P5a: QC — control metrics, thresholds, status, contamination, relatedness, locus summary
- qc/controls: canonical control names (QC-002), bkg/offset (QC-003), 13 genotyping metrics (QC-004/005), missing/zero-divisor handling (QC-007/008); controls_raw.tsv / controls_qc.tsv (QC-009/011) incl. compat CSVs - qc/thresholds + profiles: YAML schema, validation codes, precedence, shipped profiles, auto (QC-020..025) - qc/status: functional PASS/FAIL/NA, control CLEAR/FLAGGED, flags (QC-030..033) - qc/contamination: BAF regression; bit-compatible with gtc2vcf BAFregress (36/36, K-017), exact mode too - qc/relatedness: KING-robust, duplicates, Mendelian errors, pedigree from sheet (IN-018) - qc/locus_summary: OUT-072, bit-identical to BeadArrayFiles except ChiTest (K-018) - qc/suggest: suggest-thresholds (QC-026) - pipeline/qc: disk-backed cohort store; CLI qc controls|samples|contamination|relatedness|loci|suggest-thresholds - genotype: control tables (OUT-070), het_rate, QC columns, cohort QC - tests: unit + oracle (BAFregress plugin, BeadArrayFiles locus summary)
This commit is contained in:
kemal committed
1 parent
d11226ca31
commit
a8caa26aa4
28 files changed
+3355
-22
No files matched your search
+164
-11
@@ -22,6 +22,7 @@ from simurg.core.errors import ExitCode, SimurgError
|
||||
from simurg.core.log import Reporter
|
||||
from simurg.core.provenance import RunManifest
|
||||
from simurg.formats.bpm import read_bpm
|
||||
from simurg.formats.csv_manifest import assay_format
|
||||
from simurg.formats.egt import read_egt
|
||||
from simurg.formats.gtc import write_gtc
|
||||
from simurg.formats.idat import check_pair, read_idat
|
||||
@@ -37,6 +38,24 @@ from simurg.genotype.sample import (
|
||||
)
|
||||
from simurg.genotype.sex import SexParams
|
||||
from simurg.pipeline.inputs import check_paths, resolve_sample_set, sample_dir, sheet_values
|
||||
from simurg.pipeline.qc import (
|
||||
CONTROLS_QC,
|
||||
CONTROLS_RAW,
|
||||
MENDEL_COLUMNS,
|
||||
RELATEDNESS_COLUMNS,
|
||||
ControlSample,
|
||||
cohort_qc,
|
||||
control_sample,
|
||||
default_tmp,
|
||||
load_cohort,
|
||||
load_control_beads,
|
||||
pedigree_from_sheet,
|
||||
write_control_tables,
|
||||
write_tsv,
|
||||
)
|
||||
from simurg.qc.controls import ControlBead
|
||||
from simurg.qc.status import control_status, functional_status, het_zscores, sample_flags
|
||||
from simurg.qc.thresholds import Thresholds, parse_cli_pairs, resolve_thresholds
|
||||
from simurg.signal.normalization.apply import normalize, polar
|
||||
from simurg.signal.raw_xy import LocusMapCache, check_coverage
|
||||
|
||||
@@ -67,7 +86,7 @@ FIXED_COLUMNS = [
|
||||
"het_rate",
|
||||
"contamination",
|
||||
]
|
||||
QC_COLUMNS = ["functional_status", "control_status", "control_flag_codes", "flags"] # filled from phase P5
|
||||
QC_COLUMNS = ["functional_status", "control_status", "control_flag_codes", "flags"] # QC-030/031/033
|
||||
FLOAT_COLUMNS = {"autosomal_call_rate", "call_rate", "logr_dev", "gc10", "gc50", "het_rate", "contamination"}
|
||||
|
||||
|
||||
@@ -90,6 +109,14 @@ class GenotypeOptions:
|
||||
gtc_version: int | None = None
|
||||
autocall_date: str | None = None
|
||||
assay_control_probes: tuple[str, ...] = ("TGA_Ctrl_5716",)
|
||||
controls: bool = True # OUT-070
|
||||
cohort_qc: bool = True # contamination + relatedness after genotyping (QC-064)
|
||||
qc_config: Path | None = None
|
||||
qc_profile: str | None = None
|
||||
threshold_call_rate: float | None = None
|
||||
threshold_lrr_sd: float | None = None
|
||||
thresholds: list[str] = field(default_factory=list)
|
||||
bg_offset: float | None = None
|
||||
argv: list[str] = field(default_factory=list)
|
||||
|
||||
|
||||
@@ -100,6 +127,7 @@ class Job:
|
||||
grn: str
|
||||
red: str
|
||||
gtc_path: str
|
||||
sheet_values: dict[str, str] = field(default_factory=dict)
|
||||
|
||||
|
||||
_W: dict[str, Any] = {}
|
||||
@@ -111,8 +139,12 @@ def _init(
|
||||
auto_1_22: np.ndarray[Any, np.dtype[np.bool_]],
|
||||
probe_idx: dict[str, int],
|
||||
started: _dt.datetime,
|
||||
beads: list[ControlBead] | None = None,
|
||||
bg_offset: float = 3000.0,
|
||||
) -> None:
|
||||
_W.update(ctx=ctx, params=params, auto=auto_1_22, probes=probe_idx, started=started)
|
||||
_W.update(
|
||||
ctx=ctx, params=params, auto=auto_1_22, probes=probe_idx, started=started, beads=beads, bg_offset=bg_offset
|
||||
)
|
||||
|
||||
|
||||
def _process(job: Job) -> dict[str, Any]:
|
||||
@@ -134,6 +166,11 @@ def _process(job: Job) -> dict[str, Any]:
|
||||
write_gtc(g, Path(job.gtc_path))
|
||||
auto = _W["auto"] & ~ctx.manifest.intensity_only
|
||||
n_auto = int(auto.sum())
|
||||
g_auto = g.genotypes[auto]
|
||||
n_called = int((g_auto > 0).sum())
|
||||
row["het_rate"] = float((g_auto == 2).sum() / n_called) if n_called else math.nan
|
||||
if _W.get("beads"):
|
||||
row["control"] = control_sample(job.sample_id, grn, red, _W["beads"], _W["bg_offset"], job.sheet_values)
|
||||
row.update(
|
||||
autosomal_call_rate=float(np.float32(int((g.genotypes[auto] > 0).sum())) / np.float32(n_auto))
|
||||
if n_auto
|
||||
@@ -233,10 +270,22 @@ def run_genotype(opts: GenotypeOptions, rep: Reporter) -> int:
|
||||
continue
|
||||
stem = sset.stems[s.sample_id]
|
||||
gtc = sample_dir(opts.out, stem, opts.layout) / f"{stem}.gtc"
|
||||
jobs.append(Job(s.sample_id, s.sample_name, str(s.pair.grn), str(s.pair.red), str(gtc)))
|
||||
jobs.append(
|
||||
Job(
|
||||
s.sample_id,
|
||||
s.sample_name,
|
||||
str(s.pair.grn),
|
||||
str(s.pair.red),
|
||||
str(gtc),
|
||||
{k: str(v) for k, v in sheet_values(s, sset.sheet).items()},
|
||||
)
|
||||
)
|
||||
|
||||
summary_tsv, summary_json = opts.out / "sample_summary.tsv", opts.out / "sample_summary.json"
|
||||
check_outputs_absent([summary_tsv, summary_json] + [Path(j.gtc_path) for j in jobs], force=opts.force)
|
||||
extra = [opts.out / CONTROLS_RAW, opts.out / CONTROLS_QC] if opts.controls else []
|
||||
if opts.cohort_qc:
|
||||
extra.append(opts.out / "relatedness.tsv")
|
||||
check_outputs_absent([summary_tsv, summary_json, *extra] + [Path(j.gtc_path) for j in jobs], force=opts.force)
|
||||
|
||||
rep.info(f"reading manifest {opts.bpm} and cluster file {opts.egt}")
|
||||
manifest = read_bpm(opts.bpm)
|
||||
@@ -256,6 +305,10 @@ def run_genotype(opts: GenotypeOptions, rep: Reporter) -> int:
|
||||
"W_EGT_NAME", f"cluster file was built for {egt.manifest_name!r}, manifest is {manifest.manifest_name!r}"
|
||||
)
|
||||
clusters = align_clusters(egt, manifest)
|
||||
th = _thresholds(opts, manifest)
|
||||
for rec in th.as_records():
|
||||
rm.set_option(f"qc.{rec['key']}", rec["value"], rec["source"])
|
||||
