GiGs: graph-based integrated Gaussian kernel similarity for virus-drug association prediction.

Area of Science:

• Bioinformatics
• Computational Biology
• Drug Discovery

Background:

• Predicting virus-drug associations (VDAs) is vital for identifying novel antiviral therapies and repurposing existing drugs.
• Accurate VDA prediction can accelerate the drug discovery pipeline and combat emerging infectious diseases.

Purpose of the Study:

• To develop and validate a novel computational method, GiGs, for predicting VDAs to facilitate drug repositioning.
• To enhance the accuracy and reliability of VDA prediction by integrating diverse biological data and advanced graph-based techniques.

Main Methods:

• The GiGs method integrates virus sequence, drug chemical structure, and drug side effect similarities.
• It utilizes Gaussian interaction profile kernel (GIPK) for similarity integration and similarity-constrained weight graph normalization matrix factorization for prediction.
• Graph regularization is employed to prevent overfitting and improve prediction accuracy.

Main Results:

• The GiGs model demonstrated superior performance compared to five other state-of-the-art association prediction methods.
• Extensive experiments validated the model's effectiveness in predicting potential VDAs.
• A case study successfully identified broad-spectrum drugs for human coronavirus infections, with molecular docking confirming predictions.

Conclusions:

• The GiGs method provides a robust and accurate approach for predicting VDAs, significantly aiding drug repositioning efforts.
• This approach can accelerate the identification of effective antiviral drugs, particularly for challenging pathogens.
• The study highlights the potential of integrated, graph-based methods in computational drug discovery.