DeepPROTACs is a deep learning-based targeted degradation predictor for PROTACs.

Area of Science:

• Computational chemistry
• Drug discovery
• Machine learning in pharmacology

Background:

• Rational design of Proteolysis-targeting chimeras (PROTACs) is hindered by complex structure-activity relationships.
• Developing predictive models for PROTAC efficacy is crucial for accelerating drug discovery.

Purpose of the Study:

• To introduce DeepPROTACs, a deep neural network model designed to predict the degradation capacity of PROTAC molecules.
• To facilitate the rational design of potent PROTACs by predicting their efficacy based on target protein and E3 ligase structures.

Main Methods:

• Utilized Graph Convolutional Networks (GCNs) for feature extraction from ligand and binding pocket structures, represented as graphs.
• Employed a Bidirectional Long Short-Term Memory (BiLSTM) layer to process SMILES representations of PROTAC linkers.
• Trained the DeepPROTACs model on an experimental dataset curated from PROTAC-DB, labeled with DC50 and Dmax values.

Main Results:

• The DeepPROTACs model achieved an average prediction accuracy of 77.95% on the test set.
• The model demonstrated a high performance with an area under the receiver operating characteristic curve (AUC) of 0.8470.
• The study successfully developed a computational tool for predicting PROTAC degradation efficiency.

Conclusions:

• DeepPROTACs offers a powerful computational approach to predict PROTAC degradation capacity, addressing the challenges in rational PROTAC design.
• The developed model can significantly aid researchers in identifying and designing more potent PROTAC molecules.
• DeepPROTACs is accessible as a web server and open-source code, promoting its use in the scientific community.