Computational prediction of virus-human protein-protein interactions using embedding kernelized heterogeneous data.

Area of Science:

• Computational biology
• Infectious disease research
• Bioinformatics

Background:

• Pathogenic microorganisms interact with host cells, influencing disease progression and therapeutic strategies.
• Understanding pathogen-host interactions (PHIs) is vital for developing effective treatments against infectious diseases.
• Computational prediction of PHIs is increasingly important due to limited experimental data.

Purpose of the Study:

• To develop a novel computational method for predicting pathogen-host interactions (PHIs).
• To overcome the limitation of requiring non-interacting protein pairs in traditional prediction models.
• To identify novel PHIs for potential therapeutic targets.

Main Methods:

• Formulated the PHI prediction problem using kernel embedding of heterogeneous data.
• Eliminated the need for datasets of non-interacting protein pairs.
• Utilized domain-domain associations to filter predicted interactions.

Main Results:

• Identified 175 novel PHIs involving 170 human proteins and 105 viral proteins.
• Achieved over 10% improvement in accuracy and AUC compared to state-of-the-art methods.
• Demonstrated superior performance when compared using a binary classification formulation.

Conclusions:

• The proposed kernel embedding method effectively predicts PHIs without requiring negative interaction data.
• The identified novel PHIs represent potential targets for developing new anti-infective therapies.
• This approach offers a significant advancement in computational prediction of pathogen-host interactions.