Run any Skill in Manus with one click

mfsd-analysis

Stars2

Forks0

UpdatedJuly 1, 2026 at 15:57

Reference for gbcms mFSD (fragment size distribution) — the opt-in --mfsd flag, MAF/VCF column gating, KS test + LLR statistics, and Rust-native ZSTD Parquet output. Use when touching mfsd.rs, parquet_writer.rs, the mFSD report, or column-count tests.

Installation

Install with Codex or Claude Copy this prompt, paste it into Codex, Claude, or another assistant, and let it review the skill page and install it for you.

Run Skill in Manus

Source

msk-access

msk-access/gbcms

View GitHub Repository View Creator Repositories

Download

Run Skill in Manus

Related occupationsSOC

Based on SOC occupation classification

Software DevelopersComputer and Mathematical Occupations·SOC 15-1252

SKILL.md

readonly

name	mfsd-analysis
description	Reference for gbcms mFSD (fragment size distribution) — the opt-in --mfsd flag, MAF/VCF column gating, KS test + LLR statistics, and Rust-native ZSTD Parquet output. Use when touching mfsd.rs, parquet_writer.rs, the mFSD report, or column-count tests.

mFSD Analysis Patterns

Overview

mFSD (modified Fragment Size Distribution) analysis is opt-in via --mfsd.

Column Gating

Without --mfsd: 145 MAF columns (0 mfsd_*)
With --mfsd: 186 MAF columns (+41 mfsd_*)
VCF INFO: exactly 13 MFSD_* fields added only when --mfsd is set
Columns are absent when off, not NA-filled

Parquet Output

--mfsd-parquet requires --mfsd (validated at both CLI and Pydantic)
Written by write_fsd_parquet() in Rust via arrow/parquet crates
ZSTD(1) compression: parquet = { default-features = false, features = ["arrow", "zstd"] }
ref_sizes/alt_sizes are Rust-internal — no #[pyo3(get)]
Called from pipeline.py after count_bam_binned(), not inside the counting engine (count_bam is the feature-gated legacy parity oracle, not the production path)

Statistics (watch-outs)

KS p-value uses an asymptotic approximation — weak at the small ALT-fragment counts (n≈5–20) typical of low-input cfDNA.
LLR is a fragment-size Gaussian log-ratio (distinct from the PairHMM LLR); guard against ±∞ from tail fragments by using the closed-form log-ratio.

Key Files

rust/src/counting/mfsd.rs: KS test, LLR, pairwise comparisons
rust/src/counting/parquet_writer.rs: Arrow/ZSTD native Parquet
src/gbcms/io/output.py: MafWriter column gating
src/gbcms/report/mfsd_report.py: HTML report generator
tests/test_mfsd_flag.py: column count assertions

More from this repository

same repository

rust-python-ffi

msk-access/gbcms

Reference for the gbcms Rust/Python (PyO3) boundary — maturin build commands,

2026-07-012

add-feature

msk-access/gbcms

Procedure for adding a feature to gbcms following its Git Flow and Rust/Python architecture (branch from develop, plan the layer, implement, sync type stubs, add tests with the 4 counting invariants, update docs/CHANGELOG, QA, PR to develop). Use when starting new feature work.

2026-07-012

rna-annotation

msk-access/gbcms

Reference for gbcms RNA mode — GTF-based transcript annotation (--gtf), the COITree AnnotationIndex, exon-boundary distance, per-transcript counts, and ASJD (allele-specific junction detection). Use when touching annotation/, rna.rs, or RNA output columns. Strandedness/sense-antisense gotchas live here.

2026-06-302

counting-engine

msk-access/gbcms

Reference for the gbcms counting core — genomic binning (10kb bins, one fetch per bin), the 4-phase variant-check pipeline, and alignment-backend dispatch. Use when modifying engine.rs or variant_checks.rs, debugging count discrepancies, or reasoning about binning and binned↔legacy parity. For fragment/consensus see fragment-counting; for filters/quality see read-filters-qc.

2026-06-262

debug-error

msk-access/gbcms

Procedure for diagnosing gbcms build/test/CI errors — Rust compile failures, PyO3 stub mismatches, COITree SIMD (nosimd vs NEON/AVX) issues, CI black/ruff version traps, and test failures. Use when hitting an error and you need the gbcms-specific triage path.

2026-06-262

fragment-counting

msk-access/gbcms

Reference for fragment-level counting in gbcms — R1/R2 quality-weighted consensus, FAD/FAF (fragment alt depth/fraction), QNAME/UMI deduplication, duplex vs singleton UMI families, strand consistency, and amplicon split. Use when touching shared/fragment.rs or any DPF/RDF/ADF logic, or debugging fragment-level VAF.

2026-06-262

name	mfsd-analysis
description	Reference for gbcms mFSD (fragment size distribution) — the opt-in --mfsd flag, MAF/VCF column gating, KS test + LLR statistics, and Rust-native ZSTD Parquet output. Use when touching mfsd.rs, parquet_writer.rs, the mFSD report, or column-count tests.

mFSD Analysis Patterns

Overview

mFSD (modified Fragment Size Distribution) analysis is opt-in via --mfsd.

Column Gating

Without --mfsd: 145 MAF columns (0 mfsd_*)
With --mfsd: 186 MAF columns (+41 mfsd_*)
VCF INFO: exactly 13 MFSD_* fields added only when --mfsd is set
Columns are absent when off, not NA-filled

Parquet Output

--mfsd-parquet requires --mfsd (validated at both CLI and Pydantic)
Written by write_fsd_parquet() in Rust via arrow/parquet crates
ZSTD(1) compression: parquet = { default-features = false, features = ["arrow", "zstd"] }
ref_sizes/alt_sizes are Rust-internal — no #[pyo3(get)]
Called from pipeline.py after count_bam_binned(), not inside the counting engine (count_bam is the feature-gated legacy parity oracle, not the production path)

Statistics (watch-outs)

KS p-value uses an asymptotic approximation — weak at the small ALT-fragment counts (n≈5–20) typical of low-input cfDNA.
LLR is a fragment-size Gaussian log-ratio (distinct from the PairHMM LLR); guard against ±∞ from tail fragments by using the closed-form log-ratio.

Key Files

rust/src/counting/mfsd.rs: KS test, LLR, pairwise comparisons
rust/src/counting/parquet_writer.rs: Arrow/ZSTD native Parquet
src/gbcms/io/output.py: MafWriter column gating
src/gbcms/report/mfsd_report.py: HTML report generator
tests/test_mfsd_flag.py: column count assertions