DEEPScreen: high performance drug-target interaction prediction with convolutional neural networks using 2-D...

Area of Science:

• Computational chemistry and cheminformatics
• Drug discovery and development
• Bioinformatics and computational biology

Background:

• Identifying drug-candidate biomolecule interactions is crucial for drug discovery.
• Traditional screening methods are costly and time-consuming.
• Computational approaches offer efficient alternatives for predicting drug-target interactions (DTIs).

Purpose of the Study:

• To introduce DEEPScreen, a large-scale system for predicting novel DTIs in early-stage drug discovery.
• To leverage deep convolutional neural networks for enhanced DTI prediction accuracy.
• To utilize 2-D structural representations of compounds as input for improved performance.

Main Methods:

• Developed DEEPScreen using deep convolutional neural networks.
• Trained the system on 704 target proteins using curated bioactivity data.
• Employed 2-D structural representations of compounds, bypassing traditional descriptors.

Main Results:

• DEEPScreen demonstrated high prediction accuracy by learning features from 2-D compound structures.
• Performance was validated against state-of-the-art methods on benchmark datasets.
• JAK proteins were identified as novel targets for cladribine, with experimental validation via STAT3 phosphorylation in cancer cells.

Conclusions:

• DEEPScreen offers an effective computational tool for early-stage drug discovery and repurposing.
• The system facilitates in silico screening of the chemogenomic space to identify novel DTIs.
• Open-source code, models, and datasets are available for community use and further research.