A comparative analysis of mutual information methods for pairwise relationship detection in metagenomic data

Area of Science:

• Microbiology
• Bioinformatics
• Systems Biology

Background:

• Metagenomic co-occurrence networks commonly use correlation to infer microbial relationships.
• Correlation-based methods may miss complex, non-linear interactions prevalent in biological systems.

Purpose of the Study:

• To explore mutual information (MI) estimation for quantifying pairwise relationships in biological data.
• To compare MI performance against Pearson's and Spearman's correlation coefficients.

Main Methods:

• Applied various mutual information estimators.
• Tested metrics on simulated ecological data and real C. diff infection data.
• Compared MI with Pearson's (r) and Spearman's (ρ) correlation coefficients.

Main Results:

• Mutual information estimators showed superior detection ability for asymmetric relationships compared to correlation coefficients.
• MI estimators demonstrated enhanced performance in identifying exploitative relationships.
• MI can uncover complex pairwise relationships missed by traditional association measures.

Conclusions:

• Mutual information is valuable for uncovering complex pairwise relationships in biological data.
• Incorporating MI into co-occurrence network construction provides a more comprehensive analysis than correlation alone.
• MI offers a promising approach for future metagenomic studies.