DeepViral

DeepViral predicts novel virus-host protein-protein interactions (PPIs) by integrating protein sequences with infectious disease phenotypes to improve understanding of host-pathogen mechanisms.

Key Features:

• Deep Learning Framework: Uses deep learning to model and predict PPIs between humans and viruses from complex input data.
• Integration of Phenotypic Data: Incorporates disease phenotypes (observable signs and symptoms) as an additional information source alongside protein sequences.
• Shared Embedding Space: Embeds human proteins and viruses in a shared space using associated phenotypes and functions, supported by formalized background knowledge from biomedical ontologies.
• Joint Learning Approach: Jointly learns from protein sequences and phenotype features to improve prediction accuracy for intra- and inter-species PPIs and outperforms sequence-only methods.
• Novel Evaluation Setup: Employs an experimental evaluation setup designed to realistically assess prediction performance for novel viruses.

Scientific Applications:

• Infectious Disease Research: Supports the study of virus-host interactions to elucidate mechanisms underlying infectious diseases.
• Therapeutic Target Identification: Aids identification of potential therapeutic targets by predicting virus-host PPIs.
• Pathogenesis Analysis: Facilitates understanding of the pathogenesis of novel viruses by integrating molecular and phenotypic data.

Methodology:

Embeds proteins and viruses into a shared space using deep learning; jointly learns from protein sequences and disease phenotypes and functions; leverages biomedical ontologies as background knowledge; evaluates predictions with a novel experimental setup for novel viruses.