BIONIC

BIONIC integrates biological networks using graph convolutional networks (GCNs) to produce unified functional representations of genes and proteins by automatically weighting diverse input network information.

Key Features:

• Graph Convolutional Networks (GCNs): BIONIC employs graph convolutional networks to learn complex, integrated features from multiple biological networks.
• Automatic Network Weighting: It automatically weights diverse input networks to prioritize informative connections during integration.
• Unsupervised and Semisupervised Modes: BIONIC supports unsupervised and semisupervised learning and can incorporate gene function annotations in semisupervised mode.
• Scalability: BIONIC scales to integrate numerous and large networks simultaneously, including applications at the human-genome scale.

Scientific Applications:

• Network integration for functional mapping: Integrating biological networks from varied data sources to map cellular functions and improve functional representation of genes and proteins.
• Prediction of chemical-genetic interactions in yeast: Predicting and enabling experimental validation of essential gene chemical-genetic interactions from nonessential gene profiles in yeast.

Methodology:

BIONIC integrates input networks by automatically weighting them and trains graph convolutional networks to learn integrated features for genes or proteins, operating in unsupervised and semisupervised modes that can use available gene function annotations.