Logistic regression augmented community detection for network data with application in identifying autism-related...

Area of Science:

• Bioinformatics
• Computational Biology
• Genetics

Background:

• Differentiating disease-relevant genes from background noise is crucial in genetic pathway analysis.
• Auxiliary gene information can aid in this differentiation when direct disease associations are unclear.

Purpose of the Study:

• To develop a robust computational method for identifying disease-associated genes.
• To cluster relevant genes into biologically meaningful groups.
• To improve the discovery of gene sets underlying complex neurological diseases.

Main Methods:

• A logistic regression model is employed to filter out irrelevant background genes using auxiliary data.
• An expectation-maximization algorithm is modified to cluster relevant genes and model within/between-group interactions.
• A robust version of the method is derived to handle arbitrary linkage patterns in background genes.
• The stochastic blockmodel is used to prove asymptotic consistency of gene label assignments.

Main Results:

• Simulation studies demonstrate superior performance and robustness of the proposed method in finite samples.
• The robust method successfully identified previously unrecognized gene sets associated with autism spectrum disorder.
• The approach integrates diverse data sources, including de novo mutations, gene expression, and protein-protein interactions.

Conclusions:

• The developed method effectively distinguishes disease-relevant genes from background noise.
• This approach enhances the identification of gene sets implicated in neurological diseases like autism spectrum disorder.
• The integration of multiple data types provides a more comprehensive understanding of disease genetics.