Skip to main content
v2026.11,986 entries · CC-BY 4.0

MaAsLin 2 Guide: Installation, Statistical Models, and Association Analysis

MaAsLin 2 finds multivariable associations between clinical metadata and microbial features. Configure R commands, random effects, and interpret results.

Written and maintained by CASRAI Editorial Board

Published · Last updated

On this page

MaAsLin 2 (Microbiome Multivariable Associations with Linear Models) is an open-source Bioconductor R package and command-line utility developed by the Huttenhower Lab to discover statistically significant associations between complex clinical or phenotypic metadata and high-dimensional microbial meta-omics features. By fitting generalized linear models and linear mixed-effects models across taxonomic, functional, or metabolomic abundance profiles, MaAsLin 2 enables biomedical researchers to evaluate experimental exposures while rigorously controlling for confounding variables, sequencing batch effects, and repeated-measures longitudinal cohort designs.

What it is used for in research

Modern biomedical and ecological studies frequently evaluate complex microbial communities using high-throughput sequencing techniques, such as 16S ribosomal RNA (rRNA) gene amplicon sequencing, shotgun metagenomics, and metatranscriptomics. However, identifying true biological associations between microbial features (such as taxonomic clades or metabolic pathways) and phenotypic variables remains statistically challenging. Microbiome datasets are inherently high-dimensional, sparse (containing numerous structural and sampling zeros), overdispersed, and subject to compositional constraints where total sequence reads do not reflect absolute physical biomass. Furthermore, human population studies are rarely simple two-group comparisons; they involve diverse clinical covariates such as participant age, sex, body mass index (BMI), medication usage, dietary intake, and technical batch variation.

MaAsLin 2 was created to provide a comprehensive, flexible multivariable modeling framework capable of dissecting these complex interactions. Researchers deploy MaAsLin 2 across diverse study designs:

  • Cross-Sectional Epidemiology: Testing whether specific bacterial taxa or functional genes correlate with disease phenotypes (for example, inflammatory bowel disease, colorectal cancer, or metabolic syndrome) while simultaneously adjusting for patient demographics, lifestyle factors, and sequencing depth.
  • Longitudinal and Repeated-Measures Cohorts: Profiling microbiome dynamics over time during clinical drug trials, dietary interventions, or infant development. MaAsLin 2 incorporates random effects via linear mixed-effects models to account for intra-subject correlation across repeated samplings.
  • Functional Meta-Omics Profiling: Evaluating non-taxonomic features, such as metabolic pathways and gene family abundances generated by upstream pipelines like HUMAnN functional profiling, or metabolomics peak intensities from mass spectrometry assays.
  • Multi-Site and Multi-Cohort Syntheses: Integrating heterogeneous data from multi-center studies, such as the Human Microbiome Project (HMP), where technical variations across sequencing facilities and clinical collection sites must be accounted for as covariates or random grouping factors.

In extensive benchmark evaluations published by Mallick and colleagues in PLOS Computational Biology (2021), linear models applied to log-transformed relative abundances demonstrated superior overall control of false discovery rates while preserving statistical power when compared across eleven distinct association methods. This makes MaAsLin 2 a versatile default choice for computational biology laboratories seeking reproducible association discovery without the extreme false positive inflation often observed in unadjusted count models or exploratory ordination metrics calculated in packages such as the vegan package.

How to install or access it

MaAsLin 2 is officially distributed through the Bioconductor project and the bioBakery software ecosystem maintained by the Harvard T.H. Chan School of Public Health. The software requires R version 4.0 or higher. To install MaAsLin 2 from the official Bioconductor release repository, open an active R console and execute:

if (!requireNamespace("BiocManager", quietly = TRUE))
    install.packages("BiocManager")
BiocManager::install("Maaslin2")

BiocManager will resolve and install all core dependencies automatically. The primary underlying packages include lmerTest (for mixed-effects model fitting and degree-of-freedom estimation), pbapply (for progress bar tracking), car (for statistical testing and ANOVA tables), dplyr (for tabular data manipulation), vegan (for ecological data metrics), ggplot2 and pheatmap (for generating visualization plots), and logging (for runtime tracking).

