Cell Morphology-Guided De Novo Hit Design by Conditioning GANs on Phenotypic Image Features

Explore a groundbreaking approach to de novo hit design in drug discovery through a 49-minute talk on cell morphology-guided molecule generation using conditional GANs. Learn how cellular morphology data can directly inform the creation of novel bioactive compounds, potentially revolutionizing the time-consuming and costly process of drug development. Discover the methodology behind training a generative model on 30,000 compounds using cell painting morphological profiles as conditioning factors. Examine evidence for targeted generation of known agonists based on gene overexpression profiles, despite the absence of explicit biological target information during training. Delve into model evaluation, interpolation experiments, and the potential for this target-agnostic approach to facilitate knowledge transfer between biological pathways. Gain insights into the future applications of this proof-of-concept for hit generation in drug and phytopharmaceutical discovery, as well as chemical safety assessment.

- Intro
- Using Profiling Technologies to Guide the Design of Small Molecules
- Model Evaluation
- Can the Model Generate Bioactives for Specific Targets?
- Interpolation Experiments
- Conclusion & Outlook
- Q&A