Drug-target Affinity Prediction by Molecule Secondary Structure Representation Network.

Area of Science:

• Computational chemistry
• Bioinformatics
• Drug discovery

Background:

• Drug-target interaction (DTI) identification is vital for drug development, aiming for high specificity and low toxicity.
• Computer-aided prediction accelerates screening and aids in drug repositioning for new indications.
• Efficient and accurate DTI prediction algorithms are crucial for advancing pharmaceutical research.

Purpose of the Study:

• To develop a novel method for more accurate drug characterization in DTI prediction.
• To address limitations of existing methods that lose or fail to capture drug structural information.
• To enhance the efficiency and accuracy of drug-target affinity prediction models.

Main Methods:

• Proposed a novel Molecule Secondary Structure Representation Network (MSSRN) for accurate drug characterization.
• Employed relational graph convolutional networks (R-GCNs) on molecular graphs and integrated drug sequence convolutions.
• Utilized an attention mechanism to calculate spatial importance weights, guiding R-GCNs for topological information learning.

Main Results:

• Constructed the MSSRN-DTA model using MSSRN for drug structure and CNN for protein sequence.
• Demonstrated the effectiveness of the proposed MSSRN-DTA model.

Conclusions:

• The novel MSSRN method accurately captures drug structural and sequential information, enhancing DTI prediction.
• The MSSRN-DTA model shows superior performance compared to alternative methods and baseline models.
• The study validates the effectiveness of the proposed approach on benchmark datasets.