Druggability Assessment in TRAPP Using Machine Learning Approaches.

Area of Science:

• Computational biology
• Drug discovery
• Structural bioinformatics

Background:

• Protein druggability prediction is vital for identifying effective drug targets early in the drug discovery pipeline.
• Protein flexibility and conformational changes can significantly alter the druggability of binding pockets, posing a challenge for traditional prediction methods.

Purpose of the Study:

• To develop and validate statistical models for predicting protein druggability, accounting for variations in binding pocket properties.
• To integrate these models into a computational tool (TRAnsient Pockets in Proteins - TRAPP) for analyzing dynamic changes in protein binding pockets.

Main Methods:

• Development of two statistical models: a logistic regression model (TRAPP-LR) and a convolutional neural network model (TRAPP-CNN).
• Integration of TRAPP-LR and TRAPP-CNN into the TRAnsient Pockets in Proteins (TRAPP) software.
• Training and testing models on diverse datasets, including protein crystal structures and molecular dynamics simulation trajectories.

Main Results:

• TRAPP-LR and TRAPP-CNN models demonstrate performance equivalent or superior to existing methods.
• The models effectively predict druggability variations based on spatial and physicochemical properties of binding pockets.
• TRAPP successfully identifies potentially druggable protein conformations within molecular dynamics simulations.

Conclusions:

• The TRAPP tool, with its integrated LR and CNN models, provides a sensitive and accurate method for assessing protein druggability in dynamic contexts.
• TRAPP facilitates the identification of key factors influencing binding pocket druggability, supporting informed drug target selection.
• This approach enhances the early stages of drug discovery by accounting for protein conformational flexibility.