Shotgun metagenomic sequencing produces DNA reads from a sample, which may include both microbial and host DNA. The reads alone do not show which metabolic processes were active. Researchers need computational methods to connect those sequences with known or predicted genes, proteins and biochemical pathways. Functional profiling is the process of identifying and estimating the representation of microbial gene families and biological pathways from sequencing data. HUMAnN (HMP Unified Metabolic Analysis Network) is one established tool for performing this analysis.

For example, a gut microbiome may contain genes associated with carbohydrate degradation, amino acid metabolism or short-chain fatty acid production. Functional profiling can investigate the genetic potential for these processes, but it cannot independently establish whether the corresponding reactions occurred. This guide walks through a HUMAnN 3 analysis, from preparing reads to checking the output tables.

Figure 1. How HUMAnN goes from shotgun metagenomic reads to estimates of gene-family and metabolic pathway abundances. AI-assisted conceptual illustration.

What Does Functional Profiling Measure?

Functional profiling connects microbial DNA sequences to reference information about genes and biochemical functions. At the gene-family level, related sequences are grouped using a reference classification system. At the pathway level, the predicted gene functions are compared with collections of biochemical reactions to evaluate whether the sequencing evidence supports particular metabolic pathways.

These are related but distinct analytical levels. A gene family may contain proteins with similar sequences, while a pathway may require several different enzymatic reactions. For instance, a sample might contain genes encoding enzymes involved in carbohydrate metabolism. This suggests that the microbial community possesses some relevant genetic machinery. It does not establish that those enzymes were expressed, that the pathway was complete or that the expected metabolites were produced. Functional profiling is therefore most accurately described as an investigation of microbial genetic potential.

Taxonomic Profiling Versus Functional Profiling

Taxonomic profiling estimates which organisms are represented in sequencing data. Functional profiling investigates which gene families and pathways are represented. Two microbial communities may differ in taxonomic composition yet share many functional capabilities. Conversely, related organisms can possess different metabolic genes because of strain-level variation, gene gain or gene loss.

Some functional profiling tools also provide taxonomically stratified results. These associate predicted gene families or pathways with particular organisms represented in the reference framework. Such assignments are useful, but they remain computational estimates. Species-level functional attribution depends on the sequence matches and the available references. In contrast to shotgun sequencing, 16S rRNA amplicon data do not directly sample metabolic genes; predictions from these data rely on reference genomes and can miss functions that vary among related microbes.

What Is HUMAnN and How Does It Work?

HUMAnN (official project and documentation) is part of the bioBakery software ecosystem for microbiome analysis. It uses sequence matching and reference-based reconstruction to estimate microbial gene-family and pathway abundances. The established HUMAnN2/HUMAnN3 workflow uses a tiered search strategy. Taxonomic information helps identify likely reference organisms, after which reads are searched against corresponding nucleotide sequences. Reads not resolved at this stage can be examined using translated protein searches.

HUMAnN summarizes read matches into gene-family abundances, then uses them to reconstruct pathways. This read-based workflow differs from genome assembly and annotation, which reconstructs longer sequences before identifying and annotating genes. This strategy reduces the need to search every read against every available reference sequence in the same way. It also allows the analysis to recover some functional information from sequences that do not match the initial organism-specific nucleotide references. HUMAnN has evolved across software releases. Readers should use the documentation and databases corresponding to their installed version rather than assuming that commands, reference collections and outputs are identical across all releases.

The Three Important Reference Resources

ChocoPhlAn: Nucleotide reference sequences

ChocoPhlAn (HUMAnN database instructions) provides organism-associated nucleotide reference information used in the pangenome-oriented search stage of established HUMAnN workflows. This stage helps identify gene families using sequences associated with organisms detected in the microbial community. The effectiveness of this approach depends partly on the representation of those organisms in the nucleotide reference collection.

UniRef: Protein sequence clusters

UniRef groups related protein sequences into clusters at specified sequence-similarity thresholds. HUMAnN uses compatible UniRef-based reference information to support functional identification, including translated searches for reads that remain unresolved after nucleotide matching. UniRef identifiers represent protein sequence clusters. They should not automatically be treated as experimentally verified enzyme functions.

