Adaptive Immune Receptor Repertoire analysis agentic tooling

Overview

This skill provides a practical workflow for Adaptive Immune Receptor Repertoire (AIRR / VDJ-seq) data analysis with immunarch (analysis + visualization) and immundata (data ingestion + transformation + schema handling).

Use it to:

• Load AIRR data from one file, many files, glob patterns, or metadata tables.
• Define receptor semantics (chain-agnostic, single-chain, paired-chain).
• Build reproducible immutable pipelines.
• Compute repertoire statistics, clonality, diversity, and public overlap.
• Move annotations between ImmunData and single-cell objects (e.g., Seurat/AnnData).

When to Use This Skill

Use this skill when the user asks to:

• Analyze bulk or single-cell AIRR/TCR/BCR data.
• Compare repertoires across sample groups (tissue, therapy, cluster, donor, timepoint).
• Compute clonality/diversity/publicity metrics.
• Define or change receptor schema (e.g., cdr3_aa + v_call, TRA-only, TRA+TRB).
• Filter receptors by patterns or sequence distance.
• Add/propagate labels between repertoire data and scRNA metadata.
• Convert old immunarch objects to the newer ImmunData pipeline.

Quick Start

Basic import and first look

library(immunarch)

# Demo data + basic grouping
idata <- get_test_idata() |> agg_repertoires("Therapy")
idata

# Core analyses
airr_stats_genes(idata, gene_col = "v_call") |> vis()
airr_public_jaccard(idata) |> vis()
airr_clonality_prop(idata)
airr_diversity_pielou(idata) |> vis()

Optional: add clonality labels to Seurat metadata

idata <- annotate_clonality_prop(idata)
sdata <- annotate_seurat(idata, sdata, cols = "clonal_prop_bin")
Seurat::DimPlot(sdata, reduction = "umap", group.by = "clonal_prop_bin", shuffle = TRUE)

Ingest AIRR files via immundata directly

library(immundata)

md_path <- system.file("extdata/tsv", "metadata.tsv", package = "immundata")
files <- c(
  system.file("extdata/tsv", "sample_0_1k.tsv", package = "immundata"),
  system.file("extdata/tsv", "sample_1k_2k.tsv", package = "immundata")
)
md <- read_metadata(md_path)

idata <- read_repertoires(
  path     = files,
  schema   = c("cdr3_aa", "v_call"),
  metadata = md
)

Typical User Intake (what to ask/assume)

Before coding, identify:

1. Modality: bulk vs single-cell AIRR.

2. Input format: TSV/CSV/Parquet, gzipped or not, one file vs many.

3. Schema intent:

   • chain-agnostic,
   • single-chain (e.g., TRA only),
   • paired-chain (e.g., TRA+TRB, IGH + IGK|IGL).

4. Grouping variables: repertoire schema (sample/cluster/tissue/therapy).

5. Target analyses: stats, gene usage, clonality, diversity, overlap, annotation transfer.

6. Scale/performance: whether snapshots/materialization strategy is needed.

If unknown, default to conservative, reproducible choices and print intermediate summaries.

Standard Analysis Workflow

1) Ingest data and metadata

Use read_metadata() + read_repertoires().

library(immundata)

inp_files <- "path/to/airr/*.tsv.gz"
md_table  <- read_metadata("path/to/metadata.tsv")

idata <- read_repertoires(
  path              = inp_files,
  schema            = c("cdr3_aa", "v_call"),
  metadata          = md_table,
  repertoire_schema = "Sample"
)

2) Choose receptor schema explicitly

Chain-agnostic (bulk / pre-filtered)

idata <- read_repertoires(
  path   = inp_files,
  schema = c("cdr3_aa", "v_call")
)

Single-chain (e.g., TRA only)

schema <- make_receptor_schema(
  features = c("cdr3", "v_call"),
  chains   = "TRA"
)

idata <- read_repertoires(
  path        = "path/to/single_cell.csv.gz",
  schema      = schema,
  barcode_col = "barcode",
  locus_col   = "locus"
)

Paired-chain (e.g., TRA + TRB)

schema <- make_receptor_schema(
  features = c("cdr3", "v_call"),
  chains   = c("TRA", "TRB")
)

idata <- read_repertoires(
  path        = "path/to/single_cell.csv.gz",
  schema      = schema,
  barcode_col = "barcode",
  locus_col   = "locus",
  umi_col     = "umis"
)

Paired with alternative light chain (e.g., IGH + IGK|IGL)

schema <- make_receptor_schema(
  features = c("cdr3", "v_call"),
  chains   = c("IGH", "IGK|IGL")
)

3) Transform immutably (filter / annotate / mutate)

