TBAS_pipeline

We present Trio-barcoded ONT Adaptive Sampling (TBAS), a cost-efficient long-read sequencing strategy combining sample barcoding and adaptive enrichment to sequence rare-disease trios on a single PromethION/P2 flow cell. TBAS achieved near-complete variant phasing and detection of small variants, structural variants, and tandem repeats with high accuracy and 77% potential solve rate. This scalable approach retains methylation data and enables clinically relevant, phenotype-guided long-read diagnostics at a fraction of current costs.

Overview

• The software pipeline entrypoint is tbas-pipeline, implemented in tbas_pipeline/.
• The software version runs tools directly (no sbatch generation).
• The notebook TBAS_pipeline_slurm.ipynb is kept as-is for historical/reference workflow.
• The analysis/ folder contains downstream analysis notebooks organized by topic (SNV/Trio calling, coverage, methylation, tandem repeats, and variant counting).
• Benchmarking materials and example HG002 results are provided under benchmarking/, with data available at the Zenodo record below.

Analytical novelty

TBAS is not only a cheaper way to sequence trios — its analytical approach is co-designed with long-read sequencing (LRS) to extract signal that short-read trio pipelines cannot. Three points are novel:

1. Automated trio variant ranking (built into the pipeline). Small and structural variants are called per trio member and genotyped trio-aware (kanpig trio), then annotated (ANNOVAR for SNVs, AnnotSV for SVs) and ranked by the rank_snv/rank_sv stages: a four-tier SNV system that combines allele frequency, REVEL/CADD, ClinVar, and the trio inheritance pattern (de novo, recessive, compound-heterozygous), plus pathogenic SV filtering. Because reads are natively phased on a single flow cell, compound-heterozygous and parent-of-origin calls are made directly instead of inferred. See METHODS.md for the exact databases and tiering rules.

2. DNA methylation from LRS as an independent marker of pathogenic expansions — including from adaptive-sampling rejected reads. TBAS preserves native 5mC/5hmC modification tags end to end (methylation tags are carried through read extraction and emitted as phased and unphased pileups by modkit). Locus hypermethylation is then used as an orthogonal signal for repeat-expansion disorders even where the expansion is not fully spanned. To our knowledge this is the first approach to combine LRS methylation with adaptive sampling to flag pathogenic expansions such as NAXE and DIP2B, notably by recovering methylation signal from reads that adaptive sampling rejected (unblocked/off-target) rather than discarding them (worked examples in analysis/methylation/).

3. Trio-aware tandem-repeat length comparison. Long reads span full repeat alleles, so medaka tandem resolves per-sample allele lengths and tdb collates the proband and both parents into one database. This lets the pipeline compare repeat length across the trio directly and flag a proband allele that is expanded relative to the parents (de novo expansion or further somatic/germline expansion) — a comparison prior tandem-repeat pipelines, built for single samples, did not make. See the patho_ATN1_FGF14 example (ATN1/DRPLA CAG, FGF14/SCA27B GAA).

Together these turn TBAS's LRS strengths — native methylation, read phasing, full-length repeat spanning, and the trio structure itself — into diagnostic evidence, at a fraction of the cost of running trios on separate flow cells.

Requirements

• Python 3.10+.
• Standard long-read analysis command-line tools used by the pipeline: samtools, minimap2, sniffles, mosdepth, run_clair3.sh, kanpig, bedtools, bgzip, bcftools, whatshap, medaka, modkit, tdb.

The easiest way to get all of these is the bundled conda environment (see below), which installs the whole toolchain in one command.

Getting started

1. Install the external tools and the package into a single self-contained environment. With micromamba (or swap micromamba for mamba/conda):

micromamba env create -f environment.yml
micromamba activate tbas
pip install -e .

This one environment provides the entire toolchain (samtools, minimap2, sniffles, mosdepth, Clair3, kanpig, bedtools, bcftools, whatshap, medaka, modkit, tdb) — no separate per-tool envs or from-source builds. tdb installs from GitHub via pip as declared in environment.yml. For a byte-for-byte reproducible environment, create from the pinned environment.lock.yml instead.

Or, if the command-line tools are already on your PATH, just:

pip install -e .

2. Verify the toolchain is complete before running:

tbas-pipeline --manifest example_data/test_subset_chr22/manifest_example.csv --check-deps

This lists the tools required by the selected stages and reports any that are missing from PATH. A real run also performs this check automatically and aborts early if a tool is missing (pass --skip-dep-check to override).

