A deep learning method for drug-target affinity prediction based on sequence interaction information mining.

Area of Science:

• Computational Biology
• Bioinformatics
• Drug Discovery

Background:

• Accurate drug-target affinity (DTA) prediction is crucial for efficient in silico drug discovery.
• Experimental DTA determination is costly and time-consuming, necessitating advanced computational methods.
• Deep learning offers a powerful approach for DTA prediction due to its ability to process complex data.

Purpose of the Study:

• To introduce KC-DTA, a novel sequence-based deep learning framework for predicting drug-target affinity (DTA).
• To leverage sequence and graph representations for enhanced DTA prediction accuracy.
• To provide an accessible and effective tool for accelerating in silico drug discovery.

Main Methods:

• Target sequences are transformed into two matrices using k-mer analysis and Cartesian products, capturing residue interactions and evolutionary information.
• Molecular compounds are represented as graphs, with atoms as nodes and bonds as edges.
• Convolutional Neural Networks (CNNs) and Graph Neural Networks (GNNs) process these representations to extract features for DTA prediction.

Main Results:

• The KC-DTA method demonstrated high performance in predicting drug-target affinity.
• Comprehensive comparisons confirmed KC-DTA's effectiveness against state-of-the-art approaches.
• Experimental results validate KC-DTA as a significant advancement in DTA prediction.

Conclusions:

• KC-DTA is a valuable tool for in silico drug discovery, offering a computationally efficient alternative to experimental methods.
• The method shows promise for accelerating the drug development pipeline.
• The study provides open access to data and code, facilitating further research and application.