beads = load_control_beads(manifest, opts.csv, rep) if opts.controls else []
|
||||
ctx = manifest_context(manifest, clusters, cluster_file_name=Path(opts.egt).name)
|
||||
lite = manifest.lite()
|
||||
ctx = replace(ctx, manifest=lite, cache=LocusMapCache(lite)) # workers get no string columns
|
||||
@@ -282,14 +335,14 @@ def run_genotype(opts: GenotypeOptions, rep: Reporter) -> int:
|
||||
rep.info(f"genotyping {len(jobs)} sample(s) with {workers} worker(s)")
|
||||
try:
|
||||
if workers == 1:
|
||||
_init(ctx, params, auto_1_22, probes, started)
|
||||
_init(ctx, params, auto_1_22, probes, started, beads, th.bg_offset)
|
||||
results = [_process(j) for j in jobs]
|
||||
else:
|
||||
with ProcessPoolExecutor(
|
||||
workers,
|
||||
mp_context=mp.get_context("spawn"),
|
||||
initializer=_init,
|
||||
initargs=(ctx, params, auto_1_22, probes, started),
|
||||
initargs=(ctx, params, auto_1_22, probes, started, beads, th.bg_offset),
|
||||
) as ex:
|
||||
results = list(ex.map(_process, jobs))
|
||||
except KeyboardInterrupt:
|
||||
@@ -315,14 +368,11 @@ def run_genotype(opts: GenotypeOptions, rep: Reporter) -> int:
|
||||
rows[r["sample_id"]].update(r)
|
||||
|
||||
probe_cols = [f"{p}_norm_r" for p in probes]
|
||||
_run_qc(opts, rep, rm, th, beads, sset, rows, results, jobs, manifest, probe_cols, params.gtc_version)
|
||||
sheet_cols = [f"sheet:{c}" for c in (sset.sheet.passthrough_columns() if sset.sheet else [])]
|
||||
out_rows = []
|
||||
for s in sset.samples:
|
||||
r = rows[s.sample_id]
|
||||
known = {"M", "F"}
|
||||
est, rep_sex = r.get("sex_est", "U"), r.get("sex_reported", "U")
|
||||
r["sex_check"] = ("match" if est == rep_sex else "mismatch") if est in known and rep_sex in known else "unknown"
|
||||
out_rows.append(r)
|
||||
out_rows.append(rows[s.sample_id])
|
||||
columns = FIXED_COLUMNS + probe_cols + QC_COLUMNS + sheet_cols
|
||||
header = [c.removeprefix("sheet:") for c in columns]
|
||||
lines = ["\t".join(header)] + ["\t".join(_fmt(c, r.get(c, "")) for c in columns) for r in out_rows]
|
||||
@@ -350,3 +400,106 @@ def run_genotype(opts: GenotypeOptions, rep: Reporter) -> int:
|
||||
if ok == len(out_rows):
|
||||
return ExitCode.OK
|
||||
return ExitCode.PARTIAL if ok else ExitCode.ALL_FAILED
|
||||
|
||||
|
||||
def _thresholds(opts: GenotypeOptions, manifest: Any) -> Thresholds:
|
||||
cli: dict[str, Any] = parse_cli_pairs(opts.thresholds)
|
||||
if opts.threshold_call_rate is not None:
|
||||
cli["callRate"] = opts.threshold_call_rate
|
||||
if opts.threshold_lrr_sd is not None:
|
||||
cli["logRDev"] = opts.threshold_lrr_sd
|
||||
if opts.bg_offset is not None:
|
||||
cli["bgOffset"] = opts.bg_offset
|
||||
return resolve_thresholds(
|
||||
config=opts.qc_config,
|
||||
profile=opts.qc_profile,
|
||||
cli=cli,
|
||||
array_type="genotyping",
|
||||
assay_format=assay_format(manifest),
|
||||
manifest_name=manifest.manifest_name,
|
||||
)
|
||||
|
||||
|
||||
def _run_qc(
|
||||
opts: GenotypeOptions,
|
||||
rep: Reporter,
|
||||
rm: RunManifest,
|
||||
th: Thresholds,
|
||||
beads: list[ControlBead],
|
||||
sset: Any,
|
||||
rows: dict[str, dict[str, Any]],
|
||||
results: list[dict[str, Any]],
|
||||
jobs: list[Job],
|
||||
manifest: Any,
|
||||
probe_cols: list[str],
|
||||
gtc_version: int,
|
||||
) -> None:
|
||||
"""Control tables (OUT-070), het rate, cohort QC and the QC-030/031/033 columns of the summary."""
|
||||
controls: dict[str, ControlSample] = {r["sample_id"]: r.pop("control") for r in results if "control" in r}
|
||||
for r in rows.values(): # results were merged into rows already; drop the heavy objects there too
|
||||
r.pop("control", None)
|
||||
if opts.controls and beads:
|
||||
assay = {
|
||||
sid: min((float(rows[sid].get(c, math.nan)) for c in probe_cols), default=math.nan) for sid in controls
|
||||
}
|
||||
ordered = [controls[s.sample_id] for s in sset.samples if s.sample_id in controls]
|
||||
for p in write_control_tables(
|
||||
opts.out, ordered, beads, th.bg_offset, rep, assay_control=assay if probe_cols else None
|
||||
):
|
||||
rm.add_output(p, "simurg.controls/1")
|
||||
ids = [s.sample_id for s in sset.samples]
|
||||
het_z = dict(zip(ids, het_zscores([rows[i].get("het_rate", math.nan) for i in ids]), strict=True))
|
||||
dups: dict[str, list[str]] = {}
|
||||
unexpected: set[str] = set()
|
||||
ok_jobs = [j for j in jobs if rows[j.sample_id].get("status") == "OK"]
|
||||
if opts.cohort_qc and len(ok_jobs) >= 2:
|
||||
tmp = default_tmp(opts.out)
|
||||
rep.info(f"cohort QC (contamination, relatedness) on {len(ok_jobs)} sample(s)")
|
||||
store, infos = load_cohort(
|
||||
[(j.sample_id, j.gtc_path) for j in ok_jobs],
|
||||
manifest.num_loci,
|
||||
tmp,
|
||||
baf=gtc_version >= 4,
|
||||
scores=False,
|
||||
workers=max(1, opts.threads or (os.cpu_count() or 1)),
|
||||
)
|
||||
try:
|
||||
ped = pedigree_from_sheet(sset.sheet, [j.sample_id for j in ok_jobs], rep)
|
||||
res = cohort_qc(
|
||||
store, infos, manifest, rep, pedigree=ped, workers=max(1, opts.threads or (os.cpu_count() or 1))
|
||||
)
|
||||
finally:
|
||||
store.close()
|
||||
if res.contamination is not None:
|
||||
for i, s in enumerate(infos):
|
||||
rows[s.sample_id]["contamination"] = float(res.contamination.estimate[i])
|
||||
dups, unexpected = res.duplicates, res.unexpected
|
||||
write_tsv(opts.out / "relatedness.tsv", RELATEDNESS_COLUMNS, res.pairs)
|
||||
rm.add_output(opts.out / "relatedness.tsv", "simurg.relatedness/1")
|
||||
if res.mendel:
|
||||
write_tsv(opts.out / "mendel.tsv", MENDEL_COLUMNS, res.mendel)
|
||||
rm.add_output(opts.out / "mendel.tsv", "simurg.mendel/1")
|
||||
for sid in ids:
|
||||
r = rows[sid]
|
||||
known = {"M", "F"}
|
||||
est, rep_sex = r.get("sex_est", "U"), r.get("sex_reported", "U")
|
||||
r["sex_check"] = ("match" if est == rep_sex else "mismatch") if est in known and rep_sex in known else "unknown"
|
||||
if r.get("status") == "MISSING_INPUT":
|
||||
continue
|
||||
r["functional_status"] = functional_status(r, th)
|
||||
if sid in controls:
|
||||
vals = dict(controls[sid].qc.values)
|
||||
if probe_cols:
|
||||
vals["assay_control_norm_r"] = min(float(r.get(c, math.nan)) for c in probe_cols)
|
||||
cs, codes = control_status(vals, controls[sid].qc.na_metrics, th)
|
||||
r["control_status"], r["control_flag_codes"] = cs, ";".join(codes)
|
||||
r["flags"] = ";".join(
|
||||
sample_flags(
|
||||
r,
|
||||
th,
|
||||
het_z=het_z.get(sid, math.nan),
|
||||
probe_columns=probe_cols,
|
||||
duplicates=dups.get(sid, []),
|
||||
unexpected_relatedness=sid in unexpected,
|
||||
)
|
||||
)
|
||||
@@ -0,0 +1,915 @@
|
||||
"""QC orchestration and I/O (spec QC-001 … QC-064, CLI-024, CLI-025, OUT-070, OUT-072; plan §9).
