This talk summarizes our experience of developing non-human intelligent technologies for drug discovery. We created multiple temporally-validated machine learning (ML) models, and some LLM (large language model) agents to integrate and coordinate drug discovery activities. This platform includes 1) target-phenotype ML models focusing on oncology and neurodegeneration; 2) thousands of multi-task target-based and property-based ML models using proprietary data and fingerprints; 3) multiple LLM agents serving as research assistants for specific drug discovery tasks.

Discover how innovative generative AI and molecular simulation methods are revolutionizing drug discovery. This presentation will explore cutting-edge strategies for hit finding and lead optimization targeting unmet medical needs. Key highlights include AI-based ligand design, active learning absolute free energy perturbation (AQFEP) virtual screening, the Alchemical Transfer Method (ATM) for binding free energy estimation, and IDOLpro—a generative AI solution that integrates deep diffusion with multi-objective optimization.

Unlike conventional methods that focus on single-target selectivity, generative AI models leverage machine learning and deep learning algorithms to explore vast chemical spaces, optimizing molecules for polypharmacology. These models can integrate multi-target profiles, assessing potential off-target effects, efficacy, and safety considerations, ultimately facilitating the creation of compounds with desired therapeutic profiles. This study explores generative modeling for multi-target engagement and highlights its promise to address complex diseases through targeted polypharmacology.