FuseNet

FuseNet infers networks from collections of nonidentically distributed datasets using a Markov network formulation to model dependencies in high-throughput omics data, including next-generation sequencing outputs.

Key Features:

• Markov Network Formulation: Utilizes an undirected graphical model (Markov network) to infer relationships between genes and other biological entities from experimental data.
• Handling Non-Gaussian Data: Models non-Gaussian and multiple related but distinct distributions commonly encountered in high-throughput technologies such as next-generation sequencing by leveraging shared latent factors.
• Computational Efficiency and Generality: Operates on distributions from the exponential family and is implemented for computational efficiency to handle any number of such distributions.
• Shared Latent Factors: Represents model parameters through shared latent factors that define neighborhoods of network nodes and enable integration and fusion of multiple datasets.

Scientific Applications:

• Joint Network Inference: Fuses multiple datasets to infer joint networks, providing more accurate and comprehensive models than inferring separate networks for each dataset.
• Predictive Performance: Demonstrated superior predictive performance in simulation studies relative to several popular graphical models.
• Breast Cancer RNA-sequencing and Somatic Mutation Analysis: Applied to breast cancer RNA-sequencing and somatic mutation data to investigate complex genetic interactions.

Methodology:

Formulates network inference as a Markov (undirected graphical) model and jointly models collections of datasets with varying distributions from the exponential family, representing parameters via shared latent factors that define node neighborhoods.