PLATE-VS: a web server for protein-ligand assay curation and cross-target virtual screening datasets

Area of Science:

• Computational chemistry
• Drug discovery
• Bioinformatics

Background:

• High-throughput screening generates vast amounts of data, but creating reliable datasets for virtual screening and machine learning is challenging.
• Existing resources often lack integrated protein structural information, ligand activity data, and property-matched decoys, hindering robust model development.
• The bottleneck in obtaining clean, well-curated protein-ligand datasets with appropriate train/test splits limits the advancement of predictive modeling in drug discovery.

Purpose of the Study:

• To introduce PLATE-VS (Protein-Ligand Affinity-based Target Evaluation-Virtual Screening), a free, openly accessible web server.
• To provide researchers with ready-to-use, high-quality datasets for virtual screening and molecular machine learning.
• To address the limitations of existing resources by integrating diverse data types and offering principled data splitting strategies.

Main Methods:

• Integration of protein structural information, ligand activity data, and property-matched decoys.
• Development of a web server for querying, inspecting, and downloading curated datasets.
• Implementation of principled train/test splitting strategies to create datasets of varying difficulty.
• Harmonization of activity data and curation of metadata for enhanced usability.

Main Results:

• PLATE-VS provides comprehensive, training-ready datasets for protein-ligand interactions.
• The server offers interactive assay summaries, curated metadata, and stratified panels of protein-ligand complexes.
• Downloadable split tables facilitate reproducible research and model validation.
• The resource supports a range of dataset difficulties, from high similarity to challenging cross-target generalization.

Conclusions:

• PLATE-VS effectively addresses the bottleneck in generating high-quality datasets for virtual screening and molecular machine learning.
• The server's integrated approach and principled data splitting enhance the reliability and applicability of machine learning models in drug discovery.
• PLATE-VS serves as a valuable resource for studying binding activity relationships and advancing the field of computational drug design.