Improving compound-protein interaction prediction by building up highly credible negative samples.

Area of Science:

• Computational biology
• Drug discovery and development
• Bioinformatics

Background:

• Computational prediction of compound-protein interactions (CPIs) is vital for efficient drug design.
• Lack of reliable negative CPI samples hinders the performance of in silico prediction models.
• Experimental validation of CPIs is costly and time-consuming, necessitating improved computational methods.

Purpose of the Study:

• To establish a set of highly credible negative CPI samples using an in silico screening approach.
• To enhance the accuracy and reliability of computational prediction models for CPIs.
• To provide a valuable resource for identifying novel drug targets.

Main Methods:

• Developed a systematic screening framework integrating diverse data sources (chemical structures, expression profiles, side effects, protein sequences, PPI networks, functional annotations).
• Utilized the converse negative proposition: proteins dissimilar to known targets are unlikely to interact with a compound, and vice versa.
• Tested screened negative samples on classical classifiers, existing prediction models, and a drug bioactivity dataset.

Main Results:

• Six classical classifiers showed significantly improved performance with the screened negative samples compared to random samples for human and Caenorhabditis elegans.
• Three existing prediction models (bipartite local model, Gaussian kernel profile, Bayesian matrix factorization) demonstrated enhanced performance.
• Derived new interaction sets by training a support vector machine classifier on positive and screened negative CPIs.

Conclusions:

• The proposed in silico screening method effectively generates reliable negative CPI samples.
• The generated negative samples significantly improve the performance of various computational CPI prediction models.
• This resource aids researchers in identifying new drug targets and supplements existing compound-protein databases.