|
||||
|
||||
The science lives in ``simurg.qc``; this module reads IDATs/GTCs, streams cohort matrices through disk-backed
|
||||
arrays (so thousands of samples fit in a small memory budget) and writes the QC tables.
|
||||
"""
|
||||
|
||||
from __future__ import annotations
|
||||
|
||||
import datetime as _dt
|
||||
import math
|
||||
import multiprocessing as mp
|
||||
import os
|
||||
import shutil
|
||||
import tempfile
|
||||
from collections.abc import Iterator, Mapping, Sequence
|
||||
from concurrent.futures import ProcessPoolExecutor
|
||||
from dataclasses import dataclass, field
|
||||
from pathlib import Path
|
||||
from typing import IO, Any, cast
|
||||
|
||||
import numpy as np
|
||||
import numpy.typing as npt
|
||||
|
||||
from simurg.core.atomic import atomic_write_text, check_outputs_absent
|
||||
from simurg.core.errors import ExitCode, SimurgError
|
||||
from simurg.core.log import Reporter
|
||||
from simurg.core.provenance import RunManifest
|
||||
from simurg.formats.bpm import Manifest
|
||||
from simurg.formats.idat import Idat
|
||||
from simurg.formats.samplesheet import SampleSheet, fold
|
||||
from simurg.genotype.baf_lrr import chrom_core
|
||||
from simurg.qc.contamination import ContaminationResult, baf_regress
|
||||
from simurg.qc.controls import (
|
||||
ASSAY_CONTROL_COLUMN,
|
||||
COMPAT_METRIC_NAMES,
|
||||
ControlBead,
|
||||
ControlQC,
|
||||
beads_from_definitions,
|
||||
compare_definitions,
|
||||
compute_control_metrics,
|
||||
control_intensities,
|
||||
metrics_for,
|
||||
)
|
||||
from simurg.qc.relatedness import DEGREE_CUTS, DUPLICATE_CONCORDANCE, degree, mendel_errors, pair_stats, select_sites
|
||||
from simurg.qc.status import het_zscores
|
||||
from simurg.qc.thresholds import Thresholds
|
||||
|
||||
CONTROLS_RAW = "controls_raw.tsv"
|
||||
CONTROLS_QC = "controls_qc.tsv"
|
||||
COMPAT_RAW = "controls.raw_metrics.csv"
|
||||
COMPAT_QC = "controls.qc_metrics.csv"
|
||||
MIN_CONTAMINATION_SAMPLES = 10
|
||||
|
||||
|
||||
def fmt(v: Any, digits: int = 8) -> str:
|
||||
if v is None or v == "":
|
||||
return ""
|
||||
if isinstance(v, float | np.floating):
|
||||
f = float(v)
|
||||
return "NaN" if math.isnan(f) else f"{f:.{digits}f}"
|
||||
return str(v).replace("\t", " ").replace("\n", " ")
|
||||
|
||||
|
||||
def write_tsv(path: Path, columns: Sequence[str], rows: Sequence[Mapping[str, Any]], sep: str = "\t") -> None:
|
||||
lines = [sep.join(columns)] + [sep.join(fmt(r.get(c, "")) for c in columns) for r in rows]
|
||||
atomic_write_text(path, "\n".join(lines) + "\n")
|
||||
|
||||
|
||||
def autosome_mask(chroms: Sequence[str]) -> npt.NDArray[np.bool_]:
|
||||
"""Chromosomes 1–22 only (het rate, relatedness, contamination sites)."""
|
||||
auto = {str(i) for i in range(1, 23)}
|
||||
return np.array([chrom_core(c) in auto for c in chroms], dtype=np.bool_)
|
||||
|
||||
|
||||
# ================================================================================================ controls
|
||||
def load_control_beads(manifest: Manifest | None, csv_path: Path | None, rep: Reporter) -> list[ControlBead]:
|
||||
"""QC-001: BPM control config wins over the CSV ``[Controls]`` section; differences give one warning."""
|
||||
bpm_beads = beads_from_definitions(manifest.controls) if manifest is not None else []
|
||||
csv_beads: list[ControlBead] = []
|
||||
if csv_path is not None:
|
||||
from simurg.formats.csv_manifest import read_csv_manifest
|
||||
|
||||
csv_beads = beads_from_definitions(read_csv_manifest(csv_path, load_sequences=False).controls)
|
||||
if bpm_beads and csv_beads and (diffs := compare_definitions(bpm_beads, csv_beads)):
|
||||
rep.warn("W_CTRL_DEFS_DIFFER", "BPM and CSV control definitions differ (using the BPM): " + "; ".join(diffs))
|
||||
return bpm_beads or csv_beads
|
||||
|
||||
|
||||
@dataclass
|
||||
class ControlSample:
|
||||
sample_id: str
|
||||
meta: dict[str, str]
|
||||
green: npt.NDArray[np.float64]
|
||||
red: npt.NDArray[np.float64]
|
||||
qc: ControlQC
|
||||
|
||||
|
||||
def _param(pars: str, key: str) -> str:
|
||||
from simurg.genotype.sample import _param as p
|
||||
|
||||
return p(pars, key)
|
||||
|
||||
|
||||
def idat_metadata(grn: Idat, red: Idat, sheet_values: Mapping[str, str] | None = None) -> dict[str, str]:
|
||||
"""QC-011 metadata: IDAT values, plate and well overridden by the sheet."""
|
||||
scan = red.scan_entry() or grn.scan_entry()
|
||||
sv = {fold(k): v for k, v in (sheet_values or {}).items()}
|
||||
return {
|
||||
"imaging_date": red.scan_time_iso or grn.scan_time_iso,
|
||||
"sample_plate": sv.get(fold("Sample_Plate")) or red.plate or grn.plate,
|
||||
"sample_well": sv.get(fold("Sample_Well")) or red.well or grn.well,
|
||||
"scanner_id": _param(scan.block_pars, "sherlockID") if scan else "",
|
||||
"scanner_version": scan.code_version if scan else "",
|
||||
}
|
||||
|
||||
|
||||
def control_sample(
|
||||
sample_id: str,
|
||||
grn: Idat,
|
||||
red: Idat,
|
||||
beads: Sequence[ControlBead],
|
||||
bg_offset: float,
|
||||
sheet_values: Mapping[str, str] | None = None,
|
||||
) -> ControlSample:
|
||||
g, r = control_intensities(grn, red, beads)
|
||||
qc = compute_control_metrics(beads, g, r, "genotyping", bg_offset)
|
||||
return ControlSample(sample_id, idat_metadata(grn, red, sheet_values), g, r, qc)
|
||||
|
||||
|
||||
def _compat_date(iso: str) -> str:
|
||||
try:
|
||||
t = _dt.datetime.fromisoformat(iso)
|
||||
except ValueError:
|
||||
return iso
|
||||
from simurg.genotype.sample import invariant_datetime
|
||||
|
||||
return invariant_datetime(t)
|
||||
|
||||
|
||||
def write_control_tables(
|
||||
out: Path,
|
||||
samples: Sequence[ControlSample],
|
||||
beads: Sequence[ControlBead],
|
||||
bg_offset: float,
|
||||
rep: Reporter,
|
||||
*,
|
||||
array_type: str = "genotyping",
|
||||
assay_control: Mapping[str, float] | None = None,
|
||||
compat: bool = False,
|
||||
) -> list[Path]:
|
||||
"""QC-009 and QC-011 (or their compatibility CSVs). ``assay_control`` maps sample_id → assay-control norm R."""
