Recursive Indirect-Paths Modularity (RIP-M) for Detecting Community Structure in RNA-Seq Co-expression Networks.

Area of Science:

• Bioinformatics
• Computational Biology
• Systems Biology

Background:

• Gene co-expression network clusters are vital for functional analysis and machine learning.
• Existing methods like Newman Modularity often yield suboptimal cluster sizes, hindering biological interpretation.
• Identifying optimal gene clusters is crucial for accurate pathway enrichment and sample classification.

Purpose of the Study:

• To develop a generalized Newman Modularity algorithm that incorporates indirect path information from RNA-Seq data.
• To create a clustering method that allows users to constrain gene cluster sizes for improved biological relevance.
• To evaluate the performance of the new Recursive Indirect-Pathways Modularity (RIP-M) algorithm against existing methods.

Main Methods:

• Generalization of Newman Modularity using indirect path information in RNA-Seq co-expression networks.
• Implementation of a merge-and-split algorithm to control gene cluster size ranges.
• Comparative analysis using simulated co-expression networks and real RNA-Seq data from an influenza vaccine study.

Main Results:

• RIP-M demonstrated higher cluster assignment accuracy than Newman Modularity on simulated networks.
• RIP-M showed comparable accuracy to Weighted Gene Correlation Network Analysis (WGCNA), outperforming it under certain thresholds.
• Both RIP-M and WGCNA identified a similar number of immunologically relevant pathways in the vaccine response data.

Conclusions:

• RIP-M offers an effective approach for gene clustering in RNA-Seq data, balancing interpretability and inference.
• The algorithm provides a flexible alternative to existing methods, particularly when optimal cluster size is critical.
• RIP-M shows promise for advancing functional genomics and biomarker discovery in complex biological systems.