Researchers who wish to test developmental features or specific branch patches directly from source can install the package from GitHub using the remotes package:

if (!requireNamespace("remotes", quietly = TRUE))
    install.packages("remotes")
remotes::install_github("biobakery/Maaslin2")

For high-performance computing (HPC) clusters and automated bioinformatic workflows, bioBakery also provides a standalone executable command-line wrapper script named Maaslin2.R. This script is packaged inside the Bioconductor library directory under inst/tools/Maaslin2.R or can be downloaded directly from the official bioBakery repository. Once downloaded and given executable permissions (chmod +x Maaslin2.R), the script can be invoked directly from any standard Linux or macOS terminal shell without initiating an interactive R session.

Step-by-step instructions

Executing a multivariable association analysis in MaAsLin 2 involves a sequence of structured steps, from data formatting through statistical execution. Below is the complete workflow using official parameters and syntax.

1. Prepare Input Data and Metadata Tables

MaAsLin 2 requires two primary input files: a feature abundance matrix (input_data) and a sample metadata matrix (input_metadata). Both files must be formatted as tab-delimited text files (.tsv) or loaded directly into R as standard data frames:

  • Feature Abundance Table: Contains quantitative measurements (e.g., relative abundances, raw read counts, or normalized gene copies). Features (e.g., species, OTUs, pathways) are listed in columns and samples in rows, or vice versa (MaAsLin 2 automatically detects orientation based on metadata sample identifiers).
  • Metadata Table: Contains experimental and clinical variables. Samples must be listed as rows, and metadata variables (e.g., treatment group, age, sex, cohort, subject ID) must be listed as columns. The sample row names in the metadata file must match the sample identifiers in the feature table exactly.

2. Select Normalization and Transformation Schemes

Before fitting statistical models, MaAsLin 2 applies data normalization to address sequencing depth differences, followed by mathematical transformation to stabilize variance and normalize feature distributions:

  • Normalization (normalization): The default is "TSS" (Total Sum Scaling), which converts feature counts into relative abundances by dividing each feature by the sample sum. Alternative options include "CSS" (Cumulative Sum Scaling), "CLR" (Centered Log-Ratio), and "NONE" (used if input data has already undergone upstream normalization).
  • Transformation (transform): The default is "LOG", which computes the natural logarithm of normalized abundances with a pseudo-count offset. Alternative options include "LOGODDS" (logit transformation for proportion data bounded between 0 and 1) and "NONE".

3. Configure Fixed Effects, Random Effects, and Factor References

Specify which metadata columns represent primary variables of interest versus nuisance covariates:

  • Fixed Effects (fixed_effects): A vector of column names corresponding to variables tested for association (e.g., fixed_effects = c("diagnosis", "dysbiosis", "antibiotics", "age")).
  • Random Effects (random_effects): A vector of grouping columns for repeated-measures designs (e.g., random_effects = c("subject_id")). When specified, MaAsLin 2 switches its internal engine from standard linear regression (lm) to linear mixed models via lmerTest::lmer.
  • Categorical Reference Levels (reference): When evaluating categorical metadata with more than two factor levels, you must explicitly declare the baseline reference group. Format this parameter as a comma-separated string containing the variable name and the reference level, such as reference = c("diagnosis,nonIBD").
  • Standardization (standardize): Defaults to TRUE, which applies z-score transformation (subtracting the mean and dividing by the standard deviation) to continuous metadata variables, allowing effect size coefficients to be compared on a standardized scale.

4. Execute the Model in R

Open your R script or RStudio console and run the analysis using the core Maaslin2() function:

library(Maaslin2)

# Load abundance table and metadata
abundance_data <- read.table("HMP2_taxonomy.tsv", header = TRUE, sep = "t", row.names = 1, check.names = FALSE)
metadata <- read.table("HMP2_metadata.tsv", header = TRUE, sep = "t", row.names = 1, check.names = FALSE)

# Execute MaAsLin 2
fit_output <- Maaslin2(
    input_data = abundance_data,
    input_metadata = metadata,
    output = "maaslin2_results_directory",
    fixed_effects = c("diagnosis", "dysbiosis", "antibiotics", "age"),
    random_effects = c("subject_id"),
    reference = c("diagnosis,nonIBD"),
    normalization = "TSS",
    transform = "LOG",
    analysis_method = "LM",
    max_significance = 0.25,
    correction = "BH",
    standardize = TRUE,
    cores = 4
)

5. Execute the Model via Terminal CLI (Maaslin2.R)

For automated shell scripts and cluster job submission systems (e.g., Slurm or SGE), execute the identical model using the standalone command-line script:

