cmpds and other properties.
- Select an optimal number of clusters instead of a predefined one (e.g. using Silhouette or affinity propagation methods). Alternatively, using another method for maximum score-diversity selection problem (e.g. Score Erosion algorithm).
- Provide tools to analyse the chemical diversity of the final selection compound set.

Yeungdb/lig3dlens