ELECTRA-DTA: a new compound-protein binding affinity prediction model based on the contextualized sequence encoding.

Area of Science:

• Computational chemistry
• Bioinformatics
• Artificial intelligence in drug discovery

Background:

• Drug-target binding affinity (DTA) prediction is crucial for efficient drug discovery.
• Deep learning methods show promise but often require large labeled datasets, which are not always available.
• Existing methods struggle with sparse interaction data.

Purpose of the Study:

• To develop an end-to-end deep learning framework, ELECTRA-DTA, for predicting drug-target binding affinity.
• To address the challenge of limited labeled data in representation learning for drug-target interactions.
• To improve the accuracy and efficiency of drug discovery pipelines.

Main Methods:

• Employed an unsupervised learning mechanism to train ELECTRA-based contextual embedding models for proteins and compound SMILES.
• Utilized a squeeze-and-excitation (SE) convolutional neural network block with fully connected layers to capture sequential and spatial features.
• Integrated protein amino acid and compound SMILES encoding for DTA regression.

Main Results:

• ELECTRA-DTA significantly outperforms state-of-the-art DTA prediction models, particularly on the sparse BindingDB dataset.
• Demonstrated competitive performance in drug repurposing and target selection for COVID-19.
• The framework shows potential for accelerating drug discovery and generalizability in computational biology.

Conclusions:

• ELECTRA-DTA offers a robust solution for DTA prediction, overcoming limitations of data scarcity.
• The model's architecture effectively captures complex features from protein and compound representations.
• ELECTRA-DTA holds promise for advancing computational drug discovery and related bioinformatics tasks.