IILLS: predicting virus-receptor interactions based on similarity and semi-supervised learning

Area of Science:

• Computational biology
• Virology
• Bioinformatics

Background:

• Viral infectious diseases pose a significant threat to human health.
• Understanding virus-receptor interactions is crucial for treating viral infections.
• Current computational methods for predicting these interactions are limited.

Purpose of the Study:

• To propose a novel computational method, IILLS, for predicting virus-receptor interactions.
• To integrate known interactions and amino acid sequences for improved prediction accuracy.
• To evaluate the performance of IILLS against existing methods.

Main Methods:

• Developed the IILLS method integrating initial interaction scores, neighbor information, and Laplacian regularized Least Squares.
• Utilized Gaussian Interaction Profile (GIP) kernel for virus similarity calculation.
• Employed receptor sequence similarity as the final receptor similarity metric.

Main Results:

• IILLS achieved an Area Under the Curve (AUC) of 0.8675 using 10-fold cross-validation (10CV).
• IILLS achieved an AUC of 0.9061 using leave-one-out cross-validation (LOOCV).
• Performance evaluation demonstrated IILLS superiority over competing methods (BRWH, LapRLS, CMF).

Conclusions:

• The IILLS method demonstrates high accuracy and effectiveness in predicting virus-receptor interactions.
• IILLS offers a significant advancement over existing computational approaches.
• Case studies further validate the practical utility of IILLS in predicting virus-receptor interactions.