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

Abstract

Compositional analysis is based on the premise that a relatively small proportion of taxa are differentially abundant, while the ratios of the relative abundances of the remaining taxa remain unchanged. Most existing methods use log-transformed data, but log-transformation of data with pervasive zero counts is problematic, and these methods cannot always control the false discovery rate (FDR). Further, high-throughput microbiome data such as 16S amplicon or metagenomic sequencing are subject to experimental biases that are introduced in every step of the experimental workflow. McLaren et al. [ eLife 8, e46923 (2019)] have recently proposed a model for how these biases affect relative abundance data. Motivated by this model, we show that the odds ratios in a logistic regression comparing counts in two taxa are invariant to experimental biases. With this motivation, we propose logistic compositional analysis (LOCOM), a robust logistic regression approach to compositional analysis, that does not require pseudocounts. Inference is based on permutation to account for overdispersion and small sample sizes. Traits can be either binary or continuous, and adjustment for confounders is supported. Our simulations indicate that LOCOM always preserved FDR and had much improved sensitivity over existing methods. In contrast, analysis of composition of microbiomes (ANCOM) and ANCOM with bias correction (ANCOM-BC)/ANOVA-Like Differential Expression tool (ALDEx2) had inflated FDR when the effect sizes were small and large, respectively. Only LOCOM was robust to experimental biases in every situation. The flexibility of our method for a variety of microbiome studies is illustrated by the analysis of data from two microbiome studies. Our R package LOCOM is publicly available.

Impact of experimental bias on compositional analysis of microbiome data

“…In this study, we found that LOCOM was robust to not only main effect biases but also a reasonable range of interaction biases. The other methods tended to have inflated FDR even when there were only main effect biases; many of them did not control the FDR even when there was no experimental bias at all (results shown in [2]). LOCOM maintained the 6 highest sensitivity among all methods even when the other methods did not control the FDR.…”

“…We considered both binary and continuous traits of 4 interest without any confounding covariates; we also considered a binary confounder when the trait was binary. We used the two sets of causal taxa (i.e., taxa that are associated with the trait) that were used in [2], namely, a random sample of 20 taxa (referred to as M1) and the five most abundant taxa (M2); we assumed a common effect size β of the trait on all causal taxa.…”

“…We considered both binary and continuous traits of interest without any confounding covariates; we also considered a binary confounder when the trait was binary. We used the two sets of causal taxa (i.e., taxa that are associated with the trait) that were used in [2], namely, a random sample of 20 taxa (referred to as M1) and the five most abundant taxa (M2); we assumed a common effect size β of the trait on all causal taxa. We designed three scenarios for the distribution of interaction biases: sampling the error term ϵ jj′ non-differentially based on N (0.5, 0.1 2 ) for all taxa (referred to as S-nondiff), differentially based on N (0.5, 0.1 2 ) and Beta (0.5, 0.5) (a bimodal distribution with larger variance than 0.1 2 ) for null and causal taxa, respectively (S-diff-causal), and differentially for a random selection of half of the taxa and the remaining taxa (S-diff-half).…”

“…Each taxon-specific bias factor represents the accumulation of multiplicative biases over all the steps in the experimental pipeline, so that multiple sources of bias are described by a single factor. With the MWC model in mind, we developed LOCOM [2], a logistic-regression-based compositional analysis method for detecting differentially abundant taxa, that only estimates parameters that are free from bias, i.e. are not affected by bias factors.…”

coda4microbiome: compositional data analysis for microbiome cross-sectional and longitudinal studies

“…Some methods used in microbiome analysis, such as ALDEx2 [ 13 ], LinDA [ 39 , 40 ], ANCOM [ 26 ], ANCOM-BC [ 23 ], fastANCOM [ 39 ] and LOOCM [ 21 ], perform the log-ratio approach to identify differential abundant taxa between two study groups. Here we introduce coda4microbiome , a new R package for analyzing microbiome data within the CoDA framework in both, cross-sectional and longitudinal studies.…”