AI-DPAPT: a machine learning framework for predicting PROTAC activity

Area of Science:

• Computational chemistry
• Drug discovery
• Artificial intelligence in pharmacology

Background:

• Targeted protein degradation (TPD) using Proteolysis Targeting Chimeras (PROTACs) is crucial for drug development.
• Predicting small-molecule interactions with target proteins is complex and challenging.

Purpose of the Study:

• To develop an AI-powered tool for predicting PROTAC activity based on chemical structures.
• To enhance the accuracy and efficiency of identifying and optimizing PROTAC drug candidates.

Main Methods:

• Utilized machine learning algorithms and molecular fingerprinting techniques.
• Trained models on a dataset of PROTAC chemical structures and their activities.
• Evaluated classifier performance including Random Forest, achieving an AUC of 0.97.

Main Results:

• The AI-DPAPT tool accurately predicts PROTAC activities, with all models showing an ROC curve above 0.9.
• The Random Forest model demonstrated high predictive accuracy with an AUC score of 0.97.
• Identified key molecular features influencing PROTAC function and structure-activity relationships.

Conclusions:

• The developed AI platform significantly contributes to computational drug development by predicting PROTAC function.
• This tool accelerates the identification and improvement of novel therapeutic agents.
• The AI-DPAPT platform is accessible online for broader research application.