Combining DELs and machine learning for toxicology prediction.

Area of Science:

• Medicinal Chemistry
• Computational Chemistry
• Toxicology

Background:

• DNA-encoded libraries (DELs) are powerful tools for identifying novel chemical matter in drug discovery.
• The extensive experimental data generated by DELs is suitable for machine learning (ML) applications.
• ML excels at modeling intricate relationships between chemical compounds and quantitative biological endpoints.

Purpose of the Study:

• To explore the synergistic potential of combining DELs and ML for predictive toxicology.
• To model compound binding to off-targets, thereby improving safety assessments.
• To enable earlier identification of potential safety liabilities in the drug discovery pipeline.

Main Methods:

• Utilizing data from DEL screening campaigns.
• Developing and applying machine learning models to predict off-target binding.
• Leveraging ML's ability to generalize predictions across large chemical spaces.

Main Results:

• Demonstrated the feasibility of using DEL-generated data for ML-driven toxicology prediction.
• Showcased the potential for accurate prediction of off-target interactions.
• Highlighted the scalability and reusability of ML models across diverse drug discovery projects.

Conclusions:

• The integration of DELs and ML offers a cost-effective strategy for early-stage predictive toxicology.
• This combined approach can significantly improve the safety profiling of drug candidates.
• Wider application of generalized toxicology models can accelerate drug discovery by mitigating safety risks early on.