Predicting Drug-Target Affinity Based on Recurrent Neural Networks and Graph Convolutional Neural Networks

Area of Science:

• Computational chemistry
• Drug discovery
• Bioinformatics

Background:

• Drug development is costly and time-consuming, necessitating cost-reduction strategies.
• Identifying drug-target interactions (DTIs) is vital for early-stage drug discovery.
• Predicting drug-target affinity (DTA) is more informative than binary classification for assessing DTI utility.

Purpose of the Study:

• To develop an improved computational method for predicting drug-target affinity (DTA).
• To enhance the GraphDTA model for more accurate prediction of drug-target binding strength.

Main Methods:

• Modified the GraphDTA model into a triple-channel network.
• Interpreted protein sequences as time series using LSTM for feature extraction.
• Utilized BiGRU and convolutional networks to extract topological and chemical features from drug structures.
• Integrated target and drug features into a feed-forward network for DTA prediction.

Main Results:

• The enhanced triple-channel model demonstrated superior performance compared to the original GraphDTA model.
• The model achieved better predictive accuracy on the Davis and Kiba datasets.
• Performance improvements were particularly notable in benchmark database evaluations.

Conclusions:

• The altered GraphDTA model effectively predicts drug-target affinity.
• The triple-channel approach offers enhanced accuracy for DTA prediction.
• This improved method holds significant potential for optimizing drug discovery pipelines.