henhao
Sign in
MicrobiologyNature Methods
Peer-reviewed journal article

MaAsLin 3: Refining and extending generalized multivariable linear models for meta-omic association discovery

William A. NickolsiD, Thomas M. KuntziD, Jiaxian SheniD, Sagun MaharjaniD, Himel MallickiD, Eric A. FranzosaiD, Kelsey N. ThompsoniD, Jacob T. NearingiD, Curtis HuttenhoweriD

Abstract

A key question in microbial community analysis is determining which microbial features are associated with community properties such as environmental or health phenotypes. This statistical task is impeded by characteristics of typical microbial community profiling technologies, including sparsity (which can be either technical or biological) and the compositionality imposed by most nucleotide sequencing approaches. Many models have been proposed that focus on how the relative abundance of a feature (e.g. taxon or pathway) relates to one or more covariates. Few of these, however, simultaneously control false discovery rates, achieve reasonable power, incorporate complex modeling terms such as random effects, and also permit assessment of prevalence (presence/absence) associations and absolute abundance associations (when appropriate measurements are available, e.g. qPCR or spike-ins). Here, we introduce MaAsLin 3 (Microbiome Multivariable Associations with Linear Models), a modeling framework that simultaneously identifies both abundance and prevalence relationships in microbiome studies with modern, potentially complex designs. MaAsLin 3 also newly accounts for compositionality with experimental (spike-ins and total microbial load estimation) or computational techniques, and it expands the space of biological hypotheses that can be tested with inference for new covariate types. On a variety of synthetic and real datasets, MaAsLin 3 outperformed current state-of-the-art differential abundance methods in testing and inferring associations from compositional data. When applied to the Inflammatory Bowel Disease Multi-omics Database, MaAsLin 3 corroborated many previously reported microbial associations with the inflammatory bowel diseases, but notably 77% of associations were with feature prevalence rather than abundance. In summary, MaAsLin 3 enables researchers to identify microbiome associations with higher accuracy and more specific association types, especially in complex datasets with multiple covariates and repeated measures.

PDFDOICodeData
Full-text digest ↓

02 /Specification & Classification

Click any tag to browse matching papers
Citations
191
FWCI
269.60
Field percentile
top 1%
OpenAlex Topics

In-depth summary

✦ Multi-agent verified full-text synthesis

Claude Opus 5.5 · xhigh effort · 5 Nov 2026 · from the full text of version 1 (14 Dec 2024) · CC BY 4.0

MaAsLin 3 fits hurdle models separating microbial prevalence (augmented logistic) and abundance (median-adjusted log-linear) associations, improving precision over existing differential abundance tools and showing 77% of IBD associations are prevalence-driven.

Microbiome differential abundance testing is confounded by zero-inflation, compositionality, and complex study designs, and existing tools conflate presence/absence shifts with quantitative abundance changes or rely on pseudocounts that bias effect sizes. MaAsLin 3 separates prevalence and non-zero abundance modeling while adjusting for compositionality computationally via median coefficient comparison or experimentally via spike-ins and total microbial load.

Key results

  1. ShownIn synthetic SparseDOSSA 2 benchmarks, MaAsLin 3 achieved higher or equal F1 scores and less coefficient shrinkage than ALDEx2, ANCOM-BC2, and MaAsLin 2 across sample sizes above 50.

    MaAsLin 3 increased precision by up to 0.29 on average compared with MaAsLin 2 without losing recall, held average precision >= 0.82 across sample sizes (0.99 at 50 samples vs 0.89 for MaAsLin 2 and 0.54 for ANCOM-BC2), and reached 0.83 Spearman correlation with true abundance coefficients at 1,000 samples (vs 0.80 for ANCOM-BC2, 0.62 for MaAsLin 2, and 0.59 for ALDEx2). · Introduction; MaAsLin 3 improves precision and accuracy…

  2. ShownIn longitudinal designs and under ANCOM-BC structural-zero model violations, MaAsLin 3 maintained high recall and precision, with post-hoc effect thresholding removing compositionality-induced false positives.

    With 10 repeated samples per subject, thresholding coefficients at |β| > 1 restored MaAsLin 3 average precision from 0.75 to >= 0.92 with 0.80 recall (vs 0.55 recall for ALDEx2 and <= 0.33 for ANCOM-BC2); under ANCOM-BC simulations, MaAsLin 3 and ANCOM-BC2 achieved 0.96 precision at 50 samples (0.64 vs 0.59 at 1,000 samples) while ALDEx2 and MaAsLin 2 never exceeded 0.49. · MaAsLin 3 improves precision and accuracy…; MaAsLin 3 is robust to violations of modeling…

  3. ShownMaAsLin 3 relative-abundance coefficients accurately approximated experimental absolute-abundance coefficients in adult human and mouse cohorts, but diverged in a highly sparse infant gut cohort.

