Phenotypic AI-based design of cell-specific small molecule cytotoxics

Area of Science:

• Computational chemistry
• Drug discovery
• Artificial intelligence in medicine

Background:

• Traditional drug discovery often overlooks phenotypic effects, focusing instead on chemical properties or protein interactions.
• Exploring vast chemical spaces for targeted drug development remains a significant challenge.
• Developing selective toxicity against cancer cells without harming healthy cells is a key goal in oncology.

Purpose of the Study:

• To present a novel framework for designing small molecules with selective toxicity against pancreatic cancer cells using artificial intelligence.
• To explore the potential of combining predictive and generative AI models for phenotypic drug discovery.
• To demonstrate the design and validation of new chemical entities with cell-specific biological effects.

Main Methods:

• Utilized a large-scale dataset of over 11,000 compounds screened across various cancer and control cell lines.
• Trained predictive artificial intelligence (AI) models on screening data to identify key molecular features and predict compound activity.
• Integrated predictive models into a generative AI system to design novel small molecules with desired selective toxicity profiles.

Main Results:

• The AI framework successfully generated novel small molecules with predicted selective toxicity against pancreatic cancer cells.
• A significant portion of the designed molecules were structurally distinct from existing compounds.
• Experimental validation confirmed that several designed molecules exhibited the intended selective activity, outperforming conventional screening methods.

Conclusions:

• The combination of predictive and generative AI offers a powerful approach for designing compounds with complex biological effects, particularly selective toxicity.
• This AI-driven framework advances phenotypic drug discovery by enabling the design of targeted therapies without prior knowledge of specific molecular targets.
• The study underscores the potential of AI to accelerate the development of novel therapeutics for challenging diseases like pancreatic cancer.