MolAICal

MolAICal generates drug-like ligands within protein binding sites using a dual-module combination of genetic algorithms, deep learning, and molecular docking to support structure-based drug discovery.

Key Features:

• Dual-Module Architecture: Comprises two modules: the first integrates a genetic algorithm with a deep learning model trained on fragments from FDA-approved drugs and the PDBbind database and fits Vinardo scores to optimize ligand binding; the second employs a deep learning generative model trained on drug-like molecules from the ZINC database combined with molecular docking via Autodock Vina.
• Comprehensive Filtering Mechanisms: Applies Lipinski's rule of five, Pan-assay interference compounds (PAINS) filters, synthetic accessibility (SA) scoring, and user-defined rules to remove undesirable ligands.
• Demonstrated Target Applications: Shown for a membrane protein (glucagon receptor) and a non-membrane protein (SARS-CoV-2 main protease), generating diverse novel ligands with favorable binding scores and appropriate XLOGP values.

Scientific Applications:

• Structure-based drug discovery: Generation of candidate molecules directly within protein active sites using 3D structural information combined with docking and scoring.
• Targeting complex proteins: Application to membrane proteins and viral proteases, exemplified by the glucagon receptor and SARS-CoV-2 main protease.

Methodology:

Training deep learning models on datasets from PDBbind and ZINC, integrating a genetic algorithm to explore ligand space, and performing automated molecular docking with Autodock Vina alongside Vinardo score fitting.