Rscript Maaslin2.R 
    --fixed_effects="diagnosis,dysbiosis,antibiotics,age" 
    --random_effects="subject_id" 
    --reference="diagnosis,nonIBD" 
    --normalization=TSS 
    --transform=LOG 
    --analysis_method=LM 
    --max_significance=0.25 
    --correction=BH 
    --standardize=TRUE 
    --cores=4 
    HMP2_taxonomy.tsv 
    HMP2_metadata.tsv 
    maaslin2_results_directory

Output and how to read it

Upon completion of an analysis run, MaAsLin 2 generates a structured output directory containing tabular results, diagnostic execution logs, and publication-ready figures. Understanding the file organization and column definitions allows researchers to interpret biological findings accurately.

Directory Organization and File Descriptions

  • all_results.tsv: The complete tabular summary of every tested feature-metadata association evaluated by the model, ordered in ascending order of q-values (most statistically significant associations first).
  • significant_results.tsv: A filtered subset of all_results.tsv containing only those associations meeting the false discovery rate threshold defined by max_significance (default q ≤ 0.25).
  • maaslin2.log: The detailed system log tracking execution steps, filtering cutoffs, sample retention counts, model fit warnings, and computational time.
  • figures/ Directory: An automatically generated subfolder containing graphical visualizations. When plot_heatmap = TRUE (default), MaAsLin 2 generates a summary heatmap (heatmap.pdf) illustrating the effect sizes of significant associations. When plot_scatter = TRUE (default), individual boxplots (for categorical variables) and scatter plots with regression trendlines (for continuous variables) are generated for each significant feature.

Interpreting Columns in all_results.tsv

The primary result table contains nine standardized columns:

  • feature: The specific microbial clade, OTU, functional gene family, or metabolic pathway identifier tested.
  • metadata: The clinical, experimental, or demographic covariate associated with the feature.
  • value: For categorical variables, the specific non-reference factor level being compared against the baseline reference group. For continuous variables, this field typically matches the metadata name or indicates continuous slope evaluation.
  • coef: The estimated model coefficient (effect size). In linear models with log-transformed relative abundance (transform = "LOG"), the coefficient represents the change in log abundance per unit change in continuous metadata, or the log-abundance difference between the indicated category and the reference baseline. A positive coefficient indicates enrichment; a negative coefficient indicates depletion.
  • stderr: The standard error associated with the model coefficient estimate, reflecting measurement precision.
  • pval: The unadjusted nominal p-values derived from the two-tailed Student t-test or Wald test evaluating whether the coefficient differs significantly from zero.
  • qval: The adjusted p-value after controlling for multiple hypothesis testing across all evaluated feature-metadata pairs. By default, MaAsLin 2 applies the Benjamini-Hochberg (BH) false discovery rate method.
  • N: The total number of valid samples utilized in fitting the model for that specific feature after dropping samples with missing values in required covariates.
  • N.not.0: The count of samples in which the feature was detected with non-zero abundance. Features with low N.not.0 values should be inspected carefully to avoid overinterpreting associations driven by a few isolated non-zero outliers.

Calculating Fold Changes from Log Coefficients

Because MaAsLin 2 utilizes natural logarithms by default (transform = "LOG"), the reported coefficient β can be directly converted into an estimated relative fold change between groups using the exponential function:

Fold Change = exp(β)

To express effect sizes in standard log2 fold change format (commonly used in volcano plots and genomic literature), divide the natural log coefficient by the natural logarithm of 2:

log2(Fold Change) = β / ln(2) ≈ β × 1.4427

For example, an estimated coefficient of β = 0.693 corresponds to a 2.0-fold relative enrichment in the target phenotype compared to the reference control (exp(0.693) ≈ 2.0, or log2(Fold Change) = 1.0).

Troubleshooting

While MaAsLin 2 is robust and widely adopted, computational biologists and laboratory researchers commonly encounter specific configuration and data formatting issues. Below are verified diagnostic steps and solutions based on official bioBakery documentation and community support records.

1. Error: ‘Please provide the reference for the variable which includes more than 2 levels’

Root Cause: When a metadata variable is categorical and contains three or more distinct levels (for instance, a cohort variable with levels Control, Mild, and Severe), MaAsLin 2 cannot guess which category represents the baseline control group. Without an explicit baseline, contrast coefficients cannot be computed.

