nf-core/airrflow

What nf-core/airrflow is

nf-core/airrflow is a bioinformatics pipeline for analysing B-cell receptor (BCR) and T-cell receptor (TCR) adaptive immune receptor repertoire sequencing (AIRR-seq) data. Adaptive immune receptors are proteins on the surface of B and T lymphocytes responsible for recognising antigens. The pipeline evaluates targeted bulk and single-cell sequencing libraries, and can also recover BCR and TCR sequences from untargeted bulk and single-cell RNA sequencing datasets.

Originally developed by Gisela Gabernet, Susanna Marquez, and Alexander Peltzer, the workflow provides an end-to-end framework starting from either raw FASTQ reads or previously assembled sequences. It standardises quality control, sequence assembly, V(D)J germline gene segment assignment, quality filtering, clonal inference, phylogenetic lineage reconstruction, and sequence representation learning using the Immcantation software suite alongside other immune repertoire tools.

Certified as AIRR compliant by the AIRR Community, the pipeline outputs data in standardised formats compatible with community-developed downstream analytical tools. It is constructed in Nextflow DSL2 and runs inside Docker or Apptainer and Singularity containers to deliver reproducible computational workflows across varying infrastructures.

What the pipeline does

Input processing depends on the source assay. For bulk sequencing, Fastp performs adapter trimming, after which the pRESTO toolkit filters reads by base quality, masks primer sequences, pairs read mates, clusters reads, constructs consensus sequences for unique molecular identifier (UMI) libraries, and assembles paired-end reads. For single-cell data, Cell Ranger VDJ assembles contigs, annotates sequences, calls cells, and defines clonotypes. Alternatively, users can supply pre-assembled FASTA files or AIRR rearrangement tables. IgBLAST and Change-O then assign V(D)J gene segments using germline references, discarding alignments with poor quality, uninformative base lengths, non-functional frame shifts, or non-triplet junction lengths.

Following initial assignment, EnchantR and SHazaM execute quality-filtering steps to eliminate chimeric reads, detect cross-contamination, and collapse duplicates. Single-cell workflows remove cell barcodes missing heavy chains, containing multiple heavy chains, or displaying cross-sample contamination. If enabled for BCR data, TIgGER detects novel germline alleles and infers sample-specific genotypes. Clonal thresholds and definitions are subsequently established using SHazaM and SCOPer, while Dowser, IgphyML, and RAxML reconstruct clonal lineage trees. Finally, AMULETY generates sequence embeddings with pre-trained models, and QC and repertoire summary reports are compiled using Alakazam and MultiQC.

Key facts

What you can use it for

Bulk AIRR-seq analysis

Assemble raw paired-end bulk BCR or TCR amplicons with or without UMI barcodes, perform primer masking, assign V(D)J genes, and quantify clone frequency.

Single-cell immune repertoire profiling

Process 10x Genomics single-cell immune profiling FASTQ reads to assemble contigs, call cells, remove aberrant multi-heavy-chain barcodes, and infer clonotypes.

Novel BCR allele and genotype inference

Detect previously uncharacterised V-gene alleles and infer donor-specific genomic BCR genotypes from repertoire data using TIgGER.

B-cell phylogenetic lineage reconstruction

Define clonal groups, construct inferred germline sequences, and generate phylogenetic lineage trees using SCOPer, Dowser, IgphyML, and RAxML.

Receptor sequence embedding

Translate nucleotide sequences to amino acids and extract feature representations using pre-trained AMULETY models for machine learning analyses.

Run it here

The research agent runs nf-core/airrflow with Nextflow on its own machine — up to 32 cores, 256 GiB of memory and 24 hours per run, which covers the test profile and real single-sample runs. It builds the samplesheet, checks it against the pipeline's schema and reads the results back. When a run is too big for one machine, it gives you the exact command for your own cluster instead.

Strengths and limits

Questions

How to cite it

Gisela Gabernet, Susanna Marquez, Robert Bjornson, Alexander Peltzer, Hailong Meng, Edel Aron, Noah Y. Lee, Cole G. Jensen, David Ladd, Mark Polster, Friederike Hanssen, Simon Heumos, nf-core community, Gur Yaari, Markus C. Kowarik, Sven Nahnsen, Steven H. Kleinstein. (2024) PLOS Computational Biology, 20(7), e1012265. doi: https://doi.org/10.1371/journal.pcbi.1012265. Pubmed PMID: 39058741.