Kidney Week 2025 Early Program - Advances in Research Conference 2025: Generative Artificial Intelligence in Kidney Disease Research and Management-Blending Biology, Chemistry, and AI to Enable Systems Pharmacology

Video Transcription

Video Summary

Patrick Aloy describes how drug discovery struggles because human biology is a complex system: compounds optimized against isolated protein targets often fail when placed back into cellular and organismal contexts. Despite being able to screen more molecules with robotics, hit rates and successful approvals have not improved, suggesting early-stage discovery needs to incorporate biological complexity.

His group’s approach is to “format” both chemistry and biology into machine-readable vector representations. For small molecules, they built the <strong>Chemical Checker</strong>, encoding not just structure but multi-level bioactivity: targets, network/cell responses (e.g., gene expression, cell morphology), and clinical properties (indications, side effects). They also use AI to impute missing descriptors and have released code so others can build descriptors from in-house data.

For biology, they created a large <strong>knowledge graph (“Bioteque”)</strong> connecting genes, cells, tissues, diseases, functions, etc., then learned ~1,000 context-dependent embeddings.

They demonstrate signature matching for Alzheimer’s: identifying approved drugs predicted to reverse disease expression modules in mouse models, with several showing phenotypic and molecular reversal. Finally, they explore generative AI (e.g., VAE + diffusion, reinforcement learning) to design molecules with desired, specific cellular effects, outperforming simple library searching in validation rates. The long-term vision is “personalized pharmacology”: designing drugs tailored to individual patient profiles.

Asset Subtitle

Patrick Aloy

Meta Tag

Module ARC

Speaker Patrick Aloy

Keywords

drug discovery complexity

Chemical Checker

bioactivity descriptors

Bioteque knowledge graph

context-dependent embeddings

Alzheimer’s signature matching

generative AI molecular design