Proteolysis-targeting Chimera efficacy prediction using a deep-learning-QSP model.

Area of Science:

• Computational chemistry and pharmacology
• Drug discovery and development
• Biotechnology and molecular biology

Background:

• Proteolysis Targeting Chimeras (PROTACs) represent a novel therapeutic strategy for targeted protein degradation (TPD).
• PROTACs offer advantages over traditional inhibitors for targeting previously undruggable proteins.
• Predicting PROTAC efficacy is challenging due to molecular variability, necessitating advanced computational methods.

Purpose of the Study:

• To develop an integrated computational framework for predicting PROTAC molecule efficacy.
• To combine deep learning and Quantitative Systems Pharmacology (QSP) for enhanced prediction accuracy.
• To estimate key pharmacodynamic parameters, including half-maximal degradation concentration (DC50) and maximal degradation (Dmax).

Main Methods:

• Integration of a deep learning model (DeepCalici) for binding affinity prediction with a QSP Hook model.
• Utilizing curated experimental data from PROTAC-DB for model training and validation.
• Employing a supplementary deep neural network to adjust QSP model parameters based on molecular features.

Main Results:

• The integrated model demonstrated strong predictive performance for DC50, facilitating the prioritization of PROTAC candidates.
• Predictions for Dmax showed lower accuracy, attributed to experimental variability not captured in the dataset.
• The study highlights the need for comprehensive structural data and standardized experimental conditions for improved modeling.

Conclusions:

• The developed computational framework shows significant promise for accelerating PROTAC drug discovery.
• Accurate prediction of PROTAC efficacy is crucial for efficient therapeutic development.
• Future work should focus on incorporating standardized experimental data to refine predictive models for Dmax.