MCPINN

Multi-channel_PINN is a machine learning framework designed to predict compound-protein interactions (CPI) for drug discovery applications. It employs a multi-channel neural network architecture that processes molecular and protein sequence embeddings to identify potential drug candidates. This approach enables more accurate and scalable prediction of interactions between chemical compounds and target proteins.

Key Features

• Multi-channel Architecture — Combines separate neural network channels for compound and protein feature extraction before joint prediction.
• Embedding-based Inputs — Utilizes Mol2vec for molecular embeddings and ProtVec for protein sequence embeddings as input representations.
• Transfer Learning Capability — Designed to be transferable across different protein families and compound libraries.
• Scalable Design — Framework supports large-scale screening of compound-protein pairs for drug discovery pipelines.

Philosophy

The project emphasizes a modular, embedding-driven approach that separates feature learning for compounds and proteins, allowing the model to capture domain-specific patterns while remaining generalizable across biological contexts.