MDeePred: novel multi-channel protein featurization for deep learning-based binding affinity prediction in drug...

Area of Science:

• Computational chemistry
• Bioinformatics
• Drug discovery

Background:

• Identifying bioactive small molecule-target protein interactions is vital for drug discovery, repurposing, and understanding off-target effects.
• Computational approaches are essential for screening vast chemical spaces due to experimental limitations.
• Developing effective protein featurization methods for deep learning remains a challenge.

Purpose of the Study:

• To introduce a novel protein featurization approach for deep learning-based compound-target protein binding affinity prediction.
• To present MDeePred, a new computational method for drug discovery and repositioning.
• To improve the accuracy and scalability of predicting molecular interactions.

Main Methods:

• Incorporated multiple protein features (sequence, structural, evolutionary, physicochemical) into 2D vectors.
• Utilized state-of-the-art pairwise input hybrid deep neural networks.
• Adopted a proteochemometric approach using both compound and target protein features.

Main Results:

• MDeePred demonstrated sufficiently high predictive performance on benchmark datasets.
• The method achieved competitive results compared to existing state-of-the-art approaches.
• In vitro analysis showed alignment between MDeePred predictions and kinase inhibitors' effects on cancer cells.

Conclusions:

• MDeePred offers a scalable and effective solution for computational drug discovery and repositioning.
• The proposed protein featurization approach can be applied to other protein-related predictive tasks.
• This work addresses the challenge of comprehensive protein featurization for deep learning in drug discovery.