|
||||
metrics = metrics_for(array_type)
|
||||
missing: dict[str, list[str]] = {}
|
||||
for s in samples:
|
||||
missing.update(s.qc.missing)
|
||||
for col in s.qc.zero_divisor:
|
||||
rep.warn("W_CTRL_ZERO_DIVISOR", f"{s.sample_id}: {col} has a zero denominator", sample_id=s.sample_id)
|
||||
if missing:
|
||||
ids = sorted({c for v in missing.values() for c in v})
|
||||
rep.warn(
|
||||
"W_CTRL_MISSING",
|
||||
f"manifest lacks control(s) {', '.join(ids)}; not evaluable: {', '.join(sorted(missing))}",
|
||||
)
|
||||
raw_rows = [
|
||||
{
|
||||
"sample_id": s.sample_id,
|
||||
"category": b.category,
|
||||
"bead_address": b.address,
|
||||
"control_name": b.name,
|
||||
"color_label": b.color,
|
||||
"green_intensity": "NaN" if math.isnan(gv) else int(gv),
|
||||
"red_intensity": "NaN" if math.isnan(rv) else int(rv),
|
||||
}
|
||||
for s in samples
|
||||
for b, gv, rv in zip(beads, s.green.tolist(), s.red.tolist(), strict=True)
|
||||
]
|
||||
qc_rows = []
|
||||
for s in samples:
|
||||
row: dict[str, Any] = {
|
||||
"sample_id": s.sample_id,
|
||||
"array_type": array_type,
|
||||
**s.meta,
|
||||
"bg_offset": float(bg_offset),
|
||||
**{m.column: s.qc.values.get(m.column, math.nan) for m in metrics},
|
||||
"na_metrics": ";".join(s.qc.na_metrics),
|
||||
}
|
||||
if assay_control is not None:
|
||||
row[ASSAY_CONTROL_COLUMN] = assay_control.get(s.sample_id, math.nan)
|
||||
qc_rows.append(row)
|
||||
meta_cols = ["imaging_date", "sample_plate", "sample_well", "scanner_id", "scanner_version"]
|
||||
qc_cols = ["sample_id", "array_type", *meta_cols, "bg_offset", *(m.column for m in metrics)]
|
||||
if assay_control is not None:
|
||||
qc_cols.append(ASSAY_CONTROL_COLUMN)
|
||||
qc_cols.append("na_metrics")
|
||||
if not compat:
|
||||
raw_cols = list(raw_rows[0]) if raw_rows else ["sample_id"]
|
||||
write_tsv(out / CONTROLS_RAW, raw_cols, raw_rows)
|
||||
write_tsv(out / CONTROLS_QC, qc_cols, qc_rows)
|
||||
return [out / CONTROLS_RAW, out / CONTROLS_QC]
|
||||
ren_raw = {
|
||||
"sample_id": "SampleId",
|
||||
"category": "Category",
|
||||
"bead_address": "Beadtype",
|
||||
"control_name": "Control",
|
||||
"color_label": "Color",
|
||||
"green_intensity": "GreenIntensity",
|
||||
"red_intensity": "RedIntensity",
|
||||
}
|
||||
ren_qc = {
|
||||
"sample_id": "SampleId",
|
||||
"imaging_date": "ImagingDate",
|
||||
"sample_plate": "SamplePlate",
|
||||
"sample_well": "SampleWell",
|
||||
"scanner_id": "ScannerId",
|
||||
"scanner_version": "ScannerVersion",
|
||||
**COMPAT_METRIC_NAMES,
|
||||
}
|
||||
raw_c = [{ren_raw[k]: v for k, v in r.items()} for r in raw_rows]
|
||||
qc_keep = [c for c in qc_cols if c in ren_qc]
|
||||
qc_c = [
|
||||
{ren_qc[c]: (_compat_date(r[c]) if c == "imaging_date" else r.get(c, "")) for c in qc_keep} for r in qc_rows
|
||||
]
|
||||
write_tsv(out / COMPAT_RAW, list(ren_raw.values()), raw_c, sep=",")
|
||||
write_tsv(out / COMPAT_QC, [ren_qc[c] for c in qc_keep], qc_c, sep=",")
|
||||
return [out / COMPAT_RAW, out / COMPAT_QC]
|
||||
|
||||
|
||||
@dataclass
|
||||
class QcControlsOptions:
|
||||
out: Path
|
||||
bpm: Path | None = None
|
||||
csv: Path | None = None
|
||||
idats: Path | None = None
|
||||
sheet: Path | None = None
|
||||
array_type: str = "genotyping"
|
||||
bg_offset: float = 3000.0
|
||||
compat: bool = False
|
||||
force: bool = False
|
||||
on_duplicate_idat: str = "error"
|
||||
argv: list[str] = field(default_factory=list)
|
||||
|
||||
|
||||
def run_qc_controls(opts: QcControlsOptions, rep: Reporter) -> int:
|
||||
"""``simurg qc controls`` — control tables from IDATs without genotyping (QC-012, QC-014)."""
|
||||
from simurg.formats.bpm import read_bpm
|
||||
from simurg.formats.idat import check_pair, read_idat
|
||||
from simurg.pipeline.inputs import check_paths, resolve_sample_set, sheet_values
|
||||
|
||||
rm = RunManifest(command="qc controls", argv=opts.argv)
|
||||
rep.run_id = rm.run_id
|
||||
if opts.array_type == "methylation":
|
||||
raise SimurgError("E_QC_METHYLATION", "methylation control metrics are planned for v2 (QC-006)")
|
||||
if opts.bpm is None and opts.csv is None:
|
||||
raise SimurgError("E_NO_INPUT", "give --bpm or --csv for the control definitions (QC-001)")
|
||||
if opts.bg_offset < 0:
|
||||
raise SimurgError("E_QC_THRESHOLD_RANGE", f"--bg-offset must be >= 0, got {opts.bg_offset}")
|
||||
check_paths([(opts.bpm, "BPM"), (opts.csv, "CSV manifest"), (opts.idats, "IDAT folder"), (opts.sheet, "sheet")])
|
||||
sset = resolve_sample_set(opts.idats, opts.sheet, opts.on_duplicate_idat, rep, rm)
|
||||
names = (COMPAT_RAW, COMPAT_QC) if opts.compat else (CONTROLS_RAW, CONTROLS_QC)
|
||||
targets = [opts.out / n for n in names]
|
||||
check_outputs_absent(targets, force=opts.force) # QC-014: other files in the folder are fine
|
||||
manifest = read_bpm(opts.bpm) if opts.bpm else None
|
||||
beads = load_control_beads(manifest, opts.csv, rep)
|
||||
if not beads:
|
||||
raise SimurgError("E_BPM_FORMAT", "the manifest has no control definitions")
|
||||
samples: list[ControlSample] = []
|
||||
for s in sset.samples:
|
||||
if s.pair is None or not s.pair.complete:
|
||||
rep.warn("W_IDAT_UNPAIRED", f"{s.sample_id}: no complete IDAT pair; skipped", sample_id=s.sample_id)
|
||||
continue
|
||||
assert s.pair.grn is not None and s.pair.red is not None
|
||||
grn, red = read_idat(s.pair.grn), read_idat(s.pair.red)
|
||||
check_pair(grn, red)
|
||||
samples.append(control_sample(s.sample_id, grn, red, beads, opts.bg_offset, sheet_values(s, sset.sheet)))
|
||||
# QC-014: sample IDs of an existing summary must match
|
||||
summary = opts.out / "sample_summary.tsv"
|
||||
if summary.exists():
|
||||
ids = {ln.split("\t", 1)[0] for ln in summary.read_text(encoding="utf-8").splitlines()[1:] if ln}
|
||||
mine = {s.sample_id for s in samples}
|
||||
if ids ^ mine:
|
||||
rep.warn("W_QC_UNMATCHED_SAMPLES", f"{len(ids ^ mine)} sample ID(s) differ from {summary}")
|
||||
opts.out.mkdir(parents=True, exist_ok=True)
|
||||
written = write_control_tables(opts.out, samples, beads, opts.bg_offset, rep, compat=opts.compat)
|
||||
for p in written:
|
||||
rm.add_output(p, "simurg.controls/1")
|
||||
rm.summary = {"samples": len(samples)}
|
||||
rm.status = "completed"
|
||||
rm.write(opts.out / "run_manifest.qc_controls.json")
|
||||
rep.info(f"control QC for {len(samples)} sample(s) written to {opts.out}")
|
||||
return ExitCode.OK
|
||||
|
||||
|
||||
# ================================================================================================ cohort store
|
||||
@dataclass
|
||||
class SampleInfo:
|
||||
sample_id: str
|
||||
path: str
|
||||
sample_name: str = ""
|
||||
call_rate: float = math.nan
|
||||
logr_dev: float = math.nan
|
||||
sex_est: str = "U"
|
||||
has_baf: bool = False
|
||||
error: str = ""
|
||||
|
||||
|
||||
class CohortStore:
|
||||
"""Disk-backed (samples × loci) matrices: genotypes (uint8), BAF and GenCall scores (float32, optional)."""
