HUMAnN: Metagenomic Functional Profiling Pipeline

HUMAnN (HMP Unified Metabolic Analysis Network) is an open-source bioinformatics pipeline that profiles the presence, absence, and abundance of microbial metabolic pathways and molecular functions from metagenomic (DNA) and metatranscriptomic (RNA) sequencing data. Developed by the Huttenhower Lab as part of the bioBakery suite, HUMAnN quantifies what microbial communities can do and what they actively transcribe. For foundational context, see our guides on what is metagenomics and what is bioinformatics.

What it is

In microbiome research, high-throughput sequencing produces millions of short reads from complex microbial communities. When raw sequence reads are deposited in repositories like the Sequence Read Archive (SRA) with registered BioSample records, researchers need reproducible pipelines to translate nucleotide data into functional profiles.

HUMAnN addresses this through a tiered mapping strategy. Rather than searching every read against universal protein databases, HUMAnN uses taxonomic information to construct a compact, sample-specific pangenome database. Reads that map to these pangenomes are classified at the nucleotide level using Bowtie2. Only remaining unmapped reads undergo translated protein search using DIAMOND. This tiered architecture provides a 10-fold to 100-fold speedup over brute-force translated searches while preserving sensitivity.

HUMAnN also produces taxonomically stratified functional profiles. For each identified gene family and metabolic pathway, it reports total community abundance alongside the specific contribution from each detected species, showing whether a metabolic function is shared across multiple organisms or restricted to a single taxon.

Who runs it and what it covers

HUMAnN is developed and maintained by the Huttenhower Lab at the Harvard T.H. Chan School of Public Health and the Broad Institute. Originally created under the National Institutes of Health (NIH) Human Microbiome Project (HMP), the software has evolved across three major versions:

• HUMAnN 1: Established functional orthology mapping and metabolic pathway reconstruction from metagenomic reads.
• HUMAnN 2: Added tiered nucleotide-to-translated search and taxonomic stratification to connect species identity with gene abundance.
• HUMAnN 3: Integrated into the bioBakery 3 ecosystem with updated pangenomes, expanded UniRef protein clusters, MetaPhlAn 3 coupling, and refined pathway algorithms.

Released under the open-source MIT License, HUMAnN runs on Linux and macOS. It processes shotgun sequencing datasets from human body sites, environmental soils, marine habitats, and industrial bioreactors. HUMAnN handles both metagenomic DNA (measuring functional potential) and metatranscriptomic RNA (measuring functional expression) using three core reference databases:

• ChocoPhlAn: A database of species-specific microbial pangenomes used for nucleotide alignment. MetaPhlAn identifies community members, and HUMAnN extracts matching pangenomes for Bowtie2 mapping.
• UniRef (UniProt Reference Clusters): Non-redundant protein databases clustered at 50% (UniRef50) or 90% (UniRef90) identity. DIAMOND queries unmapped reads against UniRef to detect homologs in diverged taxa.
• MetaCyc: A curated database of metabolic pathways and reactions. HUMAnN maps gene families to MetaCyc reactions to compute pathway abundance and coverage, alongside functional cross-references like KEGG.

How to search or use it

HUMAnN is operated via command-line utilities. The standard workflow covers environment setup, database downloads, sample execution, and cohort post-processing across 8 operational steps:

1. Install the software environment: Install HUMAnN, MetaPhlAn, Bowtie2, and DIAMOND using Conda:

   conda create -n biobakery -c conda-forge -c bioconda -c biobakery humann

2. Download reference databases: Download ChocoPhlAn pangenomes and DIAMOND-indexed UniRef90 databases:

   humann_databases --download chocophlan full /path/to/databases
   humann_databases --download uniref uniref90_diamond /path/to/databases

   Configure database paths:

   humann_config --update database_folders nucleotide /path/to/databases/chocophlan
   humann_config --update database_folders protein /path/to/databases/uniref

3. Preprocess and decontaminate input reads: Filter raw sequencing files to remove host-derived reads and low-quality bases using quality-control tools like KneadData.
4. Run the profiling pipeline: Execute the primary command on quality-controlled FASTQ or gzip files:

   humann --input sample_01.fastq.gz --output sample_01_output --threads 8

   HUMAnN automates taxonomic prescreening, pangenome mapping, translated alignment, and pathway scoring.

5. Inspect primary output tables: HUMAnN generates three tab-delimited files per sample:
   • sample_01_genefamilies.tsv: Gene family abundances (UniRef IDs) in RPK (Reads Per Kilobase), showing community totals and species-stratified rows.
   • sample_01_pathabundance.tsv: Abundances of complete MetaCyc metabolic pathways.
   • sample_01_pathcoverage.tsv: Pathway completeness scores from 0.0 to 1.0.
6. Merge cohort tables: Combine per-sample results into unified cohort matrices:

   humann_join_tables --input ./cohort_results --output cohort_genefamilies.tsv --file_name genefamilies
   humann_join_tables --input ./cohort_results --output cohort_pathabundance.tsv --file_name pathabundance

7. Regroup gene families: Map UniRef gene families into broader functional groups, such as MetaCyc reactions or Enzyme Commission (EC) numbers:

   humann_regroup_table --input cohort_genefamilies.tsv --groups uniref90_rxn --output cohort_reactions.tsv
   humann_regroup_table --input cohort_genefamilies.tsv --groups uniref90_level4ec --output cohort_ec.tsv

8. Normalize abundance tables: Normalize raw RPK values to Copies Per Million (CPM) or relative abundance:

   humann_renorm_table --input cohort_genefamilies.tsv --units cpm --output cohort_genefamilies_cpm.tsv
   humann_renorm_table --input cohort_pathabundance.tsv --units relab --output cohort_pathabundance_relab.tsv

