BACPI

BACPI predicts compound-protein interactions and estimates binding affinities using an end-to-end bi-directional attention neural network that integrates graph attention networks (GAT) and convolutional neural networks (CNN) for molecular representation.

Key Features:

• End-to-End Neural Network Model: Predicts both compound-protein interactions (CPIs) and binding affinities directly from low-level molecular representations without intermediate handcrafted steps.
• Bi-Directional Attention Mechanism: Employs a bi-directional attention neural network to integrate learned compound and protein representations and focus on regions important for binding.
• Graph Attention Network (GAT) and Convolutional Neural Network (CNN): Utilizes GAT and CNN components to learn detailed representations of both compounds and proteins.
• Performance Evaluation: Evaluated on three CPI datasets and four binding affinity datasets, demonstrating superior performance relative to other machine learning methods on balanced and unbalanced CPIs and to state-of-the-art deep learning approaches on large binding affinity datasets.

Scientific Applications:

• Drug discovery lead identification: Prioritizes potential therapeutic compounds by predicting CPIs and estimating binding affinities for experimental follow-up.
• High-throughput virtual screening: Supports large-scale screening by handling large binding affinity datasets for candidate selection.
• Experimental prioritization: Aids in reducing time and cost of experimental validation by ranking compound-protein pairs for further testing.

Methodology:

Integrates graph attention networks (GAT) and convolutional neural networks (CNN) to extract compound and protein representations and applies a bi-directional attention mechanism within an end-to-end neural network to predict CPIs and binding affinities.