GramSeq-DTA: A Grammar-Based Drug-Target Affinity Prediction Approach Fusing Gene Expression Information.

Area of Science:

• Computational chemistry
• Bioinformatics
• Drug discovery

Background:

• Drug-target affinity (DTA) prediction is vital for drug discovery.
• Current 1D string and graph-based representations have limitations in capturing comprehensive drug and target information.
• Integrating functional genetic information can improve DTA prediction accuracy.

Purpose of the Study:

• To propose GramSeq-DTA, a novel method that integrates chemical perturbation data with structural information for enhanced DTA prediction.
• To address the limitations of existing DTA prediction models by incorporating functional genetic insights.

Main Methods:

• Utilized a Grammar Variational Autoencoder (GVAE) for drug feature extraction.
• Employed Convolutional Neural Network (CNN) and Recurrent Neural Network (RNN) for protein feature extraction.
• Integrated drug structural features, gene expression data from the L1000 project (chemical perturbation), and target features.

Main Results:

• The proposed GramSeq-DTA model demonstrated superior performance compared to state-of-the-art DTA prediction methods.
• Validation on BindingDB, Davis, and KIBA datasets confirmed the model's effectiveness.
• The integration of structural and functional (chemical perturbation) data significantly improved prediction accuracy.

Conclusions:

• GramSeq-DTA offers a novel and practical approach to DTA prediction by combining structural and functional biological data.
• This multi-modal approach advances DTA prediction accuracy and encourages further research in integrated feature representation for drug discovery.