Learning Molecular Representations from Cellular Phenotypes with Structure Preservation

Short summary

PhenMol is a structure-preserving framework for phenotype-aware molecular representation learning that disentangles molecular and cellular representations into shared and private components. It preserves chemical neighborhood organization while integrating cellular phenotype information, addressing distortion issues in existing multimodal alignment methods. Evaluated on ~30K molecule-cell morphology pairs across 270 bioactivity tasks, PhenMol improves molecular property prediction, retrieval, and clinical trial outcome prediction while better preserving molecular neighborhoods.

• •PhenMol disentangles molecular and cellular representations into shared and private components
• •Preserves chemical structure neighborhoods while integrating cellular phenotype data
• •Improves performance across 270 bioactivity tasks, retrieval, and clinical trial prediction on ~30K pairs

Generated with AI, which can make mistakes.