Patient‐aware stress testing of transcriptomic representation transportability across bone‐metastasis cohorts

Abstract Bone‐metastasis atlases reveal recurrent remodeling, but internally stable representations may fail across cohorts or controls. We evaluated a patient‐aware workflow separating discovery stability, technical reconstruction, biological comparison, and spatial organization. Discovery included 107,886 cells from 42 prostate and renal samples from 18 patients, with 11 complete anatomical triplets. Under fixed historical labels, paired compositional sensitivities retained nine myeloid and T/natural‐killer state directions; alternative annotations changed some directions and projection coverage. A historically derived 81‐gene list failed its archived external directional criterion, and a three‐ecotype representation failed its external criteria. A 26‐feature multiblock Axis1, comprising broad‐lineage, myeloid and T/natural‐killer compositions, explained 54.17% of discovery variance. Post‐review comparisons showed almost identical ordering to unbalanced PC1 (Spearman rho, 0.996) and similar patient‐held‐out reconstruction errors (0.489461 vs. 0.490098). Involved‐minus‐distal marrow differences were smaller than solid metastatic‐tissue‒marrow differences; patient‐and‐anatomy‐adjusted associations with mitochondrial percentage and predicted doublet removal left quality control‐related variation unresolved. In GSE266330, 41 bone‐metastasis patients and five healthy‐marrow donors separated on Axis1 but shared no author‐defined sequencing label. This source‐confounded contrast could not establish a disease effect independently of source or replace the unmet matched‐normal‐bone rule (one of two patients positive). Four linked mouse sections retained positive spatial contrasts under mapping sensitivities, although some contrast magnitudes were sensitive and independent animal counts were unknown. These findings support an inspectable representation stress‐testing workflow and an anatomically organized sample‐context coordinate, not a demonstrated methodological advantage, universal bone‐metastasis state, mechanism, or predictor.

Authors

Institutions

Publication Details

Journal
View
Published
2026-11-16
DOI
https://doi.org/10.1002/viw2.70225
Primary Topic
Single-cell and spatial transcriptomics
Type
article
Field-Weighted Citation Impact
0.00

Funders

Controls
|||
ALL TIME
JAN
FEB
MAR
APR
MAY
JUN
JUL
AUG
SEP
OCT
NOV
article

Patient‐aware stress testing of transcriptomic representation transportability across bone‐metastasis cohorts

Hanzhou Wang, Shengji Yu, Zhenguo Zhao, Zhangfu Li et al.
1 links
View
Single-cell and spatial transcriptomics
article

Patient‐aware stress testing of transcriptomic representation transportability across bone‐metastasis cohorts

Hanzhou Wang, Shengji Yu, Zhenguo Zhao, Zhangfu Li, Baicheng Yang, Shiqing Liu
article en

Abstract

Abstract Bone‐metastasis atlases reveal recurrent remodeling, but internally stable representations may fail across cohorts or controls. We evaluated a patient‐aware workflow separating discovery stability, technical reconstruction, biological comparison, and spatial organization. Discovery included 107,886 cells from 42 prostate and renal samples from 18 patients, with 11 complete anatomical triplets. Under fixed historical labels, paired compositional sensitivities retained nine myeloid and T/natural‐killer state directions; alternative annotations changed some directions and projection coverage. A historically derived 81‐gene list failed its archived external directional criterion, and a three‐ecotype representation failed its external criteria. A 26‐feature multiblock Axis1, comprising broad‐lineage, myeloid and T/natural‐killer compositions, explained 54.17% of discovery variance. Post‐review comparisons showed almost identical ordering to unbalanced PC1 (Spearman rho, 0.996) and similar patient‐held‐out reconstruction errors (0.489461 vs. 0.490098). Involved‐minus‐distal marrow differences were smaller than solid metastatic‐tissue‒marrow differences; patient‐and‐anatomy‐adjusted associations with mitochondrial percentage and predicted doublet removal left quality control‐related variation unresolved. In GSE266330, 41 bone‐metastasis patients and five healthy‐marrow donors separated on Axis1 but shared no author‐defined sequencing label. This source‐confounded contrast could not establish a disease effect independently of source or replace the unmet matched‐normal‐bone rule (one of two patients positive). Four linked mouse sections retained positive spatial contrasts under mapping sensitivities, although some contrast magnitudes were sensitive and independent animal counts were unknown. These findings support an inspectable representation stress‐testing workflow and an anatomically organized sample‐context coordinate, not a demonstrated methodological advantage, universal bone‐metastasis state, mechanism, or predictor.

View
Capital Medical University (CN), Chinese Academy of Medical Sciences & Peking Union Medical College (CN), Beijing Chao-Yang Hospital, Capital Medical University (CN), National Cancer Center (US), Cancer Hospital of Chinese Academy of Medical Sciences (CN)
Natural Science Foundation of Beijing Municipality
Good health and well-being
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