NG-DTA: Drug-target affinity prediction with n-gram molecular graphs

Area of Science:

• Computational chemistry
• Bioinformatics
• Drug discovery

Background:

• Drug-target affinity (DTA) prediction is vital for accelerating drug development.
• Current deep learning models often represent proteins as 1D strings, limiting their informative capacity compared to graph representations.
• Graph neural networks (GNNs) offer a more powerful approach for molecular data analysis.

Purpose of the Study:

• To develop an advanced deep learning method for accurate DTA prediction.
• To investigate the impact of protein molecular sub-graph representation on DTA prediction performance.
• To establish a novel computational tool for drug discovery pipelines.

Main Methods:

• Developed N-gram Graph DTA (NG-DTA), a deep learning model utilizing GNNs.
• Input features include molecular graphs of drugs and n-gram molecular sub-graphs of proteins.
• No protein structure prediction tools were employed.

Main Results:

• NG-DTA achieved superior performance on both Davis and Kiba datasets.
• Achieved a concordance index (CI) of 0.905 and mean square error (MSE) of 0.196 on the Davis dataset.
• Achieved a CI of 0.904 and MSE of 0.120 on the Kiba dataset.
• Demonstrated that n-gram protein sub-graphs enhance deep learning model performance for DTA prediction.

Conclusions:

• Representing proteins using n-gram molecular sub-graphs significantly improves deep learning-based DTA prediction.
• NG-DTA offers a promising and effective approach for computational drug discovery.
• The method provides accurate DTA predictions without relying on protein structure prediction.