Determining confidence of predicted interactions between HIV-1 and human proteins using conformal method.

Area of Science:

• Molecular Biology
• Bioinformatics
• Computational Biology

Background:

• Protein-protein interactions (PPIs) are fundamental to cellular processes.
• Traditional PPI prediction methods often lack confidence estimation.
• Supervised learning for PPIs requires both positive and negative interaction datasets.

Purpose of the Study:

• To develop a conformal prediction framework for PPI identification.
• To introduce confidence estimation for PPI predictions.
• To improve the identification of HIV-1 and human protein interactions.

Main Methods:

• Applied a conformal prediction framework based on the 'exchangeability' assumption.
• Utilized a function measuring the relative 'strangeness' of interacting pairs.
• Developed a method to handle a single class (interacting proteins) without a defined non-interacting set.

Main Results:

• The conformal predictor provides confidence levels for PPI predictions.
• Users can control prediction error rates by setting a desired confidence level.
• Successfully applied the framework to identify interacting pairs between HIV-1 and human proteins.

Conclusions:

• Conformal prediction offers a robust approach for PPI identification with confidence estimation.
• The confidence level aids biologists in prioritizing pairs for experimental validation.
• This method enhances the understanding of molecular mechanisms, particularly in viral-host interactions.