Enhancing the Predictive Power of Machine Learning Models through a Chemical Space Complementary DEL Screening...

Area of Science:

• Medicinal Chemistry
• Drug Discovery
• Computational Chemistry

Background:

• DNA-encoded library (DEL) technology facilitates high-throughput screening for drug discovery.
• Traditional DEL screening can yield lead compounds with high molecular weights, complicating drug development.
• Complex patterns in DEL data are challenging for human analysis.

Purpose of the Study:

• To refine DEL technology by incorporating alternative screening methods and AI for improved small molecule identification.
• To enhance chemical diversity and predictive performance in drug discovery models.
• To identify novel small molecules for therapeutic targets such as BRD4 and p300.

Main Methods:

• Integration of photocross-linking screening with traditional DEL techniques.
• Application of AI and machine learning models to analyze DEL screening data.
• Prediction and validation of active small molecules against selected protein targets.

Main Results:

• Improved predictive performance for small molecule identification models.
• Successful prediction of active small molecules for BRD4 and p300 with high hit rates (26.7% and 35.7%, respectively).
• Identified compounds possess lower molecular weights and enhanced modification potential compared to traditional DEL-derived leads.

Conclusions:

• The synergistic combination of DEL and AI technologies significantly enhances small molecule drug discovery.
• This refined approach yields drug leads with favorable physicochemical properties for further development.
• The study presents a powerful strategy for identifying potent and developable drug candidates.