Building Block-Based Binding Predictions for DNA-Encoded Libraries.

Area of Science:

• Medicinal Chemistry
• Computational Chemistry
• Drug Discovery

Background:

• DNA-encoded libraries (DELs) enable rapid screening of millions of compounds.
• DEL screening is cost-effective but prone to noise, requiring computational analysis.
• Improving signal-to-noise ratios is crucial for effective DEL data interpretation.

Purpose of the Study:

• To develop informatics tools for analyzing DEL screening data.
• To identify productive building blocks for designing new DEL libraries.
• To build a predictive model for compound binding based on building block behavior.

Main Methods:

• Utilized data from prior DEL screens to infer building block productivity.
• Developed a model based on the principle that combined building block behavior predicts compound binding.
• Applied the approach to three-cycle OpenDEL libraries screened against soluble epoxide hydrolase (sEH).

Main Results:

• Demonstrated that similar building blocks exhibit similar probabilities of forming binding compounds.
• The developed model achieved performance over an order of magnitude better than random guessing on a holdout set.
• Established a baseline for comparing machine learning models on DEL data.

Conclusions:

• The informatics workflow can enhance the design of new DEL libraries.
• The model provides a valuable tool for optimizing DEL campaigns and improving drug discovery efficiency.
• This approach offers a significant advancement in leveraging DEL data for identifying potent drug candidates.