ML-PLIC

ML-PLIC automates characterization of protein–ligand interactions and generates machine learning-based scoring functions to improve virtual screening and identification of potential binders.

Key Features:

• Automated Characterization of PLI: Computes protein–ligand interaction representations using physical and biochemical terms such as energy terms and interaction fingerprints.
• Machine Learning-Based Scoring Functions (MLSFs): Trains MLSFs to identify potential binders for given protein targets and to score ligand binding.
• Modular Architecture: Comprises Docking, Descriptors, Modeling, Screening, and Pipeline modules that perform ligand docking, generate interaction fingerprints, train MLSFs on descriptors and docking data, conduct ML-based virtual screening, and integrate these steps into a cohesive workflow.
• Validation and Performance: Validates MLSFs across benchmark datasets Directory of Useful Decoys-Enhanced (DUD-E), Active as Decoys (AAD), and TocoDecoy, reporting performance superior to traditional docking tools and competitive with deep learning-based scoring functions.
• Case Study Application: Demonstrates application to Serine/threonine-protein kinase WEE1 by developing MLSFs using the ML-based virtual screening pipeline.

Scientific Applications:

• Scoring function development: Development and evaluation of ML-derived scoring functions for protein–ligand binding assessment.
• Virtual screening and binder identification: ML-based virtual screening to identify potential small-molecule binders for target proteins.
• Drug discovery and design: Prediction of ligand binding affinity to inform selection of potential drug candidates and structure-based design decisions.

Methodology:

Performs ligand docking; computes energy terms and interaction fingerprints as descriptors; trains machine learning scoring functions on descriptors and docking data; conducts virtual screening with trained MLSFs; validates performance on DUD-E, AAD, and TocoDecoy.