DoubleSG-DTA: Deep Learning for Drug Discovery: Case Study on the Non-Small Cell Lung Cancer with EGFR Mutation

Area of Science:

• Computational chemistry
• Bioinformatics
• Machine learning in drug discovery

Background:

• Drug-targeted therapies are crucial for cancer treatment.
• Accurate prediction of drug-target affinities (DTAs) is essential for efficient drug development.
• Existing methods for DTA prediction require improvement in accuracy and interpretability.

Purpose of the Study:

• To develop a novel deep learning model, DoubleSG-DTA, for predicting drug-target affinities.
• To enhance the accuracy and interpretability of computational drug discovery models.
• To identify potential drug candidates from natural products for specific cancer mutations.

Main Methods:

• Developed DoubleSG-DTA, a mechanism-driven deep learning model utilizing drug sequences, protein sequences, and drug graphs.
• Employed graph isomorphism networks for molecular structure analysis and squeeze-and-excitation networks for feature enhancement.
• Integrated cross-multi-head attentions to model non-covalent molecular docking interactions.

Main Results:

• DoubleSG-DTA demonstrated superior performance over existing methods across multiple datasets.
• The model successfully identified promising hit compounds for Non-Small Cell Lung Cancer with EGFRT790M mutation from natural products.
• The interpretability analysis highlighted key molecular structures contributing to predicted drug-target interactions.

Conclusions:

• DoubleSG-DTA provides a powerful and interpretable framework for predicting drug-target affinities.
• The model facilitates the discovery of novel therapeutic compounds by extrapolating potential drug candidates.
• This approach can significantly accelerate the development of targeted therapies for various diseases.