MetaCyc: Metabolic reactions and pathways

MetaCyc contains curated information about metabolic reactions, enzymes and pathways. HUMAnN uses pathway definitions to evaluate whether the detected gene-family evidence supports particular metabolic processes. A pathway documented in MetaCyc may be experimentally characterized in certain organisms. Its computational prediction in another microbial community does not establish experimental confirmation in that community.

Practical HUMAnN Workflow: From FASTQ Reads to Functional Tables

These command templates follow HUMAnN 3 and assume a Unix-like shell, a working installation and compatible databases. Check the official manual for setup instructions for your release.

Step 1. Prepare sequencing data

A typical Illumina paired-end sample has two FASTQ files, such as sample_R1.fastq.gz and sample_R2.fastq.gz. Both contain sequencing reads and quality scores. Before functional profiling, researchers should evaluate sequencing quality, adapter contamination and the presence of unwanted host-derived DNA. Read quality control can be performed with tools such as fastp. Host-read filtering may be appropriate for human or animal samples.

For the single-file input shown below, concatenate the quality-controlled R1 and R2 reads from the same sample into one FASTQ. Concatenation keeps the reads from both files without merging overlapping mates or making HUMAnN use their pairing information.

Step 2. Check the software installation

After installing a compatible HUMAnN release using the official installation instructions, inspect the available command-line options:

humann --version
humann --help

These commands allow readers to confirm that HUMAnN is accessible and inspect the options supported by their installation. A typical installation also requires compatible nucleotide and protein reference databases, plus the taxonomic profiling dependencies specified for that release.

Step 3. Check database configuration

The HUMAnN utility can display the configured database locations:

humann_config --print

The database-management utility provides download and configuration options (see the official database instructions):

humann_databases --help

Before downloading large reference collections, readers should check disk-space requirements, available memory, software compatibility and institutional computing policies. Database versions should be recorded because they can influence which microbial genes and pathways are identified.

Step 4. Run functional profiling

Once the reads and databases are prepared, a basic HUMAnN command has the following structure:

humann \
  --input sample.fastq.gz \
  --output humann_output \
  --threads 8 \
  --nucleotide-database /path/to/chocophlan \
  --protein-database /path/to/uniref

Here, sample.fastq.gz represents a correctly prepared input file. The two database paths are placeholders, not real installed directories. The --threads option specifies the requested number of computational threads. Actual runtime and memory requirements depend on the dataset, databases and computing environment.

HUMAnN may invoke a compatible taxonomic profiling component during the workflow. The relevant taxonomic database and software versions must also be available and compatible. This command illustrates the structure of the analysis. It should not be copied unchanged without replacing the input and database paths.

Step 5. Examine the output files

For an input named sample.fastq.gz, the HUMAnN 3 output convention includes these files:

humann_output/
    sample_genefamilies.tsv
    sample_pathabundance.tsv
    sample_pathcoverage.tsv
    sample_humann_temp/
        sample.log

The three TSVs are the principal biological output tables. In a default HUMAnN 3.9 run, the log is in the sample’s temporary output folder and can help identify processing problems. The exact files and naming conventions should be checked against the installed version. For example, the HUMAnN 4 documentation uses sample_2_genefamilies.tsv for gene families and sample_4_pathabundance.tsv for pathway abundance. See official HUMAnN 4 output examples.

Understanding Gene-Family Abundance

The gene-family table contains reference-associated gene-family identifiers and estimated abundances. HUMAnN commonly reports gene-family abundance in reads per kilobase (RPK), which adjusts for reference sequence length. RPK does not normalize for sequencing depth across samples. A simplified, hypothetical table might look like this:

Gene Family          sample_1
UNMAPPED             120.0
UniRef90_EXAMPLE_A    45.0
UniRef90_EXAMPLE_B    18.0

These identifiers and values are illustrative, not actual HUMAnN results. UNMAPPED represents sequence evidence that was not successfully assigned to the relevant reference gene families. It should not be interpreted as proof that those sequences lack biological functions. Some gene-family outputs contain both community-level entries and taxonomically stratified entries, allowing users to examine estimated contributions from particular microbial groups.

Understanding Pathway Abundance and Coverage

Pathway abundance and pathway coverage answer different questions. Pathway abundance estimates the representation of a pathway using evidence from the gene families associated with its component reactions. Pathway coverage evaluates the extent to which the observed functional evidence supports the pathway under the software's reconstruction model.