|
||||
|
||||
def __init__(self, root: Path, n_samples: int, n_loci: int, *, baf: bool, scores: bool) -> None:
|
||||
self.root, self.shape = root, (n_samples, n_loci)
|
||||
root.mkdir(parents=True, exist_ok=True)
|
||||
self.gt = np.lib.format.open_memmap(root / "gt.npy", "w+", np.uint8, self.shape)
|
||||
self.baf = np.lib.format.open_memmap(root / "baf.npy", "w+", np.float32, self.shape) if baf else None
|
||||
self.score = np.lib.format.open_memmap(root / "score.npy", "w+", np.float32, self.shape) if scores else None
|
||||
|
||||
def close(self) -> None:
|
||||
for name in ("gt", "baf", "score"):
|
||||
a = getattr(self, name)
|
||||
if a is not None:
|
||||
a.flush()
|
||||
self.gt = self.baf = self.score = None # type: ignore[assignment]
|
||||
shutil.rmtree(self.root, ignore_errors=True)
|
||||
|
||||
|
||||
def _load_one(task: tuple[int, str, str, str, bool, bool]) -> SampleInfo:
|
||||
from simurg.formats.gtc import read_gtc
|
||||
|
||||
i, sid, path, root, want_baf, want_scores = task
|
||||
info = SampleInfo(sid, path)
|
||||
try:
|
||||
g = read_gtc(path)
|
||||
info.sample_name, info.call_rate, info.logr_dev, info.sex_est = (
|
||||
g.sample_name,
|
||||
float(g.call_rate),
|
||||
float(g.logr_dev) if g.version >= 4 else math.nan,
|
||||
g.gender_char,
|
||||
)
|
||||
gt = np.load(Path(root) / "gt.npy", mmap_mode="r+")
|
||||
if len(g.genotypes) != gt.shape[1]:
|
||||
raise SimurgError("E_GTC_UNSUITABLE", f"{path}: {len(g.genotypes)} loci, manifest has {gt.shape[1]}")
|
||||
gt[i] = g.genotypes
|
||||
gt.flush()
|
||||
if want_baf:
|
||||
baf = np.load(Path(root) / "baf.npy", mmap_mode="r+")
|
||||
info.has_baf = len(g.b_allele_freqs) == gt.shape[1]
|
||||
baf[i] = g.b_allele_freqs if info.has_baf else np.nan
|
||||
baf.flush()
|
||||
if want_scores:
|
||||
sc = np.load(Path(root) / "score.npy", mmap_mode="r+")
|
||||
sc[i] = g.genotype_scores
|
||||
sc.flush()
|
||||
except SimurgError as e:
|
||||
info.error = f"{e.code}: {e.message}"
|
||||
except Exception as e:
|
||||
info.error = f"E_INTERNAL: {type(e).__name__}: {e}"
|
||||
return info
|
||||
|
||||
|
||||
def load_cohort(
|
||||
samples: Sequence[tuple[str, str]], n_loci: int, tmp: Path, *, baf: bool, scores: bool, workers: int
|
||||
) -> tuple[CohortStore, list[SampleInfo]]:
|
||||
"""Read GTCs (sample_id, path) into a :class:`CohortStore` with a process pool."""
|
||||
store = CohortStore(tmp, len(samples), n_loci, baf=baf, scores=scores)
|
||||
store.gt.flush()
|
||||
tasks = [(i, sid, p, str(tmp), baf, scores) for i, (sid, p) in enumerate(samples)]
|
||||
if workers <= 1 or len(tasks) <= 1:
|
||||
infos = [_load_one(t) for t in tasks]
|
||||
else:
|
||||
with ProcessPoolExecutor(workers, mp_context=mp.get_context("spawn")) as ex:
|
||||
infos = list(ex.map(_load_one, tasks, chunksize=4))
|
||||
return store, infos
|
||||
|
||||
|
||||
class SiteView:
|
||||
"""(sites × samples) row access onto a (samples × loci) matrix, optionally restricted to some samples."""
|
||||
|
||||
def __init__(self, m: npt.NDArray[Any], samples: npt.NDArray[np.intp] | None = None) -> None:
|
||||
self.m, self.samples = m, samples
|
||||
self.shape = (m.shape[1], m.shape[0] if samples is None else len(samples))
|
||||
|
||||
def __getitem__(self, rows: Any) -> npt.NDArray[Any]:
|
||||
block = np.asarray(self.m[:, rows])
|
||||
return (block if self.samples is None else block[self.samples]).T
|
||||
|
||||
|
||||
def _blocks(n: int, size: int) -> Iterator[slice]:
|
||||
for s in range(0, n, size):
|
||||
yield slice(s, min(n, s + size))
|
||||
|
||||
|
||||
def genotype_counts(
|
||||
gt: npt.NDArray[np.uint8], samples: npt.NDArray[np.intp] | None = None, block: int = 65536
|
||||
) -> npt.NDArray[np.int64]:
|
||||
"""(loci × 4) counts of no-call/AA/AB/BB over (a subset of) samples, streamed in locus blocks."""
|
||||
out = np.zeros((gt.shape[1], 4), np.int64)
|
||||
for sl in _blocks(gt.shape[1], block):
|
||||
g = np.asarray(gt[:, sl])
|
||||
if samples is not None:
|
||||
g = g[samples]
|
||||
for k in range(4):
|
||||
out[sl, k] = (g == k).sum(axis=0)
|
||||
return out
|
||||
|
||||
|
||||
def het_rates(gt: npt.NDArray[np.uint8], loci: npt.NDArray[np.intp], block: int = 65536) -> npt.NDArray[np.float64]:
|
||||
het = np.zeros(gt.shape[0], np.int64)
|
||||
called = np.zeros(gt.shape[0], np.int64)
|
||||
for s in range(0, len(loci), block):
|
||||
g = np.asarray(gt[:, loci[s : s + block]])
|
||||
het += (g == 2).sum(axis=1)
|
||||
called += (g > 0).sum(axis=1)
|
||||
with np.errstate(invalid="ignore", divide="ignore"):
|
||||
return np.where(called > 0, het / called, np.nan)
|
||||
|
||||
|
||||
# ================================================================================================ pedigree
|
||||
@dataclass
|
||||
class Pedigree:
|
||||
parents: dict[str, list[str]] # child sample_id -> genotyped parents (0–2)
|
||||
replicates: dict[str, str] # sample_id -> declared replicate sample_id
|
||||
|
||||
def expected(self, a: str, b: str) -> str:
|
||||
if b in self.parents.get(a, []) or a in self.parents.get(b, []):
|
||||
return "parent-offspring"
|
||||
if self.replicates.get(a) == b or self.replicates.get(b) == a:
|
||||
return "replicate"
|
||||
pa, pb = set(self.parents.get(a, [])), set(self.parents.get(b, []))
|
||||
if pa & pb:
|
||||
return "sibling"
|
||||
return ""
|
||||
|
||||
|
||||
def pedigree_from_sheet(sheet: SampleSheet | None, sample_ids: Sequence[str], rep: Reporter) -> Pedigree:
|
||||
"""IN-018: Parent1/Parent2/Replicate values name Sample_ID, Sample_Name or sample_id (in that order)."""