# Filtering
idata_f <- idata |>
  filter(v_call == "TRBV2") |>
  filter(imd_proportion >= 0.0002)

# Sequence-distance filter
idata_seq <- idata |>
  filter(seq_options = make_seq_options(
    patterns  = "CASSELAGYRGEQYF",
    query_col = "cdr3",
    method    = "lev",
    max_dist  = 3
  ))

# Annotation join
idata_ann <- annotate(
  idata       = idata,
  annotations = cells[c("barcode", "ident")],
  by          = c(imd_barcode = "barcode"),
  keep_repertoires = FALSE
)

# Mutations
idata_mut <- idata |>
  mutate(big_chains = umis >= 10) |>
  mutate(dist_to_pattern = dd$levenshtein(cdr3, "CASSSVSGNSPLHF"))

4) Aggregate repertoires for reporting strata

idata_grp <- idata |>
  agg_repertoires("Tissue")

5) Compute immune repertoire statistics

# Chain and gene statistics
chains <- airr_stats_chains(idata_grp)
genes  <- airr_stats_genes(idata_grp, gene_col = "v_call", level = "receptor")

chains |> vis()
genes  |> vis()

6) Clonality analysis

cl_line <- airr_clonality_line(idata_grp)
cl_prop <- airr_clonality_prop(idata_grp)
cl_rank <- airr_clonality_rank(idata_grp, bins = c(10, 100, 1000))

cl_prop |> vis()
cl_rank |> vis()

7) Diversity analysis

d50      <- airr_diversity_dxx(idata_grp, perc = 50)
chao1    <- airr_diversity_chao1(idata_grp)
shannon  <- airr_diversity_shannon(idata_grp)
pielou   <- airr_diversity_pielou(idata_grp)
hill1    <- airr_diversity_index(idata_grp)
hillprof <- airr_diversity_hill(idata_grp, q = c(0, 1, 2))

pielou |> vis()

8) Public overlap

m_pub <- airr_public_intersection(idata_grp)
m_jac <- airr_public_jaccard(idata_grp)

m_pub |> vis()
m_jac |> vis()

9) Snapshot expensive steps

# Save intermediate immutable snapshot to avoid recomputing expensive transforms
idata_cached <- immundata::write_immundata(idata_mut, "path/to/snapshot_folder")

Common Pitfalls and Best Practices

1. Changing receptor definition mid-analysis

   • Receptor schema is foundational. If receptor definition changes, re-read data into a new ImmunData.

2. Using wrong mode for single-cell

   • Paired analysis requires barcode_col, locus_col, and typically umi_col.
   • Single-chain mode does not necessarily collapse multiple chains per barcode.

3. Skipping metadata strategy

   • Prefer metadata-driven ingestion (path = "<metadata>") when file provenance and sample mapping matter.

4. Direct internal-table edits

   • Avoid low-level manual edits. Use high-level verbs (filter, annotate, mutate, agg_repertoires).

5. Ignoring immutable pipeline behavior

   • Each step returns a new object; persist expensive steps with snapshots.

6. Non-canonical columns

   • In scripts, rely on canonical imd_* columns.
   • In package code, prefer schema keys/aliases (e.g., via imd_schema()).

7. Distance-heavy transforms on large data

   • Pattern/Levenshtein operations can be expensive; run once, snapshot, then reuse.

Bundled Resources

TBD

Additional Resources

• immunomind docs home: https://immunomind.github.io/docs/
• Quick Start: https://immunomind.github.io/docs/intro/quick_start/
• Reading repertoire files: https://immunomind.github.io/docs/guides/io/ingesting/
• Reading single-/paired-chain data: https://immunomind.github.io/docs/guides/io/modes/
• Filter: https://immunomind.github.io/docs/guides/transform/filter/
• Annotate: https://immunomind.github.io/docs/guides/transform/annotate/
• Mutate: https://immunomind.github.io/docs/guides/transform/mutate/
• Workflow phase 1 (ingestion): https://immunomind.github.io/docs/concepts/workflow/phase_ingestion/
• Workflow phase 2 (transformation): https://immunomind.github.io/docs/concepts/workflow/phase_transformation/
• ImmunData structure: https://immunomind.github.io/docs/guides/data_schema/
• Single-cell end-to-end tutorial: https://immunomind.github.io/docs/tutorials/single_cell/

Tips for Effective Analysis

• Start with a small subset and verify schema + grouping before scaling up.
• Print object summaries after ingestion and after major transforms.
• Use explicit variable names for stages (idata_raw, idata_qc, idata_ann, idata_stats).
• Prefer pipelines that can be re-executed end-to-end from raw inputs.
• Keep biologically meaningful grouping variables in repertoire schema early.
• Use vis() early and often for sanity checks before formal interpretation.