PreBINDS

PreBINDS generates customizable compound–protein interaction (CPI) datasets from ChEMBL to support development and evaluation of deep learning models for CPI prediction and in silico screening.

Key Features:

• Customizable dataset generation: Constructs CPI datasets from the ChEMBL database with adjustable positive and negative activity thresholds for defining labels.
• Activity value visualization: Provides histograms of activity value distributions to inform and refine threshold selection for positive and negative examples.
• Target protein selection: Enables selection of target proteins based on Pfam families, ChEMBL classifications, and sequence similarity criteria.
• Deep learning support: Outputs dataset attributes tailored for training and evaluating deep learning models for CPI prediction and in silico drug design.

Scientific Applications:

• Bioinformatics and computational biology: Generation of curated CPI datasets for computational analysis of ligand–target relationships.
• Pharmacology and drug discovery: Preparation of datasets to support target-based screening and lead identification in drug development.
• Machine learning model development: Creation of labeled datasets and feature sets for training and benchmarking deep learning models for CPI prediction and virtual screening.

Methodology:

Leverages data from the ChEMBL database to construct CPI datasets, applies user-set positive/negative thresholds, selects targets by Pfam, ChEMBL classification, or sequence similarity, and uses histograms of activity values to refine dataset parameters.