Solution: Explicitly pass the reference level in the reference argument as a vector formatted as c("variable_name,reference_level"). Ensure there are no stray spaces around the comma:

# Correct syntax:
reference = c("DiseaseStatus,Control", "Site,Boston")

Alternatively, pre-format categorical columns as factors in R and assign the baseline level using the base relevel() function prior to invoking Maaslin2():

metadata$DiseaseStatus <- factor(metadata$DiseaseStatus)
metadata$DiseaseStatus <- relevel(metadata$DiseaseStatus, ref = "Control")

2. Singular Fit Warnings and NA Results in Random Effects Models

Root Cause: When fitting linear mixed-effects models via random_effects, the underlying lmerTest package may issue boundary (singular) fit: see help('isSingular') warnings, resulting in NA entries for standard errors and p-values. This happens when the random effect grouping variable has near-zero variance, when the grouping factor is improperly formatted as a continuous numeric vector, or when subjects possess only a single observation (lacking repeated measures).

Solution: Ensure that the grouping variable is explicitly converted into a factor (metadata$subject_id <- as.factor(metadata$subject_id)). Verify your study design: if participants were only sampled once, random subject intercepts cannot be estimated and random_effects must be omitted. If certain subjects have only single timepoints within a multi-timepoint cohort, ensure there are sufficient multi-sample subjects to allow parameter estimation.

3. Sample Identifier Mismatches Between Data and Metadata

Root Cause: If row names in input_metadata do not match sample column or row names in input_data, MaAsLin 2 will either throw an error (No common samples found) or silently drop mismatched samples, severely reducing sample size N.

Solution: In R, verify intersection prior to analysis:

common_samples <- intersect(rownames(metadata), colnames(abundance_data))
if (length(common_samples) == 0) {
    common_samples  0)

Also check for hidden carriage returns in files exported from Microsoft Excel or Windows text editors.

4. Model Convergence Failures in CPLM and ZINB Models

Root Cause: Alternative statistical engines such as Compound Poisson Linear Models (analysis_method = "CPLM") or Zero-Inflated Negative Binomial models (analysis_method = "ZINB") rely on iterative numerical optimization algorithms that frequently fail to converge on highly sparse microbiome tables or datasets with modest sample sizes (N under 100).

Solution: For general microbiome epidemiology, use the benchmarked default analysis_method = "LM" with normalization = "TSS" and transform = "LOG". Mallick et al. (2021) demonstrated that linear regression on log-transformed relative abundance matches or outperforms zero-inflated models in statistical power and empirical FDR control while avoiding optimization crashes.

5. Unexpected Loss of Rare Features

Root Cause: By default, MaAsLin 2 applies filtering parameters: min_abundance = 0.0 and min_prevalence = 0.1. This removes any feature not present in at least 10% of samples, which may exclude rare taxa of interest.

Solution: If you intentionally wish to evaluate rare taxa, lower the prevalence cutoff (e.g., min_prevalence = 0.05 for 5% prevalence, or min_prevalence = 0.0 to disable prevalence filtering entirely).

Free alternatives

While MaAsLin 2 is a standard multivariable framework for meta-omics association analysis, several alternative open-source tools exist within the R and Bioconductor bioinformatics ecosystem. Choosing the most suitable package depends on your experimental design, whether you are analyzing raw read counts or relative abundances, and whether you require random-effects modeling for repeated measurements.

The data table below outlines five established open-source tools for differential abundance analysis, comparing their core statistical models, handling of zero inflation and compositionality, and practical strengths.

Bioconductor / GitHub
General Linear Models (LM) and Linear Mixed Models (LMM)
Complex multivariable epidemiological studies with continuous and categorical covariates and repeated measures
MIT License

Bioconductor
Log-linear models with sampling fraction bias correction
Cross-sectional or longitudinal microbiome studies requiring explicit bias-corrected absolute abundance inference
Artistic-2.0

Bioconductor
Dirichlet-multinomial sampling with Centered Log-Ratio (CLR) and Welch/Wilcoxon tests
Small to medium sample size cohorts where compositional data analysis principles must be strictly enforced
GPL-3.0

CRAN / GitHub
Linear models with bias-corrected Centered Log-Ratio (CLR)
High-dimensional microbiome datasets requiring rapid mixed-effects modeling with rigorous compositional control
GPL-2.0

Bioconductor
Negative Binomial Generalized Linear Models with dispersion shrinkage
Unscaled raw sequence count tables from RNA-seq adapted to microbiome profiling without zero inflation
LGPL (≥ 2)

Software Tool Primary Package Repository Statistical Modeling Paradigm Optimal Use Case Open Source License
MaAsLin 2
ANCOM-BC2
ALDEx2
LinDA
DESeq2

Researchers conducting large-scale epidemiological studies with dozens of clinical covariates typically select MaAsLin 2 due to its fast execution, straightforward covariate adjustment, and native support for both R scripts and cluster-friendly command-line execution. When compositional sampling bias is the primary theoretical concern, tools such as ANCOM-BC2 and LinDA provide valuable complementary frameworks.

