PhenoScreen

PhenoScreen leverages contrastive learning to connect chemical space with cellular morphological profiles, enabling accurate prediction of compound-induced phenotypic changes for phenotypic drug discovery.

Key Features:

• Dual-space contrastive learning: Connects chemical perturbations to cellular morphology through large-scale compound-phenotype association data.
• Lead compound identification: Accurately identifies lead compounds that induce user-defined phenotypic shifts with novel scaffolds using different levels of phenotypic information.
• Cross-cancer generalization: Demonstrates strong generalization across cancer cell types including osteosarcoma and rhabdomyosarcoma.
• Virtual screening capability: Supports compound virtual screening for phenotypic drug discovery.

Scientific Applications:

• Phenotypic drug discovery: Screens compounds in cellular models representing disease-relevant phenotypes as an alternative to target-based approaches.
• First-in-class drug discovery: Identifies compounds exerting therapeutic effects through multiple targets and mechanisms for diseases with poorly understood molecular mechanisms or lacking validated therapeutic targets.
• Polypharmacology exploration: Uncovers multi-target drugs by enabling broader exploration of biological systems.
• Osteosarcoma screening: Predicts active compounds inducing user-specified phenotypes with varying inhibitory effects in osteosarcoma phenotypic models.
• Rhabdomyosarcoma screening: Identifies active compounds with IC50 values up to 1.842 μM in rhabdomyosarcoma models.
• Novel therapeutic pathway identification: Accelerates drug discovery by identifying novel therapeutic pathways and increasing diversity of viable drug candidates.

Methodology:

PhenoScreen utilizes contrastive learning to connect chemical space with cellular morphological profiles through large-scale compound-phenotype association data.