LDM

LDM tests presence-absence associations in microbiome studies while accounting for confounding by library size and unifying community-level and individual-taxon analyses.

Key Features:

• Presence-absence association testing: Tests associations based on taxa presence or absence at both community and individual-taxon levels.
• Library-size confounding control: Addresses confounding introduced by total sample read counts (library size) that may correlate with covariates.
• Non-stochastic rarefaction extension: Applies LDM across multiple rarefied taxa count tables and averages residual sum-of-squares (RSS) terms instead of averaging F-statistics.
• F-statistic construction from averaged RSS: Constructs an F-statistic using averaged RSS values from rarefaction replicates for hypothesis testing.
• Adjustment for covariates: Allows adjustment for confounding covariates in analyses of discrete and continuous traits and their interactions.
• Support for discrete and continuous traits and interactions: Tests associations for both discrete and continuous variables and their interactions.
• Robustness and power: Demonstrates robustness to systematic differences in library size and improved power compared to alternative methods in simulation studies.

Scientific Applications:

• Microbiome presence-absence studies: Detects taxa associated with phenotypes based on presence or absence in microbiome data.
• Confounding assessment due to library size: Evaluates and mitigates bias arising from differences in total read counts across samples.
• Testing traits and interactions: Tests discrete and continuous traits and their interactions while adjusting for covariates.
• Inflammatory bowel disease (IBD) analysis: Applied to IBD data to analyze presence-absence associations when cases have systematically smaller library sizes than controls.

Methodology:

Apply rarefaction (subsampling to a common library size) to generate multiple rarefied taxa count tables, fit the LDM to each replicate to obtain residual sum-of-squares (RSS) terms, average the RSS across replicates, and construct an F-statistic from the averaged RSS for hypothesis testing.