|
||||
ped = Pedigree({}, {})
|
||||
if sheet is None:
|
||||
return ped
|
||||
ids = set(sample_ids)
|
||||
by_col: dict[str, dict[str, str]] = {"sample_id_col": {}, "name": {}}
|
||||
rows = [r for r in sheet.rows if r.sample_id in ids]
|
||||
cols = {fold(c): c for c in sheet.columns}
|
||||
for r in rows:
|
||||
if (c := cols.get(fold("Sample_ID"))) and r.values.get(c):
|
||||
by_col["sample_id_col"].setdefault(r.values[c], r.sample_id)
|
||||
if r.sample_name:
|
||||
by_col["name"].setdefault(r.sample_name, r.sample_id)
|
||||
|
||||
def resolve(v: str) -> str | None:
|
||||
v = v.strip()
|
||||
if v in ("", "0", "NA", "na", "-"):
|
||||
return None
|
||||
for table in (by_col["sample_id_col"], by_col["name"]):
|
||||
if v in table:
|
||||
return table[v]
|
||||
if v in ids:
|
||||
return v
|
||||
rep.warn("W_QC_PEDIGREE", f"{v!r} does not name a sample in this set")
|
||||
return None
|
||||
|
||||
for r in rows:
|
||||
for key in ("Parent1", "Parent2"):
|
||||
if (c := cols.get(fold(key))) and (p := resolve(r.values.get(c, ""))) and p != r.sample_id:
|
||||
ped.parents.setdefault(r.sample_id, []).append(p)
|
||||
if (c := cols.get(fold("Replicate"))) and (p := resolve(r.values.get(c, ""))) and p != r.sample_id:
|
||||
ped.replicates[r.sample_id] = p
|
||||
return ped
|
||||
|
||||
|
||||
# ================================================================================================ cohort QC
|
||||
@dataclass
|
||||
class CohortQC:
|
||||
het_rate: npt.NDArray[np.float64]
|
||||
het_z: list[float]
|
||||
contamination: ContaminationResult | None = None
|
||||
pairs: list[dict[str, Any]] = field(default_factory=list)
|
||||
duplicates: dict[str, list[str]] = field(default_factory=dict)
|
||||
unexpected: set[str] = field(default_factory=set)
|
||||
mendel: list[dict[str, Any]] = field(default_factory=list)
|
||||
|
||||
|
||||
def cohort_qc(
|
||||
store: CohortStore,
|
||||
infos: Sequence[SampleInfo],
|
||||
manifest: Manifest,
|
||||
rep: Reporter,
|
||||
*,
|
||||
pedigree: Pedigree | None = None,
|
||||
contamination: bool = True,
|
||||
relatedness: bool = True,
|
||||
all_pairs: bool = False,
|
||||
median: str = "reference",
|
||||
workers: int = 1,
|
||||
) -> CohortQC:
|
||||
ids = [s.sample_id for s in infos]
|
||||
ok = np.array([not s.error for s in infos])
|
||||
auto = np.flatnonzero(autosome_mask(manifest.chroms) & ~manifest.intensity_only)
|
||||
het = het_rates(store.gt, auto)
|
||||
het[~ok] = np.nan
|
||||
res = CohortQC(het_rate=het, het_z=het_zscores(het.tolist()))
|
||||
if contamination and store.baf is not None:
|
||||
usable = ok & np.array([s.has_baf for s in infos])
|
||||
for s in infos:
|
||||
if not s.error and not s.has_baf:
|
||||
rep.warn("W_QC_NO_BAF", f"{s.sample_id}: GTC has no BAF; contamination not estimated")
|
||||
if usable.sum() < MIN_CONTAMINATION_SAMPLES:
|
||||
rep.warn(
|
||||
"W_QC_TOO_FEW",
|
||||
f"contamination needs >= {MIN_CONTAMINATION_SAMPLES} samples with BAF (have {int(usable.sum())})",
|
||||
)
|
||||
else:
|
||||
rows_idx = np.flatnonzero(usable)
|
||||
sel = None if usable.all() else rows_idx
|
||||
c = genotype_counts(store.gt, sel)[:, 1:].astype(np.float64)
|
||||
an = 2 * c.sum(axis=1)
|
||||
site_af = np.full(manifest.num_loci, np.nan)
|
||||
with np.errstate(invalid="ignore", divide="ignore"):
|
||||
af = np.where(an > 0, (c[:, 1] + 2 * c[:, 2]) / an, np.nan)
|
||||
site_af[auto] = af[auto] # contamination sites: autosomes only
|
||||
with _pool(workers) as ex:
|
||||
cr = baf_regress(
|
||||
SiteView(store.gt, sel), # type: ignore[arg-type]
|
||||
SiteView(store.baf, sel), # type: ignore[arg-type]
|
||||
site_af,
|
||||
median=median,
|
||||
map_fn=ex.map if ex is not None else map,
|
||||
)
|
||||
est = np.full(len(ids), np.nan)
|
||||
nh = np.zeros(len(ids), np.int64)
|
||||
est[rows_idx], nh[rows_idx] = cr.estimate, cr.n_hom
|
||||
res.contamination = ContaminationResult(est, nh, cr.n_sites)
|
||||
if relatedness and ok.sum() >= 2:
|
||||
usable_loci = np.zeros(manifest.num_loci, np.bool_)
|
||||
usable_loci[auto] = True
|
||||
g_ok = np.flatnonzero(ok)
|
||||
sites = select_sites(genotype_counts(store.gt, None if ok.all() else g_ok), usable_loci)
|
||||
G = SiteView(store.gt, None if ok.all() else g_ok)[sites]
|
||||
ps = pair_stats(G)
|
||||
ped = pedigree or Pedigree({}, {})
|
||||
sub = [ids[i] for i in g_ok]
|
||||
for a in range(len(sub)):
|
||||
for b in range(a + 1, len(sub)):
|
||||
phi, conc = float(ps.kinship[a, b]), float(ps.concordance[a, b])
|
||||
exp = ped.expected(sub[a], sub[b])
|
||||
dup = conc >= DUPLICATE_CONCORDANCE
|
||||
if dup:
|
||||
res.duplicates.setdefault(sub[a], []).append(sub[b])
|
||||
res.duplicates.setdefault(sub[b], []).append(sub[a])
|
||||
rel = math.isfinite(phi) and phi > DEGREE_CUTS[2][0] # second degree or closer
|
||||
if rel and not exp:
|
||||
res.unexpected.update((sub[a], sub[b]))
|
||||
if all_pairs or dup or exp or (math.isfinite(phi) and phi > DEGREE_CUTS[3][0]):
|
||||
res.pairs.append(
|
||||
{
|
||||
"sample_a": sub[a],
|
||||
"sample_b": sub[b],
|
||||
"n_sites": int(ps.n_sites[a, b]),
|
||||
"kinship": phi,
|
||||
"concordance": conc,
|
||||
"degree": degree(phi) if math.isfinite(phi) else "",
|
||||
"expected": exp,
|
||||
}
|
||||
)
|
||||
for child, parents in (ped.parents or {}).items():
|
||||
if child not in ids:
|
||||
continue
|
||||
ci = ids.index(child)
|
||||
prow = [np.asarray(store.gt[ids.index(p)])[auto] for p in parents[:2]]
|
||||
e, n = mendel_errors(
|
||||
np.asarray(store.gt[ci])[auto], prow[0] if prow else None, prow[1] if len(prow) > 1 else None
|
||||
)
|
||||
res.mendel.append(
|
||||
{
|
||||
"child": child,
|
||||
"parent1": parents[0] if parents else "",
|
||||
"parent2": parents[1] if len(parents) > 1 else "",
|
||||
"n_errors": e,
|
||||
"n_informative": n,
|
||||
"error_rate": e / n if n else math.nan,
|
||||
}
|
||||
)
|
||||
return res
|
||||
|
||||
|
||||
class _NullPool:
|
||||
def __enter__(self) -> None:
|
||||
return None
|
||||
|
||||
def __exit__(self, *a: object) -> None:
|
||||
return None
|
||||
|
||||
|
||||
def _pool(workers: int) -> Any:
|
||||
if workers <= 1:
|
||||
return _NullPool()
|
||||
return ProcessPoolExecutor(workers, mp_context=mp.get_context("spawn"))
|
||||
|
||||
|
||||
def default_tmp(out: Path) -> Path:
|
||||
base = Path(os.environ.get("SIMURG_TMP", out))
|
||||
base.mkdir(parents=True, exist_ok=True)
|
||||
return Path(tempfile.mkdtemp(prefix=".simurg_cohort_", dir=base))
|
||||
|
||||
|
||||
def contamination_rows(ids: Sequence[str], cr: ContaminationResult | None) -> dict[str, dict[str, Any]]:
|
||||
if cr is None:
|
||||
return {}
|
||||
return {sid: {"contamination": float(cr.estimate[i]), "n_hom_sites": int(cr.n_hom[i])} for i, sid in enumerate(ids)}
|
||||
|
||||
|
||||
def status_from_thresholds(th: Thresholds) -> list[dict[str, Any]]:
|
||||
return th.as_records()
|
||||
|
||||
|
||||
# ================================================================================================ locus summary
|
||||
def write_locus_summary(
|
||||
path: Path, store: CohortStore, manifest: Manifest, egt: Any, samples: npt.NDArray[np.intp] | None = None
|
||||
) -> None:
|
||||
"""OUT-072, streamed in locus blocks. ``egt`` is a parsed cluster file (records joined by locus name)."""
