DeNovo: virus-host sequence-based protein-protein interaction prediction.

Area of Science:

• Computational Biology and Bioinformatics
• Virology and Host-Pathogen Interactions
• Machine Learning in Biological Systems

Background:

• Predicting protein-protein interactions (PPIs) for novel viruses is challenging due to limited data, high costs, and sequence dissimilarity across viral families.
• Existing methods often fail or are inefficient when dealing with viruses lacking known PPIs or exhibiting significant sequence divergence.
• Understanding virus-host PPIs is crucial for deciphering viral mechanisms and developing targeted therapeutics.

Purpose of the Study:

• To develop a novel computational framework, DeNovo, for predicting protein-protein interactions (PPIs) of a novel virus with its host.
• To overcome limitations of existing methods by leveraging PPI data from diverse viruses and shared host proteins.
• To assess the generalization capability of the developed framework across different biological domains.

Main Methods:

• DeNovo employs a sequence-based negative sampling and machine learning approach.
• The framework learns from known PPIs of various viruses to predict interactions for a novel virus, exploiting conserved host proteins.
• The model's performance and generalization were rigorously tested on PPI datasets from different biological contexts.

Main Results:

• DeNovo achieved high prediction accuracy, reaching 81% for viruses with no sequence similarity and 86% for those with distant similarity to training data.
• These results are comparable to state-of-the-art methods in single virus-host and intra-species PPI prediction.
• The framework demonstrated strong generalization, achieving near-optimal accuracy on bacteria-human interactions.

Conclusions:

• DeNovo effectively predicts protein-protein interactions for novel viruses, even with limited or dissimilar training data.
• The developed framework significantly advances the capability to predict PPIs for virtually any virus infecting humans.
• DeNovo's robust generalization suggests its broad applicability in predicting host-pathogen interactions across diverse biological systems.