[theontho]

Closes #166.

Ultima Genomics (UG 100) reads come from MyHeritage's low-pass WGS and Nebula/DNA360's 30x. They are single-end, ~300 bp, and their errors are mostly homopolymer lengths. Each read carries flow-space probabilities in tp/t0 tags, and its read group has PL:ULTIMA and the flow order (FO). Ultima's and Broad's pipelines need those to call well. WGS Extract now recognizes Ultima data and keeps that information through the whole pipeline.

Detection: an Ultima read group (PL:ULTIMA), or --platform ultima (needed for SRA FASTQs, which have neither). info recognizes Ultima read names.

align / realign:

• Aligns with BWA-MEM (-Y), as Ultima's and Broad's pipelines do. --gpu, minimap2/pbmm2 and --r2 are refused.
• tp/t0 ride in the FASTQ comment (samtools fastq -T → bwa -C). A streaming filter reverses them on reverse-strand reads before and after BWA; Picard's ATTRIBUTE_TO_REVERSE does this in Ultima's pipeline, and samtools reset doesn't.
• Read groups (with the flow order) are kept via bwa -H. realign streams the BAM/CRAM into BWA without writing FASTQs.
• Duplicates are marked with GATK MarkDuplicates --FLOW_MODE, using Ultima's flags. Without GATK it falls back to samtools markdup and warns.

Callers:

• vcf snp/indel use bcftools' -X ultima profile and call's default prior. WGS Extract's -P 0 more than halved Ultima indel precision (0.13 vs 0.29 on a chr20 slice).
• vcf gatk uses --flow-mode ADVANCED.
• vcf deepvariant uses the ULTIMA model from Google's experimental google/deepvariant:gh1100-ultima image (Support Ultima sequencing data google/deepvariant#1100), configurable as deepvariant_ultima_image.
• vcf freebayes and the GPU HaplotypeCaller warn. GPU DeepVariant refuses Ultima reads.

bam to-ubam keeps tp, t0, tm, rq and a3 in read orientation.

Example genomes: GIAB HG008 Ultima data (an SRA FASTQ, and the 234x GRCh38 CRAM, which samtools can read over HTTPS by region).

Validation (real data)

GIAB HG008-N-D, whole chr20. Ultima's CRAM was downsampled to ~30x and run through realign to hs38DH's chr20, then each caller. The comparison set is the same tissue's Illumina BAM (~40x, bcftools). Both are scored inside GIAB's "not in any difficult region" stratification (50.8 Mb). Alleles and genotype class must match.

Caller on Ultima reads	SNP P	SNP R	SNP F1	Indel P	Indel R	Indel F1
bcftools (-X ultima)	0.998	0.994	0.996	0.354	0.984	0.520
GATK --flow-mode ADVANCED	0.985	0.996	0.991	0.396	0.971	0.563
DeepVariant ULTIMA	0.997	0.994	0.995	0.969	0.950	0.959

Also checked:

• Flow tags after realign match the input exactly: 349k reads, 0 mismatches; reverse-strand reads are flipped correctly.
• tm/a3 tag counts are preserved.
• Ultima's DeepVariant pipeline keeps duplicates; duplicate flags made no difference to any caller here.
• The full unit suite passes (1935 passed). New tests are in tests/test_ultima.py, including real BWA + GATK flow-mode runs.

A whole-genome realign wasn't run: on the shared 8-core host it would take about a day. Whole chr20 at production depth exercised the same code paths.

MyHeritage's 2x low-pass data is too shallow for reliable genotypes without imputation, which is out of scope (noted in the README).

[coderabbitai]

Important

• 🔍 Trigger review

This repository does not receive automatic reviews because it has fewer than 10 stars.

Run configuration

• Source mode: Auto
• Repository: theontho/wgsextract-cli
• Branch: ultima-support
• File: .coderabbit.yaml
• Commit: 370a8ad
• Layers: Repository YAML
• Review profile: CHILL
• Plan: Advanced
• Run ID: a27b1c45-7c6f-4a3d-8fd6-ee1491d21a66

• Autofix · Keep fixing CodeRabbit findings and required CI, and resolving merge conflicts

Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out.

❤️ Share

• X
• Mastodon
• Reddit
• LinkedIn

_{Comment @coderabbitai help to get the list of available commands.}

[Copilot]

🟡 Changes recommended

Ultima handling can incorrectly reverse read-oriented flow arrays and overlook valid FASTQ comment tags, misassigning probabilities to bases or silently dropping flow metadata.

2 open issues

• Preserve read-group orientation for aligned Ultima flow arrays · New
• Detect FASTQ tags in space-delimited headers · New

What changed in this PR

Adds Ultima Genomics support across alignment, variant calling, and uBAM conversion, addressing #166 with platform-specific processing and public example datasets.

Changes:

• Detects Ultima reads and adds BWA alignment with flow-tag preservation and flow-aware duplicate marking.
• Selects Ultima-specific caller settings and documents unsupported GPU paths.
• Adds regression tests, GIAB examples, usage guidance, and chr20 validation results.

File	Description
tests/​test_ultima.py	Tests detection, flow tags, alignment, and model selection.
src/​wgsextract_cli/​core/​ultima.py	Adds shared Ultima detection and flow-tag utilities.
src/​wgsextract_cli/​core/​constants.py	Recognizes Ultima read names.
src/​wgsextract_cli/​core/​config.py	Adds the Ultima DeepVariant image setting.
src/​wgsextract_cli/​commands/​vcf/​deepvariant.py	Selects the ULTIMA model and experimental image.
src/​wgsextract_cli/​commands/​vcf/​callers.py	Enables GATK flow mode and caller warnings.
src/​wgsextract_cli/​commands/​vcf/​basic.py	Applies Ultima bcftools settings and default prior.
src/​wgsextract_cli/​commands/​vcf/​__init__.py	Exposes the ULTIMA model option.
src/​wgsextract_cli/​commands/​realign.py	Detects Ultima inputs and streams realignment.
src/​wgsextract_cli/​commands/​align.py	Adds Ultima alignment and duplicate marking.
src/​wgsextract_cli/​commands/​_microarray_vcf.py	Applies the Ultima profile to microarray calling.
src/​wgsextract_cli/​commands/​_examples_catalog.py	Adds public GIAB Ultima FASTQ and CRAM examples.
src/​wgsextract_cli/​commands/​_bam_ubam.py	Preserves Ultima tags during uBAM conversion.
site_src/​workflows.md	Documents Ultima alignment workflows.
README.md	Summarizes support, limitations, and validation results.

_{Review effort: Balanced}