|
||||
from simurg.core.atomic import atomic_open
|
||||
from simurg.qc.locus_summary import COLUMNS, summarize_block
|
||||
|
||||
assert store.score is not None
|
||||
idx = egt.name_index()
|
||||
rec = np.array([idx.get(nm, -1) for nm in manifest.names], dtype=np.int64)
|
||||
has = rec >= 0
|
||||
safe = np.where(has, rec, 0)
|
||||
|
||||
def take(a: npt.NDArray[Any]) -> npt.NDArray[np.float64]:
|
||||
v = np.asarray(a)[safe].astype(np.float64)
|
||||
v[~has] = np.nan
|
||||
return v
|
||||
|
||||
cl = {
|
||||
"locus_id": np.where(has, np.asarray(egt.addresses)[safe], 0),
|
||||
"gentrain_score": take(egt.gentrain_score),
|
||||
"cluster_sep": take(egt.cluster_sep),
|
||||
}
|
||||
for j, gname in enumerate(("aa", "ab", "bb")):
|
||||
cl[f"{gname}_t_mean"] = take(egt.theta_mean[:, j])
|
||||
cl[f"{gname}_t_sd"] = take(egt.theta_dev[:, j])
|
||||
cl[f"{gname}_r_mean"] = take(egt.r_mean[:, j])
|
||||
cl[f"{gname}_r_sd"] = take(egt.r_dev[:, j])
|
||||
strands = [s if s in ("+", "-") else "U" for s in (manifest.ref_strands or [""] * manifest.num_loci)]
|
||||
|
||||
def f(v: Any) -> str:
|
||||
if isinstance(v, float | np.floating):
|
||||
return "NaN" if math.isnan(v) else f"{float(v):.8g}"
|
||||
return str(v)
|
||||
|
||||
with atomic_open(path, "w", encoding="utf-8") as raw:
|
||||
fh = cast(IO[str], raw)
|
||||
fh.write("\t".join(COLUMNS) + "\n")
|
||||
for sl in _blocks(manifest.num_loci, 65536):
|
||||
g = np.asarray(store.gt[:, sl])
|
||||
s = np.asarray(store.score[:, sl])
|
||||
if samples is not None:
|
||||
g, s = g[samples], s[samples]
|
||||
stats = summarize_block(g, s)
|
||||
for k in range(sl.stop - sl.start):
|
||||
i = sl.start + k
|
||||
row = {"locus_name": manifest.names[i], "plus_minus_strand": strands[i]}
|
||||
vals = [row[c] if c in row else (cl[c][i] if c in cl else stats[c][k]) for c in COLUMNS]
|
||||
fh.write("\t".join(f(v) for v in vals) + "\n")
|
||||
|
||||
|
||||
# ================================================================================================ GTC-based commands
|
||||
SAMPLE_QC_COLUMNS = [
|
||||
"sample_id",
|
||||
"sample_name",
|
||||
"status",
|
||||
"call_rate",
|
||||
"autosomal_call_rate",
|
||||
"logr_dev",
|
||||
"sex_est",
|
||||
"sex_reported",
|
||||
"sex_check",
|
||||
"het_rate",
|
||||
"het_z",
|
||||
"contamination",
|
||||
"n_hom_sites",
|
||||
"functional_status",
|
||||
"flags",
|
||||
]
|
||||
RELATEDNESS_COLUMNS = ["sample_a", "sample_b", "n_sites", "kinship", "concordance", "degree", "expected"]
|
||||
MENDEL_COLUMNS = ["child", "parent1", "parent2", "n_errors", "n_informative", "error_rate"]
|
||||
|
||||
|
||||
@dataclass
|
||||
class QcCohortOptions:
|
||||
"""Options shared by ``qc samples | contamination | relatedness | loci`` (CLI-025)."""
|
||||
|
||||
command: str
|
||||
out: Path
|
||||
bpm: Path
|
||||
gtcs: Path | None = None
|
||||
sheet: Path | None = None
|
||||
egt: Path | None = None
|
||||
threads: int = 0
|
||||
force: bool = False
|
||||
all_pairs: bool = False
|
||||
pedigree_from_sheet: bool = True
|
||||
median: str = "reference"
|
||||
thresholds: Thresholds | None = None
|
||||
argv: list[str] = field(default_factory=list)
|
||||
|
||||
|
||||
def _gtc_samples(opts: QcCohortOptions, rep: Reporter) -> tuple[list[tuple[str, str]], SampleSheet | None]:
|
||||
"""(sample_id, path) in sheet order (sheet given) or discovery order (IN-030, IN-016)."""
|
||||
from simurg.formats.discovery import discover_gtcs
|
||||
from simurg.formats.samplesheet import parse_samplesheet
|
||||
|
||||
sheet = parse_samplesheet(opts.sheet) if opts.sheet else None
|
||||
root = opts.gtcs or (sheet.root_folder if sheet else None)
|
||||
if root is None:
|
||||
raise SimurgError("E_NO_INPUT", "give --gtcs or a sample sheet with a RootFolder")
|
||||
found = discover_gtcs(root)
|
||||
if sheet is None:
|
||||
return [(k, str(v)) for k, v in found.items()], None
|
||||
out = []
|
||||
for r in sheet.rows:
|
||||
if r.sample_id in found:
|
||||
out.append((r.sample_id, str(found[r.sample_id])))
|
||||
else:
|
||||
rep.warn("W_QC_UNMATCHED_SAMPLES", f"{r.sample_id}: no GTC under {root}", sample_id=r.sample_id)
|
||||
if not out:
|
||||
raise SimurgError("E_NO_INPUT", f"none of the sheet's samples has a GTC under {root}")
|
||||
return out, sheet
|
||||
|
||||
|
||||
def run_qc_cohort(opts: QcCohortOptions, rep: Reporter) -> int:
|
||||
"""``simurg qc samples | contamination | relatedness | loci`` (CLI-025)."""
