Pre-registration: whose proportion is it? Denominator substitution in spatial transcriptomics cell-type composition (D study, M0)

What this is. A pre-registration for a methodological audit of how "cell type proportions" are computed and reported in spatial transcriptomics, deposited before any result existed. All 60 census judgement cells were empty at freeze time. The question. In spatial transcriptomics with multi-cell spots, the number published as a "cell type proportion" is frequently a fraction of spots, not of cells. The anchor paper (Clevenger et al., Cell 2026, doi:10.1016/j.cell.2026.03.006) illustrates the mechanism precisely: each spot is assigned the cell type with the highest inferred proportion (an argmax hard call), and the reported proportions are computed over all spots. Three distinct distortions follow, and this study measures them separately rather than as one composite factor: (i) the inferred proportion is a share of transcripts, (ii) the argmax collapses a compositional vector to a label, (iii) one spot contains several cells, with the cells-per-spot ratio differing by organ. What is frozen. Criteria J1-J4 (thresholds for the anchor audit and for the simulation) and census items P1-P4 (what counts as "reported a proportion" and "stated the denominator") are fixed verbatim. The only design parameter that changed after v2.0 is the census sample size (40 -> 60; stratum quota 10/20/10 -> 15/30/15), changed before any paper was read. Wilson half-widths motivating it: +/-14.8 pp at n=40 versus +/-12.3 pp at n=60, against a 15 pp decision step. See PREREG_DECISION_LOG.md. Verified before freezing (gate results, included with quotes and arithmetic). G1 PASS: the spot denominator is stated in the figure legend and implemented in the authors' released code (Cell_Count = n(), i.e. the variable named as a cell count holds spot rows). G2 FAIL: the per-spot proportion matrix is not public — the Zenodo record is code only and the proportion tables are referenced by internal drive paths — so the simulation route is used and the anchor's own argmax is not recomputed. G3 FAIL: cells per spot is not measurable per organ; the paper reports no spot dimension, and its "5 million nuclei" appears in two mutually inconsistent readings (2.03-fold apart). Deliberately absent. No result of any kind: no census coding, no simulation output, no organ-level coefficient of variation. The census sampling frame is the 276 eligible papers from a 366-hit PubMed query (2024-2026), stratified into three journals tiers; the drawn list, the frame and the sampling script are included so the list can be regenerated exactly. Package contents and per-file SHA256 are in MANIFEST.json; reproduction commands are in README.md.

Authors

Publication Details

Journal
Zenodo (CERN European Organization for Nuclear Research)
Published
2026-11-21
DOI
https://doi.org/10.5281/zenodo.24409774
Primary Topic
Single-cell and spatial transcriptomics
Type
preprint
Controls
|||
ALL TIME
JAN
FEB
MAR
APR
MAY
JUN
JUL
AUG
SEP
OCT
NOV
preprint

Pre-registration: whose proportion is it? Denominator substitution in spatial transcriptomics cell-type composition (D study, M0)

Chaohua Tang
Zenodo (CERN European Organization for Nuclear Research)
Single-cell and spatial transcriptomics
preprint

Pre-registration: whose proportion is it? Denominator substitution in spatial transcriptomics cell-type composition (D study, M0)

Chaohua Tang
preprint en

Abstract

What this is. A pre-registration for a methodological audit of how "cell type proportions" are computed and reported in spatial transcriptomics, deposited before any result existed. All 60 census judgement cells were empty at freeze time. The question. In spatial transcriptomics with multi-cell spots, the number published as a "cell type proportion" is frequently a fraction of spots, not of cells. The anchor paper (Clevenger et al., Cell 2026, doi:10.1016/j.cell.2026.03.006) illustrates the mechanism precisely: each spot is assigned the cell type with the highest inferred proportion (an argmax hard call), and the reported proportions are computed over all spots. Three distinct distortions follow, and this study measures them separately rather than as one composite factor: (i) the inferred proportion is a share of transcripts, (ii) the argmax collapses a compositional vector to a label, (iii) one spot contains several cells, with the cells-per-spot ratio differing by organ. What is frozen. Criteria J1-J4 (thresholds for the anchor audit and for the simulation) and census items P1-P4 (what counts as "reported a proportion" and "stated the denominator") are fixed verbatim. The only design parameter that changed after v2.0 is the census sample size (40 -> 60; stratum quota 10/20/10 -> 15/30/15), changed before any paper was read. Wilson half-widths motivating it: +/-14.8 pp at n=40 versus +/-12.3 pp at n=60, against a 15 pp decision step. See PREREG_DECISION_LOG.md. Verified before freezing (gate results, included with quotes and arithmetic). G1 PASS: the spot denominator is stated in the figure legend and implemented in the authors' released code (Cell_Count = n(), i.e. the variable named as a cell count holds spot rows). G2 FAIL: the per-spot proportion matrix is not public — the Zenodo record is code only and the proportion tables are referenced by internal drive paths — so the simulation route is used and the anchor's own argmax is not recomputed. G3 FAIL: cells per spot is not measurable per organ; the paper reports no spot dimension, and its "5 million nuclei" appears in two mutually inconsistent readings (2.03-fold apart). Deliberately absent. No result of any kind: no census coding, no simulation output, no organ-level coefficient of variation. The census sampling frame is the 276 eligible papers from a 366-hit PubMed query (2024-2026), stratified into three journals tiers; the drawn list, the frame and the sampling script are included so the list can be regenerated exactly. Package contents and per-file SHA256 are in MANIFEST.json; reproduction commands are in README.md.

Zenodo (CERN European Organization for Nuclear Research)
Single-cell and spatial transcriptomics
AI Navigator

Ask Laika to Summarize, Analyze, and Connect papers live on the map.

Summarize Papers & Methodologies

Extract key findings, datasets, and comparative methods across publications.

Benchmark Rankings & Visual Analytics

Rank top research institutions, authors, funders, topics, and journals by Field-Weighted Citation Impact (FWCI) and paper volume with instant charts.

Connect Distant Disciplines

Bridge topological clusters on the map to find hidden collaborative intersections.

Laika access coming soon