DrugPred: a structure-based approach to predict protein druggability developed using an extensive nonredundant data...

Area of Science:

• Biochemistry
• Structural Biology
• Computational Chemistry

Background:

• Assessing protein druggability is vital for drug discovery and target validation.
• Existing methods for predicting druggability often lack accuracy or robustness.

Purpose of the Study:

• To develop a structure-based computational method for predicting protein druggability.
• To create a reliable predictor that outperforms previous approaches.

Main Methods:

• Compiled a comprehensive dataset of protein crystal structures, categorizing them by druggability.
• Trained a machine learning model, DrugPred, using partial least-squares projection to latent structures discriminant analysis (PLS-DA).
• Validated the predictor using both internal and external datasets, assessing its robustness against conformational changes.

Main Results:

• DrugPred demonstrated high accuracy in distinguishing between druggable and less druggable protein binding sites.
• The method showed robustness against variations in protein binding site conformations.
• DrugPred outperformed previously published protein druggability prediction methods.

Conclusions:

• The developed structure-based method, DrugPred, offers a significant advancement in predicting protein druggability.
• The predictor's superior performance is attributed to its training dataset, which focuses on evolutionarily optimized binding cavities.
• DrugPred provides a valuable tool for early-stage drug discovery and target assessment.