A high pathway-coverage score reflects support under the model; it is not the percentage of the pathway experimentally confirmed and does not show that its enzymes were active. Similarly, a pathway-abundance value is not a direct measurement of metabolic flux. Another important distinction concerns UNINTEGRATED, which is used in pathway-abundance outputs for gene-family evidence that is not incorporated into the reconstructed pathway assignments.

This differs from UNMAPPED, which concerns reads left unaligned. In the pathway-abundance table, both special rows are scaled to pathway-abundance units; neither row is a raw read count. These categories help researchers understand the limitations of reference-based reconstruction.

Normalizing HUMAnN Abundance Tables

HUMAnN provides utilities for converting abundance tables into alternative normalization units. Run the following examples from the project directory containing humann_output:

humann_renorm_table \
  --input humann_output/sample_genefamilies.tsv \
  --output sample_genefamilies_relab.tsv \
  --units relab

The relab option produces relative-abundance values under the selected normalization procedure. Another supported option is:

humann_renorm_table \
  --input humann_output/sample_genefamilies.tsv \
  --output sample_genefamilies_cpm.tsv \
  --units cpm

Here, cpm refers to HUMAnN's copies per million, a total-sum-scaled, TPM-like measure (not raw sequencing counts per million). It must not be confused with experimentally measured gene copies per million microbial cells. The denominator and treatment of special categories such as unmapped or unintegrated evidence can affect interpretation. Researchers should document the selected normalization settings.

Combining Functional Profiles Across Samples

Microbiome studies usually contain multiple samples. HUMAnN provides a utility for combining individual output tables into a multi-sample matrix. For example:

humann_join_tables \
  --input humann_output \
  --output combined_genefamilies.tsv \
  --file_name genefamilies

This example joins raw gene-family tables. Keep one table per sample in the input directory, all at the same feature level and in the same units. The --file_name filter matches any filename containing the given text, so genefamilies also matches genefamilies_relab. Keep normalized files in a separate directory and, when joining them, use the corresponding filter, such as genefamilies_relab. After joining, the table can be imported into R or another statistical environment for visualization and analysis.

A Short Practice Exercise Using Official Example Data

Before using a large study dataset, open the HUMAnN manual and find its demonstration or test-data instructions for the version you installed. Run the documented example, then identify the gene-family, pathway-abundance and pathway-coverage tables. Record the HUMAnN version, database versions and the percentage of reads still unaligned after the final search, as reported in the run log. A normalized UNMAPPED value is not a read percentage. Next, compare the community-level pathway entries with any species-stratified entries. These may not sum to identical values because community reconstruction can combine reaction evidence across taxa. No dataset or results are supplied by this article, and the HUMAnN templates have not been run on a dataset for this article.

A Simple R Exercise: Inspecting a Functional Table

A useful first exercise is to import a HUMAnN table and examine its structure before performing any statistical analysis.

functional <- read.delim(
  "combined_genefamilies.tsv",
  check.names = FALSE,
  comment.char = "",
  quote = ""
)

dim(functional)
head(functional)
colnames(functional)

The output will depend on the actual table. The first column typically contains feature identifiers, while the remaining columns represent samples. Researchers should check for special rows, taxonomically stratified features, missing values and the normalization units before calculating summaries.

A simple visualization can then be created from a properly selected subset of comparable features. However, community-level and stratified features should not be summed together because doing so may double-count the same functional evidence. For more advanced comparisons, the statistical method should account for compositionality, study design and repeated measurements where relevant. For a first exercise, examine a single community-level feature rather than adding stratified rows to its unstratified parent. If you are new to R, see Getting Started with R and RStudio for Microbiome Analysis.

Interpreting Functional Results in a Microbiome Study

Consider a hypothetical study comparing the gut microbiomes of dogs receiving different diets. After sequencing and functional profiling, researchers observe differences in the estimated relative abundance of gene families associated with carbohydrate metabolism.

This may suggest differences in microbial genetic potential between the dietary groups. However, the result does not independently establish that one group produced more short-chain fatty acids, absorbed more energy or experienced better health outcomes.

Those conclusions require additional evidence, such as appropriately validated metabolite measurements and a suitable experimental design. If the study includes repeated sampling from individual animals, the statistical analysis should also account for within-animal dependence.