Strengths and limits

Methodological strengths and operational boundaries of HUMAnN include:

• Tiered alignment efficiency: Nucleotide screening against sample pangenomes reduces reliance on translated searches, lowering compute times on large cohorts.
• Taxonomic stratification: Quantifies functional contributions at both community and species levels, revealing taxonomic drivers of metabolic pathways.
• Coordinated DNA and RNA profiling: Uses identical functional definitions across metagenomic and metatranscriptomic inputs for direct comparison of potential versus expression.
• Integrated post-processing: Command-line utilities handle table merging, functional regrouping, and normalization.
• Storage and memory requirements: Uncompressed ChocoPhlAn and DIAMOND-indexed UniRef90 databases require substantial disk space and server RAM.
• Reference database dependence: Functional assignment relies on reference genomes and UniRef clusters; sequences from novel lineages accumulate in unmapped categories.
• Metabolic focus: Pathway reconstruction relies on MetaCyc, emphasizing microbial metabolism over uncharacterized secondary pathways.

Alternatives

Researchers evaluate functional profiling tools based on input data types, compute infrastructure, and required outputs:

Tool / Platform	Sequencing Input	Profiling Mechanism	Primary Databases	Taxonomic Stratification
HUMAnN	Shotgun metagenomes (DNA) & metatranscriptomes (RNA)	Tiered nucleotide (Bowtie2) and translated (DIAMOND) read mapping	ChocoPhlAn, UniRef50/UniRef90, MetaCyc	Yes (community-total and species-stratified abundances)
PICRUSt2	Marker gene amplicon sequences (16S, 18S, ITS)	Phylogenetic placement into reference tree and gene copy imputation	Greengenes, SILVA, MetaCyc, KEGG, EC	Limited (inferred through ancestral genome reconstruction)
MEGAN6	Shotgun reads, long reads, or translated alignments	DIAMOND/BLAST alignment parsed by Lowest Common Ancestor (LCA) algorithms	NCBI Taxonomy, SEED, eggNOG, InterPro2GO	Yes (interactive co-exploration of taxonomic and functional trees)
eggNOG-mapper	Assembled contigs, predicted proteins, or raw reads	Precomputed hidden Markov model (HMM) profile search and diamond scoring	eggNOG orthology database, SMART, GO, KEGG	No (annotates input sequences directly without species decomposition)
MG-RAST	Shotgun reads or amplicon libraries (web platform)	Automated cloud pipeline executing similarity searches against M5NR	M5NR (GenBank, SEED, KEGG, UniProt, RefSeq)	Yes (integrated web charts linking taxonomy to functional subsystems)

Frequently Asked Questions

What does HUMAnN stand for?

HUMAnN stands for the HMP Unified Metabolic Analysis Network. It was originally established under the National Institutes of Health (NIH) Human Microbiome Project (HMP) to provide standard computational infrastructure for functional metagenomic profiling.

What is the difference between MetaPhlAn and HUMAnN?

MetaPhlAn and HUMAnN are companion tools in the bioBakery suite that address complementary questions. MetaPhlAn performs taxonomic profiling by mapping sequencing reads against unique clade-specific marker genes to determine organism relative abundance. HUMAnN performs functional profiling by mapping reads against pangenomes and protein databases to quantify metabolic pathways and gene families.

What reference databases does HUMAnN require?

HUMAnN requires three primary databases: the ChocoPhlAn nucleotide database (containing species-specific microbial pangenomes), the UniRef protein database (UniRef50 or UniRef90 indexed for DIAMOND), and functional mapping tables that link UniRef IDs to MetaCyc pathways and reactions.

What are the primary output files produced by HUMAnN?

For each input sequencing sample, HUMAnN generates three tab-delimited (TSV) files: genefamilies.tsv (abundance of individual UniRef gene families in Reads Per Kilobase), pathabundance.tsv (abundance of complete metabolic pathways), and pathcoverage.tsv (completeness score of each pathway from 0.0 to 1.0).

Why are abundance values in HUMAnN reported in RPK?

HUMAnN reports raw gene family abundances in RPK (Reads Per Kilobase) to normalize for gene length, because longer genes generate more sequencing reads at equal biological copy number. Researchers use the companion utility humann_renorm_table to further normalize RPK values to Copies Per Million (CPM) or relative abundance to account for differences in sequencing depth across samples.

Sources

• bioBakery. “HUMAnN User Manual.” Huttenhower Lab, Harvard T.H. Chan School of Public Health and Broad Institute, https://github.com/biobakery/humann (November 7, 2026).
• Beghini, F., McIver, L.J., Blanco-Míguez, A., Dubois, L., Asnicar, F., Maharjan, S., et al. “Integrating taxonomic, functional, and strain-level profiling of diverse microbial communities with bioBakery 3.” eLife 2021;10:e65088. DOI: 10.7554/eLife.65088. (November 7, 2026).
• Franzosa, E.A., McIver, L.J., Rahnavard, G., Thompson, L.R., Schirmer, M., Weingart, G., et al. “Species-level functional profiling of metagenomes and metatranscriptomes.” Nature Methods 2018;15(11):962–968. DOI: 10.1038/s41592-018-0176-y. (November 7, 2026).
• Huttenhower Lab. “HUMAnN: HMP Unified Metabolic Analysis Network.” Harvard T.H. Chan School of Public Health, http://huttenhower.sph.harvard.edu/humann (November 7, 2026).