LOCOM: A logistic regression model for testing differential abundance in compositional microbiome data with false...

Area of Science:

• Microbiome research
• Bioinformatics
• Statistical modeling

Background:

• Compositional analysis of microbiome data is crucial for understanding microbial communities.
• Existing methods struggle with zero counts and experimental biases, leading to unreliable false discovery rate (FDR) control.
• Log-transformation of microbiome data with zero counts is problematic and can affect analytical outcomes.

Purpose of the Study:

• To develop a robust compositional analysis method that is invariant to experimental biases.
• To address limitations of existing methods in controlling FDR and handling zero counts in high-throughput microbiome data.
• To introduce a novel logistic regression-based approach for microbiome compositional analysis.

Main Methods:

• Proposed Logistic Compositional Analysis (LOCOM), a robust logistic regression approach.
• Utilized odds ratios invariant to experimental biases, avoiding problematic log-transformation and pseudocounts.
• Employed permutation for inference to manage overdispersion and small sample sizes, supporting various trait types and confounder adjustment.

Main Results:

• LOCOM consistently preserved FDR and demonstrated significantly improved sensitivity compared to existing methods in simulations.
• ANCOM and ANCOM-BC/ALDEx2 showed inflated FDR with small and large effect sizes, respectively.
• LOCOM proved robust to experimental biases across all tested scenarios, outperforming other methods.

Conclusions:

• LOCOM is a reliable and robust method for compositional analysis of microbiome data, unaffected by experimental biases.
• The method offers superior FDR control and sensitivity, making it suitable for diverse microbiome studies.
• An R package for LOCOM is publicly available, facilitating its application in the research community.