3. Create a manifest CSV/TSV with at least:

• sample_id (example: 4_6_Gregor_Trio)
• bed_file (example: Epilepsy, CMRG, WGS, or a direct BED path)
• proband_gender (used by medaka stages, example: female)

Optional columns:

• calls_bam (explicit path to calls*.bam; if omitted, the pipeline searches under <output_folder>/<sample_id>/calls*.bam)
• read_group_prefix (skip BAM-header inference during demultiplex stage)
• tr_bed_file (TR catalog BED for medaka_local; if omitted, the pipeline derives this from bed_file via built-in mapping)

4. Run a dry run starting from demultiplexing:

tbas-pipeline \
  --manifest example_data/test_subset_chr22/manifest_example.csv \
  --output-folder demo_output \
  --stages demultiplex \
  --dry-run

5. Run full pipeline:

tbas-pipeline \
  --manifest example_data/test_subset_chr22/manifest_example.csv \
  --output-folder demo_output

Example data

Two real-data bundles under example_data/ (subsets of the 4_6_Gregor_Trio sample) let you smoke-test the pipeline end to end:

• example_data/test_subset_chr22/ — CMRG genes on chr22:42.6-42.74 Mb. Exercises stages 1-16 fully; medaka_patho is skipped because this window contains no known pathogenic STR loci.
• example_data/patho_ATN1_FGF14/ — windows around ATN1 (DRPLA, CAG) and FGF14 (SCA27B, GAA). These loci are in the pathogenic catalog, so this bundle also exercises medaka_patho and tdb on real disease repeats.

Both runs need a GRCh38 reference, a Clair3 model directory, and the pathogenic TR BED, passed via --reference, --clair3-model, and --adotto-pheno.

Pipeline stages

Default stage order:

1. demultiplex
2. fastq_extract
3. minimap2
4. bam_sort
5. sniffles_global
6. bam_mosdepth
7. clair3_local
8. kanpig_pileup
9. kanpig_gt
10. kanpig_trio
11. whatshap_single_sample_local_phasing
12. whatshap_haplotag
13. medaka_local
14. medaka_patho
15. modkit
16. modkit_nohp
17. tdb
18. merge_trio_snv — merge the trio's small-variant VCFs (proband first)
19. annovar — annotate small variants (ANNOVAR)
20. rank_snv — tier and rank SNVs → <sample>.snv.ranked.tsv
21. annotsv — annotate structural variants (AnnotSV)
22. rank_sv — keep (likely) pathogenic SVs → <sample>_<barcode>.sv.pathogenic.tsv
23. rank_tr — trio tandem-repeat comparison → <sample>.tr.trio_distinct.tsv

You can run a subset with --stages stage1,stage2,....

The annotation/ranking stages (18–23) turn raw calls into ranked candidates: an ANNOVAR-based SNV tiering system, AnnotSV pathogenic SV filtering, and a STRchive-restricted trio tandem-repeat comparison that flags proband repeat alleles distinct from both parents. They need ANNOVAR (table_annovar.pl + a humandb directory), AnnotSV, and a STRchive BED — configured via --annovar-humandb, --annotsv-build, and --strchive-bed. See METHODS.md for the exact databases, tiering rules, and filtering criteria.

Notebook reference

• TBAS_pipeline_slurm.ipynb is preserved and unchanged.
• Use it as reference for the original SLURM-oriented workflow.

Downstream analysis notebooks

The analysis/ directory contains topic-focused notebooks and resources:

• SNV/Trio calling

  • analysis/snv_trio_analysis.ipynb: downstream analysis notebook for small variants and trio evaluation.

• Coverage

  • analysis/covearge/mosdepth_analysis.ipynb: coverage analysis with mosdepth.
  • analysis/covearge/mosdepth_summary_totals.csv: example summary output.

• Methylation

  • analysis/methylation/methylation_analysis.ipynb: methylation analysis notebook.
  • analysis/methylation/*/: per-sample directories with supporting files.

• Tandem repeats (TR)

  • analysis/tr/adaptive_sampling_tr_analysis.ipynb: TR analysis on adaptive sampling data.
  • analysis/tr/tr_regions/: TR region catalogs used by the analysis, including:
    • adotto_catalog.hg38.lite.*.bed
    • strchive/STRchive-disease-loci.hg38.bed

• Variant counts

  • analysis/variant_counts/count_variants_by_barcode.sh
  • analysis/variant_counts/count_variants_by_barcode_parallel.sh
  • analysis/variant_counts/variant_counts.tsv

Benchmarking

• Folder: benchmarking/
  • adaptive_methylation_benchmark.ipynb: methylation benchmarking notebook.
  • benchmarking/hg002_variant_calls/: example HG002 variant call files and indices for reference.

Benchmarking data DOI: https://doi.org/10.5281/zenodo.17398577

Data availability and citation

If you use these materials, please reference the Zenodo record:

• TBAS benchmarking data: https://doi.org/10.5281/zenodo.17398577

License

This project is distributed under the terms of the license in LICENSE.