Gene network inference by fusing data from diverse distributions.

Area of Science:

• Computational Biology
• Statistical Genetics
• Bioinformatics

Background:

• Markov networks are standard for gene relation inference but assume Gaussian data.
• High-throughput omics data often violates the Gaussian assumption.
• Jointly analyzing multiple, non-identical datasets reveals common network features.

Purpose of the Study:

• Develop a novel Markov network formulation for inferring gene networks from diverse data distributions.
• Address limitations of current methods with non-Gaussian and multi-dataset omics data.
• Improve accuracy and efficiency in biological network inference.

Main Methods:

• Introduced FuseNet, a computationally efficient and general Markov network formulation.
• Utilized shared latent factors to model parameters across non-identically distributed datasets.
• Applied to RNA-sequencing and somatic mutation data for breast cancer analysis.

Main Results:

• FuseNet demonstrated superior predictive performance against popular graphical models in simulations.
• Fusion of multiple datasets yielded substantial gains over single-dataset network inference.
• Validated effectiveness in a novel application to breast cancer omics data.

Conclusions:

• FuseNet provides a principled statistical approach for non-Gaussian and multi-dataset network inference.
• The method accurately models high-throughput omics data, enhancing biological discovery.
• Fusion of datasets is crucial for robust gene network inference.