Frequently asked questions

What is the primary difference between MaAsLin 2 and ANCOM-BC?

MaAsLin 2 uses generalized linear models and linear mixed models on transformed relative abundances (typically Total Sum Scaling followed by log transformation), offering high computational speed, robust false discovery rate control, and flexible covariate modeling. In contrast, ANCOM-BC (Analysis of Compositions of Microbiomes with Bias Correction) explicitly estimates a sample-specific sampling fraction bias parameter to infer changes in absolute microbial abundance rather than relative proportions. While ANCOM-BC is theoretically tailored to compositional bias correction, MaAsLin 2 is considerably faster and natively accommodates multi-factor random effects designs across large cohort sizes.

How does MaAsLin 2 handle zero values in abundance tables?

Microbiome sequencing matrices frequently contain zero values due to both true biological absence (structural zeros) and insufficient sequencing depth (sampling zeros). When using the default log transformation (transform = "LOG"), MaAsLin 2 adds a half-minimum non-zero positive offset (pseudo-count) to zero values before taking the logarithm. This prevents mathematical undefined values (-∞) while preserving rank ordering across low-abundance features.

Can MaAsLin 2 analyze longitudinal or repeated-measures microbiome data?

Yes. MaAsLin 2 has native support for longitudinal cohort designs and repeated samplings via the random_effects parameter. When one or more metadata grouping variables (such as subject_id or family_id) are passed to random_effects, MaAsLin 2 internally invokes lmerTest::lmer to fit linear mixed-effects models, estimating random intercepts to account for intra-subject dependency across temporal observations.

What is the default significance threshold in MaAsLin 2?

The default false discovery rate (FDR) significance threshold is controlled by max_significance = 0.25. All tested feature-metadata associations are written to all_results.tsv, but only those associations with an adjusted q-value less than or equal to 0.25 are written to significant_results.tsv and visualized in diagnostic plots. In clinical and discovery studies, researchers frequently adjust this parameter to max_significance = 0.05 or 0.10 to enforce stricter statistical confidence.

How should continuous metadata variables be scaled before running MaAsLin 2?

By default, MaAsLin 2 automatically standardizes continuous metadata columns by setting standardize = TRUE. This applies a z-score transformation (subtracting the variable mean and dividing by the standard deviation), ensuring that model coefficients represent the change in log abundance per standard deviation of the continuous variable. This standardization enables direct comparison of effect sizes across continuous variables measured on different numerical scales, such as age in years, BMI in kg/m², and blood laboratory markers.

What input file formats are supported by MaAsLin 2?

MaAsLin 2 accepts standard tab-delimited text files (.tsv or .txt) as well as comma-separated value files (.csv). Within an R session, inputs can also be passed directly as R data.frame or matrix objects. The abundance table can be oriented with samples as rows and features as columns or vice-versa, as MaAsLin 2 automatically matches sample identifiers against metadata row names.

