Cell Morphology-guided, Scalable and Controllable Molecule Design via Generative AI - Project Summary This project aims to advance phenotypic drug discovery (PDD) by developing a generative AI-driven platform for innovative, cell morphology-guided, scalable, and controllable small molecule design without relying on drug targets. By integrating tri-modal contrastive learning, controllable morphology generation, and retrieval- augmented molecule design, the platform addresses key PDD challenges, including limited understanding of structure–morphology relationships, inconclusive data generation, and lack of explainability in molecule design. Aim 1 will establish relationships among molecular structures, induced morphological changes, and textual descriptions to create a unified representation space. Aim 2 will develop a scalable and controllable diffusion model to generate morphology images that accurately reflect biological responses to chemical treatments. Aim 3 will implement a retrieval-augmented platform for traceable de novo small molecule design, leveraging external data sources to enhance explainability. This project will enable a target-agnostic approach to drug design, expanding therapeutic options for diseases without identified drug targets.