|
||||
from simurg.formats.bpm import read_bpm
|
||||
from simurg.pipeline.inputs import check_paths
|
||||
from simurg.qc.status import functional_status, sample_flags
|
||||
from simurg.qc.thresholds import resolve_thresholds
|
||||
|
||||
rm = RunManifest(command=f"qc {opts.command}", argv=opts.argv)
|
||||
rep.run_id = rm.run_id
|
||||
check_paths([(opts.bpm, "BPM"), (opts.gtcs, "GTC folder"), (opts.sheet, "sample sheet"), (opts.egt, "EGT")])
|
||||
if opts.command == "loci" and opts.egt is None:
|
||||
raise SimurgError("E_NO_INPUT", "qc loci needs --egt (cluster statistics, OUT-072)")
|
||||
names = {
|
||||
"samples": ["sample_qc.tsv", "contamination.tsv", "relatedness.tsv"],
|
||||
"contamination": ["contamination.tsv"],
|
||||
"relatedness": ["relatedness.tsv"],
|
||||
"loci": ["locus_summary.tsv"],
|
||||
}[opts.command]
|
||||
outputs = [opts.out / n for n in names]
|
||||
check_outputs_absent(outputs, force=opts.force)
|
||||
samples, sheet = _gtc_samples(opts, rep)
|
||||
rep.info(f"reading manifest {opts.bpm}")
|
||||
manifest = read_bpm(opts.bpm)
|
||||
workers = max(1, opts.threads or (os.cpu_count() or 1))
|
||||
opts.out.mkdir(parents=True, exist_ok=True)
|
||||
tmp = default_tmp(opts.out)
|
||||
want_baf = opts.command in ("samples", "contamination")
|
||||
rep.info(f"loading {len(samples)} GTC(s) with {workers} worker(s)")
|
||||
store, infos = load_cohort(
|
||||
samples, manifest.num_loci, tmp, baf=want_baf, scores=opts.command == "loci", workers=workers
|
||||
)
|
||||
try:
|
||||
for s in infos:
|
||||
if s.error:
|
||||
rep.error(s.error.split(":", 1)[0], f"{s.sample_id}: {s.error}", sample_id=s.sample_id)
|
||||
ids = [s.sample_id for s in infos]
|
||||
if opts.command == "loci":
|
||||
egt_path = opts.egt
|
||||
assert egt_path is not None
|
||||
from simurg.formats.egt import read_egt
|
||||
|
||||
ok = np.flatnonzero([not s.error for s in infos])
|
||||
write_locus_summary(opts.out / "locus_summary.tsv", store, manifest, read_egt(egt_path), ok)
|
||||
else:
|
||||
ped = pedigree_from_sheet(sheet, ids, rep) if opts.pedigree_from_sheet else None
|
||||
res = cohort_qc(
|
||||
store,
|
||||
infos,
|
||||
manifest,
|
||||
rep,
|
||||
pedigree=ped,
|
||||
contamination=opts.command in ("samples", "contamination"),
|
||||
relatedness=opts.command in ("samples", "relatedness"),
|
||||
all_pairs=opts.all_pairs,
|
||||
median=opts.median,
|
||||
workers=workers,
|
||||
)
|
||||
crow = contamination_rows(ids, res.contamination)
|
||||
if opts.command in ("samples", "contamination"):
|
||||
write_tsv(
|
||||
opts.out / "contamination.tsv",
|
||||
["sample_id", "contamination", "n_hom_sites"],
|
||||
[{"sample_id": sid, **crow.get(sid, {})} for sid in ids],
|
||||
)
|
||||
if opts.command in ("samples", "relatedness"):
|
||||
write_tsv(opts.out / "relatedness.tsv", RELATEDNESS_COLUMNS, res.pairs)
|
||||
if res.mendel:
|
||||
write_tsv(opts.out / "mendel.tsv", MENDEL_COLUMNS, res.mendel)
|
||||
outputs.append(opts.out / "mendel.tsv")
|
||||
if opts.command == "samples":
|
||||
th = opts.thresholds or resolve_thresholds()
|
||||
auto = autosome_mask(manifest.chroms) & ~manifest.intensity_only
|
||||
auto_idx = np.flatnonzero(auto)
|
||||
sex_rep = {r.sample_id: r.reported_sex for r in sheet.rows} if sheet else {}
|
||||
rows = []
|
||||
for i, s in enumerate(infos):
|
||||
g = np.asarray(store.gt[i])[auto_idx]
|
||||
row: dict[str, Any] = {
|
||||
"sample_id": s.sample_id,
|
||||
"sample_name": s.sample_name,
|
||||
"status": "FAILED" if s.error else "OK",
|
||||
"call_rate": s.call_rate,
|
||||
"autosomal_call_rate": float((g > 0).mean()) if len(g) and not s.error else math.nan,
|
||||
"logr_dev": s.logr_dev,
|
||||
"sex_est": s.sex_est,
|
||||
"sex_reported": sex_rep.get(s.sample_id, "U"),
|
||||
"het_rate": float(res.het_rate[i]),
|
||||
"het_z": res.het_z[i],
|
||||
**crow.get(s.sample_id, {}),
|
||||
}
|
||||
known = {"M", "F"}
|
||||
row["sex_check"] = (
|
||||
("match" if row["sex_est"] == row["sex_reported"] else "mismatch")
|
||||
if row["sex_est"] in known and row["sex_reported"] in known
|
||||
else "unknown"
|
||||
)
|
||||
row["functional_status"] = functional_status(row, th)
|
||||
row["flags"] = ";".join(
|
||||
sample_flags(
|
||||
row,
|
||||
th,
|
||||
het_z=res.het_z[i],
|
||||
duplicates=res.duplicates.get(s.sample_id, []),
|
||||
unexpected_relatedness=s.sample_id in res.unexpected,
|
||||
)
|
||||
)
|
||||
rows.append(row)
|
||||
write_tsv(opts.out / "sample_qc.tsv", SAMPLE_QC_COLUMNS, rows)
|
||||
finally:
|
||||
store.close()
|
||||
for p in outputs:
|
||||
if p.exists():
|
||||
rm.add_output(p, f"simurg.{p.stem}/1")
|
||||
rm.summary = {"samples": len(infos), "failed": sum(bool(s.error) for s in infos)}
|
||||
rm.status = "completed"
|
||||
rm.write(opts.out / f"run_manifest.qc_{opts.command}.json")
|
||||
failed = sum(bool(s.error) for s in infos)
|
||||
rep.info(f"done: qc {opts.command} on {len(infos) - failed}/{len(infos)} sample(s); outputs in {opts.out}")
|
||||
if failed == 0:
|
||||
return ExitCode.OK
|
||||
return ExitCode.PARTIAL if failed < len(infos) else ExitCode.ALL_FAILED
|
||||
|
||||
|
||||
# ================================================================================================ suggest-thresholds
|
||||
def _read_table(path: Path) -> list[dict[str, str]]:
|
||||
import csv
|
||||
|
||||
text = path.read_text(encoding="utf-8-sig")
|
||||
delim = "\t" if "\t" in text.split("\n", 1)[0] else ","
|
||||
return list(csv.DictReader(text.splitlines(), delimiter=delim))
|
||||
|
||||
|
||||
def dataset_values(folder: Path) -> dict[str, list[float]]:
|
||||
"""Per-key metric values of one QC dataset folder (QC-040 file contract)."""
|
||||
from simurg.qc.thresholds import KEY_TO_COLUMN
|
||||
|
||||
def num(v: str | None) -> float:
|
||||
try:
|
||||
return float(v) if v not in (None, "") else math.nan
|
||||
except ValueError:
|
||||
return math.nan
|
||||
|
||||
out: dict[str, list[float]] = {}
|
||||
qc = folder / CONTROLS_QC
|
||||
if not qc.exists():
|
||||
raise SimurgError(
|
||||
"E_QC_DATASET_INCOMPLETE", f"Dataset {folder.name} is missing required file(s): {CONTROLS_QC}"
|
||||
)
|
||||
rows = _read_table(qc)
|
||||
for key, col in KEY_TO_COLUMN.items():
|
||||
if rows and col in rows[0]:
|
||||
out[key] = [num(r.get(col)) for r in rows]
|
||||
summary = folder / "sample_summary.tsv"
|
||||
if summary.exists():
|
||||
srows = [r for r in _read_table(summary) if r.get("status", "OK") == "OK"]
|
||||
out["callRate"] = [num(r.get("autosomal_call_rate")) for r in srows]
|
||||
out["logRDev"] = [num(r.get("logr_dev")) for r in srows]
|
||||
return out
|
||||
|
||||
|
||||
def run_suggest_thresholds(
|
||||
data: Sequence[Path], out_yaml: Path, th: Thresholds, rep: Reporter, *, k: float = 4.0, min_samples: int = 30
|
||||
) -> int:
|
||||
from simurg.qc.suggest import render_yaml, suggest
|
||||
|
||||
if not data:
|
||||
raise SimurgError("E_QC_NO_DATASETS", "No valid dataset folders detected.")
|
||||
merged: dict[str, list[float]] = {}
|
||||
for d in data:
|
||||
if not d.exists():
|
||||
raise SimurgError("E_QC_NO_DATASETS", f"Input data folder does not exist: {d}", path=str(d))
|
||||
for key, vals in dataset_values(d).items():
|
||||
merged.setdefault(key, []).extend(vals)
|
||||
sugs = suggest(merged, th, k=k, min_samples=min_samples)
|
||||
for s in sugs:
|
||||
if s.note:
|
||||
rep.warn("W_QC_TOO_FEW", f"{s.key}: {s.note}; left unset")
|
||||
out_yaml.parent.mkdir(parents=True, exist_ok=True)
|
||||
atomic_write_text(out_yaml, render_yaml(sugs, k))
|
||||
write_tsv(
|
||||
out_yaml.with_suffix(".tsv"),
|
||||
["key", "n", "centre", "spread", "value", "pct_flagged_current", "pct_flagged_suggested", "note"],
|
||||
[s.__dict__ for s in sugs],
|
||||
)
|
||||
rep.info(f"suggested thresholds written to {out_yaml}")
|
||||
return ExitCode.OK
|
||||
Reference in new issue
Block a user