Deep Learning Protocols for Predicting Drug Mechanism of Action and Drug-Target Interactions

Understanding drug mechanisms of action (MOA) and predicting drug-target interactions (DTIs) are fundamental challenges in modern drug discovery and development, hindered by high costs, long development timelines, and limited knowledge of compound activity and molecular targets. Here, we present two deep learning-based computational protocols designed to address these challenges. The first framework employs directed message passing neural networks (D-MPNN) to predict drug MOA from chemical-genetic interaction profiles (CGIPs), by learning how molecular structures perturb biological pathways through systematic profiling across genetically sensitized strains. The second framework, iNGNN-DTI, utilizes interpretable nested graph neural networks combined with pretrained molecule models to predict DTIs, leveraging cross-attention mechanisms to provide insights into binding determinants. We highlight the application of these methods to key therapeutic areas, including antibacterial drug discovery and drug repurposing for COVID-19 therapeutics. Each protocol provides comprehensive guidance on data preparation, model implementation, validation strategies, and result analysis. These computational approaches offer scalable, cost-effective tools for accelerating therapeutic development by bridging chemical structure, molecular interactions, and systems-level biological responses.