Sources

  • Mallick, H., Rahnavard, A., McIver, L. J., Ma, S., Zhang, Y., Nguyen, L. H., Tickle, T. L., Weingart, G., Ren, B., Schwager, E. H., Chatterjee, S., Thompson, K. N., Wilkinson, J. E., Subramanian, A., Lu, Y., Waldron, L., Paulson, J. N., Franzosa, E. A., Corrada Bravo, H., and Huttenhower, C. (2021). Multivariable association discovery in population-scale meta-omics studies. PLOS Computational Biology, 17(11), e1009442. https://doi.org/10.1371/journal.pcbi.1009442 (Accessed 2026-11-15).
  • Huttenhower Lab. (2024). MaAsLin 2 User Manual. Harvard T.H. Chan School of Public Health. http://huttenhower.sph.harvard.edu/maaslin2 (Accessed 2026-11-15).
  • bioBakery. (2024). Maaslin2: Microbiome Multivariable Association with Linear Models (GitHub Repository and Documentation). https://github.com/biobakery/Maaslin2 (Accessed 2026-11-15).
  • Bioconductor. (2024). Maaslin2: Multivariable Association Discovery in Population-scale Meta-omics Studies. Bioconductor Package Repository. https://bioconductor.org/packages/release/bioc/html/Maaslin2.html (Accessed 2026-11-15).
  • Lin, H., and Peddada, S. D. (2020). Analysis of compositions of microbiomes with bias correction. Nature Communications, 11(1), 4114. https://doi.org/10.1038/s41467-020-17041-7 (Accessed 2026-11-15).
  • Fernandes, A. D., Reid, J. N., Macklaim, J. M., McMurrough, T. A., Edgell, D. R., and Gloor, G. B. (2014). Unifying the analysis of high-throughput sequencing datasets: characterizing RNA-seq, 16S rRNA gene sequencing and selective growth experiments by composition. Microbiome, 2, 15. https://doi.org/10.1186/2049-2618-2-15 (Accessed 2026-11-15).
  • Zhou, C., Hu, J., Zhou, Y., and Shen, N. (2022). LinDA: linear models for differential abundance analysis of microbiome compositional data. Genome Biology, 23(1), 95. https://doi.org/10.1186/s13059-022-02655-5 (Accessed 2026-11-15).

Data: Microbiome Differential Abundance Statistical Frameworks Comparison

Comparative synthesis of open-source statistical software packages used for differential abundance and multivariable association analysis in microbiome and meta-omics research, compiled from peer-reviewed software publications and official Bioconductor documentation.

Microbiome Differential Abundance Statistical Frameworks Comparison
software_packageprimary_statistical_modeldefault_normalizationcompositionality_handlinglongitudinal_supportcli_availability
MaAsLin 2General Linear Model (LM) / Linear Mixed Model (LMM)Total Sum Scaling (TSS)Log-transformed relative abundance or CLRYes (via lmerTest random effects)Yes (Maaslin2.R script)
ANCOM-BC2Linear Regression with Sampling Fraction Bias CorrectionIterative sample-specific bias estimationSample-specific log-scale bias correctionYes (random intercept/slope models)No (R package interface only)
ALDEx2ANOVA-like Welch t-test / Wilcoxon / GLMCentered Log-Ratio (CLR)Monte Carlo Dirichlet distribution samplingNo (primarily standard GLM fixed effects)No (R package interface only)
LinDALinear Regression with Mode-based Bias CorrectionCentered Log-Ratio (CLR)Mode-based compositional bias correctionYes (via lme4 mixed-effects integration)No (R package interface only)
DESeq2Negative Binomial Generalized Linear Model (GLM)Median-of-ratios library scalingLog-ratio scaling / variance stabilizing transformNo (standard fixed-effects GLM only)No (R package interface only)

Sources

Compiled by CASRAI from the sources above. Free to reuse under CC BY 4.0 with attribution.

Follow CASRAI

Research-administration guidance, standards updates and independent tool reviews.

Ask CASRAI · free to try

Ask about MaAsLin 2 Guide: Installation, Statistical Models, and Association Analysis

2 free questions a day, no account.

An AI assistant for research: funder rules, compliance, methods and publishing. It links the sources behind its answers and labels answers that come from a web search.

Answers draw on CASRAI's guides and dictionary plus the federal and funder documents we index: Federal Register, Grants.gov, Regulations.gov and UKRI.

Works on this site and inside Claude, Cursor and the AI tools you already use.

Everything CASRAI publishes — this page, the dictionary, the guides and the news — stays open to read, with no account and no card.

Referenced across the research world

University of Cambridge logoColumbia University logoCrossref logoUniversity of Edinburgh logoHarvard University logoUniversity of Oxford logoPrinceton University logoStanford School of Medicine logoUniversity College London logoORCID logoUniversity of Cambridge logoColumbia University logoCrossref logoUniversity of Edinburgh logoHarvard University logoUniversity of Oxford logoPrinceton University logoStanford School of Medicine logoUniversity College London logoORCID logo
  • University of Cambridge logo
  • Columbia University logo
  • Crossref logo
  • University of Edinburgh logo
  • Harvard University logo
  • University of Oxford logo
  • Princeton University logo
  • Stanford School of Medicine logo
  • University College London logo
  • ORCID logo

View CASRAI adoption →