PICRUSt2

Phylogenetic Investigation of Communities by Reconstruction of Unobserved States

What is PICRUSt2?

PICRUSt2 (Phylogenetic Investigation of Communities by Reconstruction of Unobserved States 2) predicts the functional composition of a metagenome using 16S rRNA amplicon data. It places your 16S sequences on a reference phylogenetic tree and then predicts gene family and pathway abundances based on characterized genomes.

PICRUSt2 answers the question: “What metabolic functions might this community be performing?” — from 16S amplicon data alone.

📄 GitHub
📖 Documentation
🗞️ Paper: Douglas et al. 2020, Nature Biotechnology

When to Use PICRUSt2

Use PICRUSt2 when you have 16S rRNA amplicon data (not shotgun metagenomics) and want functional predictions:

You don’t have shotgun metagenomic data
You want a quick functional annotation from existing 16S surveys
Budget or compute constraints prevent shotgun sequencing

Warning

PICRUSt2 predictions are inferred, not measured. For accurate functional profiling, use HUMAnN with actual shotgun metagenomics data.

Installation

Via conda (recommended)

conda create -n picrust2 -c bioconda -c conda-forge picrust2
conda activate picrust2

Via pip

pip install picrust2

Basic Usage

PICRUSt2 requires as input: 1. A table of representative 16S sequences (FASTA) 2. An ASV/OTU abundance table (biom format)

picrust2_pipeline.py \
  -s seqs.fna \
  -i feature-table.biom \
  -o picrust2_out_pipeline \
  -p 4

Step-by-step workflow

# Step 1: Place sequences on reference tree
place_seqs.py \
  -s seqs.fna \
  -o out.tre \
  -p 4

# Step 2: Predict hidden state (gene family copy numbers)
hsp.py -i 16S -t out.tre -o 16S_predicted.tsv.gz -p 4
hsp.py -i EC -t out.tre -o EC_predicted.tsv.gz -p 4

# Step 3: Compute metagenome abundances
metagenome_pipeline.py \
  -i feature-table.biom \
  -m 16S_predicted.tsv.gz \
  -f EC_predicted.tsv.gz \
  -o EC_metagenome_out

# Step 4: Predict pathway abundances
pathway_pipeline.py \
  -i EC_metagenome_out/pred_metagenome_contrib.tsv.gz \
  -o pathways_out \
  -p 4

Output Files

File	Contents
`EC_predicted.tsv.gz`	Predicted EC (enzyme) copy numbers per ASV
`EC_metagenome_out/pred_metagenome_unstrat.tsv.gz`	Community-level EC abundances
`pathways_out/path_abun_unstrat.tsv.gz`	MetaCyc pathway abundances
`pathways_out/path_abun_strat.tsv.gz`	Pathway abundances stratified by contributing taxon

Tips & Gotchas

Warning

NSTI scores — PICRUSt2 reports the Nearest Sequenced Taxon Index (NSTI) for each ASV. High NSTI values (>2) indicate your sequences are phylogenetically distant from reference genomes, making predictions less reliable. Filter out high-NSTI ASVs for better results.

Tip

Use DADA2 or DEBLUR output as input to PICRUSt2 — exact sequence variants work better than OTUs clustered at 97% similarity.

Tip

Add descriptions to output files using add_descriptions.py:

add_descriptions.py -i pathways_out/path_abun_unstrat.tsv.gz -m METACYC -o path_abun_unstrat_descrip.tsv.gz

--- title: "PICRUSt2" subtitle: "Phylogenetic Investigation of Communities by Reconstruction of Unobserved States" --- ## What is PICRUSt2? **PICRUSt2** (Phylogenetic Investigation of Communities by Reconstruction of Unobserved States 2) predicts the functional composition of a metagenome using 16S rRNA amplicon data. It places your 16S sequences on a reference phylogenetic tree and then predicts gene family and pathway abundances based on characterized genomes. PICRUSt2 answers the question: **"What metabolic functions might this community be performing?"** — from 16S amplicon data alone. - 📄 [GitHub](https://github.com/picrust/picrust2) - 📖 [Documentation](https://github.com/picrust/picrust2/wiki) - 🗞️ [Paper: Douglas et al. 2020, *Nature Biotechnology*](https://www.nature.com/articles/s41587-020-0548-6) --- ## When to Use PICRUSt2 Use PICRUSt2 when you have **16S rRNA amplicon data** (not shotgun metagenomics) and want functional predictions: - You don't have shotgun metagenomic data - You want a quick functional annotation from existing 16S surveys - Budget or compute constraints prevent shotgun sequencing ::: {.callout-warning} PICRUSt2 predictions are **inferred, not measured**. For accurate functional profiling, use [HUMAnN](humann.qmd) with actual shotgun metagenomics data. ::: --- ## Installation ### Via conda (recommended) ```bash conda create -n picrust2 -c bioconda -c conda-forge picrust2 conda activate picrust2 ``` ### Via pip ```bash pip install picrust2 ``` --- ## Basic Usage PICRUSt2 requires as input: 1. A table of representative 16S sequences (FASTA) 2. An ASV/OTU abundance table (biom format) ```bash picrust2_pipeline.py \ -s seqs.fna \ -i feature-table.biom \ -o picrust2_out_pipeline \ -p 4 ``` ### Step-by-step workflow ```bash # Step 1: Place sequences on reference tree place_seqs.py \ -s seqs.fna \ -o out.tre \ -p 4 # Step 2: Predict hidden state (gene family copy numbers) hsp.py -i 16S -t out.tre -o 16S_predicted.tsv.gz -p 4 hsp.py -i EC -t out.tre -o EC_predicted.tsv.gz -p 4 # Step 3: Compute metagenome abundances metagenome_pipeline.py \ -i feature-table.biom \ -m 16S_predicted.tsv.gz \ -f EC_predicted.tsv.gz \ -o EC_metagenome_out # Step 4: Predict pathway abundances pathway_pipeline.py \ -i EC_metagenome_out/pred_metagenome_contrib.tsv.gz \ -o pathways_out \ -p 4 ``` --- ## Output Files | File | Contents | |------|----------| | `EC_predicted.tsv.gz` | Predicted EC (enzyme) copy numbers per ASV | | `EC_metagenome_out/pred_metagenome_unstrat.tsv.gz` | Community-level EC abundances | | `pathways_out/path_abun_unstrat.tsv.gz` | MetaCyc pathway abundances | | `pathways_out/path_abun_strat.tsv.gz` | Pathway abundances stratified by contributing taxon | --- ## Tips & Gotchas ::: {.callout-warning} **NSTI scores** — PICRUSt2 reports the Nearest Sequenced Taxon Index (NSTI) for each ASV. High NSTI values (>2) indicate your sequences are phylogenetically distant from reference genomes, making predictions less reliable. Filter out high-NSTI ASVs for better results. ::: ::: {.callout-tip} **Use DADA2 or DEBLUR** output as input to PICRUSt2 — exact sequence variants work better than OTUs clustered at 97% similarity. ::: ::: {.callout-tip} **Add descriptions** to output files using `add_descriptions.py`: ```bash add_descriptions.py -i pathways_out/path_abun_unstrat.tsv.gz -m METACYC -o path_abun_unstrat_descrip.tsv.gz ``` ::: --- ## Further Reading - [PICRUSt2 tutorial](https://github.com/picrust/picrust2/wiki/PICRUSt2-Tutorial-(v2.5.2)) - [Douglas et al. 2020, *Nature Biotechnology*](https://www.nature.com/articles/s41587-020-0548-6)