    In mouse diet (66% zeros, dPCR) and human IBD/PSC (80% zeros, flow cytometry) cohorts, MaAsLin 3 and ANCOM-BC2 reached mean Spearman r of 0.96 and 0.95 and slopes of 0.97 and 0.97 against experimental absolute coefficients (vs r = 0.52 and 0.59 and slopes = 0.50 and 0.48 for ALDEx2 and MaAsLin 2), while correlation dropped to 0.47 in the infant gut spike-in dataset (97% zeros). · MaAsLin 3 accurately approximates absolute…; Real absolute abundance data

  4. ShownApplying MaAsLin 3 to the HMP2 Inflammatory Bowel Disease Multi-omics Database revealed that species-level IBD associations are predominantly prevalence losses, whereas metatranscriptomic pathway shifts are mainly abundance changes.

    Across 1,637 HMP2 metagenomes (750 CD, 459 UC, 428 non-IBD from 130 individuals), 287 of 372 significant associations (77%) were prevalence associations (89% negative, including Dysosmobacter welbionis prevalence β = -4.14 and -3.46, q = 1.5 × 10^-4 and 5.1 × 10^-9 vs abundance q = 0.24 and 1), while 142 of 209 metatranscriptomic pathway associations (68%, DNA-adjusted) were abundance associations. · Abstract; MaAsLin 3’s linear model extensions enable new…; MaAsLin 3 clarifies and refines biomarker…; HMP2-IBDMDB analysis

Methods, data and reproducibility

MaAsLin 3 normalizes feature tables (via total-sum, spike-in, or total biomass scaling), splits data into binary presence/absence and log2 non-zero abundances, and fits Diaconis-Ylvisaker logistic models (with abundance-induced prevalence screening) plus linear or lme4 mixed models tested against the median coefficient. Supports omnibus ANOVA/LRT, ordered predictors, contrast tests, and feature-specific covariates.

  • Synthetic microbiome datasets from SparseDOSSA 2 (Stool template) and ANCOM-BC generator (soil template) · human and soil microbial communities (simulated) · n = 100 synthetic datasets per simulation setting (20 to 1,000 samples)
  • Real microbiome datasets with binary metadata for permutation null benchmarking · microbial communities · n = 38 datasets randomized 100 times each (3,800 replications)
  • Experimental absolute-abundance cohorts: infant gut (spike-in), mouse diet (16S digital PCR), and human IBD/PSC (flow cytometry) · human, mouse, gut microbiome · n = 3 datasets with 97%, 66%, and 80% zero sparsity
  • HMP2 Inflammatory Bowel Disease Multi-omics Database (IBDMDB) stool metagenomes and metatranscriptomes · human gut microbiome · n = 1,637 metagenomes (750 CD, 459 UC, 428 non-IBD) + metatranscriptomes (130 hosts)
Methods & tools
MaAsLin 3, MaAsLin 2, ALDEx2, ANCOM-BC2, SparseDOSSA 2, MetaPhlAn 4, MetaPhlAn 3, HUMAnN 3, KneadData, lme4, lmerTest, multcomp
Code
https://huttenhower.sph.harvard.edu/maaslin3https://github.com/biobakery/MaAsLin3https://github.com/willnickols/MaAsLin3_benchmarkhttps://github.com/biobakery/biobakery/wiki/MaAsLin3/https://github.com/biobakery/biobakery/wiki/MTX-model-3
Data & accessions
https://github.com/willnickols/MaAsLin3_benchmark
Software & environment
R/Bioconductor: MaAsLin 3 v3.0.0, ALDEx2 v1.34.0, ANCOM-BC2 v2.4.0, MaAsLin 2 v1.16.0, MetaPhlAn 4.0.6 (Oct22 database), MetaPhlAn 3.0, HUMAnN 3, KneadData, SparseDOSSA 2, lme4, lmerTest, multcomp
Limitations (3)
  • At small sample sizes (e.g., 50 samples), MaAsLin 3 has lower recall (0.19) than MaAsLin 2 (0.27) or ANCOM-BC2 (0.23) because logistic regression requires more observations and linear models use only non-zero values. · stated by the authors
  • Finite sequencing read depth can cause rare features with true abundance decreases to fall below the detection threshold, yielding missed associations or spurious prevalence associations. · stated by the authors
  • Inferring absolute abundance effects from relative abundances via median comparison becomes biased when more than half of features change in absolute abundance or when total community biomass depends on blooming taxa (as in the infant gut). · stated by the authors

Who should read:Read this if you analyze microbiome taxonomic, functional, or metatranscriptomic profiles and need multivariable mixed-effects models that separate presence/absence from abundance associations.

Reproducibility ledger

Nobody has reported trying to reproduce this paper yet. Did you replicate an experiment or run the code? Your report saves the next lab a week.

Sign in to file a reproducibility report.

Related papers

Similarity Map
SelectedSame fieldCross-field
THISMallick '21Yang '16Yang '22Hu '21Cougoul '19Kurtz '19
84% semantic matchMicrobiology
H. Mallick, A. Rahnavard, L. J. McIver+17 more·bioRxiv·Jan 20, 2021