hg38_62celltypes, hg38_33celltypes: .msdref version 3, with a confusion
Wanding Zhou
ef6b77a
Both human deconvolution references now carry an MSDCONF1 trailer, which
`deconv --group-threshold` reads to report near-collinear classes under one
joined label rather than splitting mass between columns the data cannot
separate. The mouse reference has carried one since v11; these are the two
most readers fetch.
The M/U block of each is byte-identical to the version-2 artifact it
replaces. `cmp -l` over the old file's length reports exactly ONE differing
byte in each -- offset 9, the version word, 2 to 3 -- and the added bytes are
the appended trailer (20 + 4*n^2 + len(prov)): 15,624 for the 62-class,
4,584 for the 33-class. So this is not a rebuilt model that happens to agree;
it is the same model plus a trailer.
Measured on held-out data, because a shipped build uses everything and has
none left. The 62-class twin was rebuilt FROM CELLS -- pools over folds 1-9
of the Zhou 2025 hash split, fold 0 held out as mixture components, with
Hepatocyte and Granulocyte held out by donor sin…
26 days ago
mm10_41celltypes.msdref: the first mouse deconvolution reference
Wanding Zhou
43b7a27
A pure addition -- every v10 file ships unchanged, nothing withdrawn.
41 Liu 2021 mouse brain major types over the 6,765,325 CpGs that could ever
separate some class subset (30.9% of mm10), packed as uint16 M/U from
pseudobulks of all 54,447 cells with no fold held back. Until now every
.msdref was human and mouse shipped only classifiers plus the wg decoder.
It is the first version-3 artifact: it carries a validation confusion
trailer, which `deconv --group-threshold` reads to report near-collinear
classes under one joined label (CA3/DG-po) rather than splitting mass between
columns the data cannot separate. The trailer was measured on a
fold-held-out twin, since an all-cells build has no held-out cells to measure
with. A binary older than that format refuses the file rather than misreading
it, so the registry pin moves only with a YAME release that can read it.
Validated before staging: all 41 pools self-deconvolve to 100.0%; on P21
spatial DMT mouse brain (2,235 in-tissue pixels)…
28 days ago
SHA256SUMS: hg38_wg and both lites rehashed, two withdrawn files dropped
Wanding Zhou
76413e9
models: new hg38_wg decoder, footprint lites, routing trees withdrawn
Wanding Zhou
e17499d
hg38_wg.updecx is retrained on a Loyfer-67 reference pooled from training
cells only plus the bank's 100 resolvers, to the step cap at patience 400.
It fixes the hepatocyte ELF3/MIR200C undercall (dip MAE 0.309 -> 0.047 at
29k input) and is better than the previous model at both loci at every
coverage.
hg38_celltype_lite.clfx and mm10_brain_lite.clfx are the footprint-picked
lites: resolver pairs chosen by hard-block footprint rather than the
retired gate, a fifth the size of the full banks.
hg38_celltype.clfx and mm10_celltype_brain.clfx are withdrawn, superseded
by the bank classifiers published at v9. They stay in the history and in
every earlier tag.
mm10_wg.updecx is unchanged: its tissue-reference pilot was inconclusive,
so the mouse decoder is carried forward and revisited later.
README: a pointer to the methscope docs, not a second copy of them
Wanding Zhou
52ee8e2
The model table, the bank vs routing-tree classifiers, the deconvolution
references and the violation rule now live on the Models and Reference
tabs of zhou-lab.github.io/methscope-cli, generated from the binary and
from YAME's asset catalogue. This card keeps the front-matter, one
paragraph, the methscope fetch lines and the link, so there is one place
to maintain.
README: bank classifiers, held-out mouse accuracy, classify argument order
Wanding Zhou
2f8ebda
- rows and a size table for the four bank models; _full supersedes the
routing trees for new work
- mm10_brain_{full,lite} scored on 49,113 Liu 2021 cells outside the
training cap: 0.9660 / 0.9482 micro
- every classify example now reads model-first, matching the binary
SHA256SUMS: add the four bank classifiers
Wanding Zhou
25352fb
bank classifiers: hg38_celltype_{full,lite}, mm10_brain_{full,lite}
Wanding Zhou
61e38dd
Pooled-bank cell-type classifiers (one booster, no routing) from the
20260904 bank-full recipe. _full carries a calibrated resolver for every
class pair; _lite only for pairs no LCA binstring separates. Human 63
classes, mouse 41 brain classes.
README: hg38_62celltypes.msdref is the primary deconvolution reference
hg38_celltype is now a violation-framework model with no fitted parameters: it reads each class's expected methylation state off the MRMP binstrings and calls the class the query contradicts least, weighting patterns by sqrt(n_cpg). Confidence is the margin to the runner-up, which collapses when no class fits.
External test 0.842 / macro-F1 0.853 at >=100k covered CpGs. Colorectal cancer single cells call 100% gastrointestinal from liver and lymph metastases alike, and the sink that sent globally demethylated tumours to Acinar.(Pancreas) drops from 30.0% of TCGA calls to 2.3%.
Classifier bundles are renamed .ubjx -> .clfx: .ubjx named a payload format rather than a role and had already stopped being accurate (hg38_sex carried a plain-text linear spec, no UBJSON). Bundles are detected by MSBNDL1 magic, never by extension, so .ubjx stays accepted. The previous filenames remain available at --revision v5.
mm10_celltype.ubjx is unchanged, pending a Liu 2021 rebuild.
