Functional Prediction with PICRUSt2

Functional Prediction with PICRUSt2

Prepare Input Files

library(phyloseq)
library(Biostrings)

ps <- readRDS('phyloseq_object.rds')

# Export ASV table (samples as columns)
otu <- as.data.frame(otu_table(ps))
if (!taxa_are_rows(ps)) otu <- t(otu)
write.table(otu, 'asv_table.tsv', sep = '\t', quote = FALSE)

# Export ASV sequences as FASTA
seqs <- refseq(ps)  # Or extract from ASV names if stored there
writeXStringSet(seqs, 'asv_seqs.fasta')

Run PICRUSt2 Pipeline

# Full pipeline (place sequences, predict functions, metagenome inference)
picrust2_pipeline.py \
    -s asv_seqs.fasta \
    -i asv_table.tsv \
    -o picrust2_output \
    -p 4 \
    --stratified \
    --per_sequence_contrib

# Output files:
# - pathway_abundance.tsv (MetaCyc pathways)
# - KO_metagenome_out/​pred_metagenome_unstrat.tsv (KEGG orthologs)
# - EC_metagenome_out/​pred_metagenome_unstrat.tsv (EC numbers)

Step-by-Step Pipeline

# 1. Place sequences in reference tree
place_seqs.py -s asv_seqs.fasta -o placed_seqs.tre -p 4

# 2. Hidden state prediction (gene content)
hsp.py -i 16S -t placed_seqs.tre -o marker_nsti_predicted.tsv -m pic -n

# 3. Predict gene families (KO)
hsp.py -i KO -t placed_seqs.tre -o KO_predicted.tsv -m pic

# 4. Metagenome inference
metagenome_pipeline.py \
    -i asv_table.tsv \
    -m marker_nsti_predicted.tsv \
    -f KO_predicted.tsv \
    -o KO_metagenome_out \
    --strat_out

# 5. Pathway inference
pathway_pipeline.py \
    -i KO_metagenome_out/​pred_metagenome_contrib.tsv \
    -o pathway_output \
    -p 4

Quality Control: NSTI

import pandas as pd

# NSTI = Nearest Sequenced Taxon Index
# Lower = more reliable prediction (< 2 is acceptable)
nsti = pd.read_csv('marker_nsti_predicted.tsv', sep='\t')
print(f'Mean NSTI: {nsti["metadata_NSTI"].mean():.3f}')
print(f'ASVs with NSTI > 2: {(nsti["metadata_NSTI"] > 2).sum()}')

Analyze Pathway Output

library(ggplot2)

pathways <- read.delim('picrust2_output/​pathways_out/​path_abun_unstrat.tsv', row.names = 1)
metadata <- read.csv('sample_metadata.csv', row.names = 1)

# Normalize to relative abundance
pathways_rel <- sweep(pathways, 2, colSums(pathways), '/')

# Differential pathway analysis (use ALDEx2 or similar)
library(ALDEx2)
groups <- metadata[colnames(pathways), 'Group']
pathway_aldex <- aldex(as.data.frame(t(pathways)), groups, mc.samples = 128)

Add Pathway Descriptions

# Map pathway IDs to names
add_descriptions.py \
    -i pathway_abundance.tsv \
    -m METACYC \
    -o pathway_abundance_described.tsv

KEGG Module Analysis

# Analyze KEGG modules instead of individual KOs
ko_table <- read.delim('KO_metagenome_out/​pred_metagenome_unstrat.tsv', row.names = 1)

# Use KEGGREST for module mapping
library(KEGGREST)
modules <- keggLink('module', 'ko')

Limitations

• Predictions based on phylogenetic placement
• Novel taxa (high NSTI) have unreliable predictions
• 16S resolution limits species-level accuracy
• Cannot detect horizontal gene transfer events

Related Skills

• amplicon-processing - Generate ASV input
• metagenomics/​functional-profiling - Direct shotgun-based profiling
• pathway-analysis/​kegg-pathways - KEGG pathway enrichment