Polypharmacology Browser PPB3: A Web-Based Deep Learning Tool for Target Prediction Using ChEMBL Data.

Area of Science:

• Computational chemistry
• Pharmacology
• Machine learning

Background:

• Drug-like molecules frequently interact with multiple biological targets, a phenomenon known as polypharmacology.
• Assessing polypharmacology is crucial for effective drug development and understanding drug mechanisms.

Purpose of the Study:

• To develop and validate deep neural network models for predicting drug-target interactions.
• To assess the polypharmacology of bioactive molecules using a large-scale dataset.

Main Methods:

• Trained deep neural networks using binary substructure fingerprints of molecules from ChEMBL 34.
• Associated molecules with targets based on activity data (≥50% active at ≤10 μM).
• Utilized a dataset comprising over 2.4 million interactions between 1.1 million molecules and 7546 targets.

Main Results:

• The models demonstrated good performance in terms of recall and precision for both molecules and targets.
• The dataset size and scope (including various target types beyond proteins) significantly exceed previous models.
• A case study highlighted the models' predictive capabilities compared to other online tools.

Conclusions:

• Deep neural networks can effectively predict drug-target interactions and polypharmacology.
• The developed models offer a valuable tool for drug discovery and development.
• Online prediction services (PPB3) are available for public use.