Common Problems and Troubleshooting

A large proportion of reads remains unmapped. This may reflect limited database coverage, low-quality sequence data, non-target DNA or organisms poorly represented in the reference collections. The appropriate response is to examine quality-control and mapping reports rather than assume the reads are biologically uninformative.

Few pathways are reconstructed. This may arise from insufficient sequence evidence, incomplete reference annotations or limitations of the pathway reconstruction model. It does not necessarily mean the microbial community lacks metabolic capabilities.

Different HUMAnN versions produce different results. Changes in reference databases, software dependencies, algorithms or parameters can affect output. Version and database documentation should be preserved.

Relative abundance is mistaken for biological activity. A predicted increase in a pathway does not demonstrate increased RNA expression, enzyme activity or metabolite production.

Results cannot be reproduced. Missing software versions, database releases, commands and preprocessing details are common barriers. A complete analysis record should include these items.

Where Can Readers Practise Functional Profiling?

The official HUMAnN repository and detailed manual provide software documentation, demonstration resources and command examples. Readers should begin with its supported example data rather than immediately processing a large metagenomic dataset. The bioBakery project also provides information about the wider software ecosystem. Readers using Finnish research infrastructure can consult the CSC HUMAnN guide, noting that its instructions are specific to CSC computing environments.

For pathway interpretation, MetaCyc can be used to explore reaction and pathway descriptions. UniProt's UniRef documentation explains the protein clustering system; that site uses JavaScript, so it may not display in text-only browsers. These resources allow readers to connect the computational workflow with the underlying biological references.

What Should Researchers Report?

A reproducible functional profiling analysis should document the sequencing platform, sample preprocessing, host-read filtering, HUMAnN version, reference database releases, taxonomic profiling dependencies and command-line parameters. The report should identify whether results were analyzed at the gene-family or pathway level, whether taxonomically stratified entries were included and which normalization method was used.

Researchers should also explain how unmapped and unintegrated evidence was handled and describe the statistical approach used for comparisons. Where possible, scripts and computational environment information should be preserved.

Conclusion

Functional profiling uses shotgun metagenomic sequencing data to investigate the genes and metabolic pathways represented in microbial communities. HUMAnN provides a practical framework for connecting sequencing reads with reference gene families and curated pathway definitions. Its output tables can support comparisons across samples and biological conditions.

The value of these results depends on understanding the computational workflow, reference databases, normalization procedures and limits of sequence-based inference. A predicted microbial function indicates genetic potential. Combining functional profiles with appropriate experimental and molecular measurements provides a stronger basis for interpreting microbial metabolism.

References

1. Beghini, F., McIver, L. J., Blanco-Míguez, A., Dubois, L., Asnicar, F., Maharjan, S., Mailyan, A., Manghi, P., Scholz, M., Thomas, A. M., Valles-Colomer, M., Weingart, G., Zhang, Y., Zolfo, M., Huttenhower, C., Franzosa, E. A., & Segata, N. (2021). Integrating taxonomic, functional, and strain-level profiling of diverse microbial communities with bioBakery 3. eLife, 10, Article e65088. https://doi.org/10.7554/eLife.65088
2. Caspi, R., Billington, R., Fulcher, C. A., Keseler, I. M., Kothari, A., Krummenacker, M., Latendresse, M., Midford, P. E., Ong, Q., Ong, W. K., Paley, S., Subhraveti, P., & Karp, P. D. (2018). The MetaCyc database of metabolic pathways and enzymes. Nucleic Acids Research, 46(D1), D633–D639. https://doi.org/10.1093/nar/gkx935
3. Franzosa, E. A., McIver, L. J., Rahnavard, G., Thompson, L. R., Schirmer, M., Weingart, G., Lipson, K. S., Knight, R., Caporaso, J. G., Segata, N., & Huttenhower, C. (2018). Species-level functional profiling of metagenomes and metatranscriptomes. Nature Methods, 15(11), 962–968. https://doi.org/10.1038/s41592-018-0176-y
4. Matchado, M. S., Rühlemann, M., Reitmeier, S., Kacprowski, T., Frost, F., Haller, D., Baumbach, J., & List, M. (2024). On the limits of 16S rRNA gene-based metagenome prediction and functional profiling. Microbial Genomics, 10(2), Article 001203. https://doi.org/10.1099/mgen.0.001203