Selected author evidence guide

Cancer omics: subtype transfer, prognosis and network-state evidence

How should a new cancer cohort be used for molecular subtyping, prognostic scoring, network-state analysis or immunotherapy-response assessment?

A source-linked worked comparison and reusable reporting worksheet for choosing analyses in a new cancer cohort. Six examples separate molecular subtyping, survival prognosis, stage-associated network states and treatment-response evidence, using reviewed study facts and identified methodological guidance.

Start with the intended output, cohort and observation unit. Then identify the model/template/network object, its implementation and updating, independent evaluation data, endpoint/time and response measurements. Keep the separate analyses of one publication in separate worksheet rows and cite the original DOI for study findings.

Which analysis and evidence should be carried into a new external cancer cohort?

为新的外部癌症队列选择分析时,应迁移什么对象,怎样报告其验证证据?

Source-linked methodological synthesis: Define the output first. HCCcrds provides a molecular-subtype and cohort-role example; the thyroid study separates a stage-associated DNB state from a compact subtype classifier; LPAS contributes a retrospective survival-score example; the TNBC work distinguishes molecular-state profiling, a mixed-cancer response-model pipeline and target-cancer proxy analyses. ConsensusClusterPlus can assess resampling stability, and SubMap can assess molecular correspondence across independently defined subclasses. For prediction models, record unchanged implementation versus updating/refitting. For treatment claims, identify actual outcome-linked response data separately from molecular proxies. The worksheet makes these choices explicit before a new-cohort analysis. [1] [2] [3] [4] [5] [6] [7]

先确定所需输出。HCCcrds提供分子亚型及队列用途的实例;甲状腺研究区分分期相关DNB网络状态与紧凑亚型分类器;LPAS提供回顾性生存风险评分实例;TNBC研究则需分开分子状态、混合癌种响应模型和目标癌种中的预测proxy。ConsensusClusterPlus可评估重采样下的聚类稳定性,SubMap可评估独立数据中既定亚类的分子对应。对预测模型,记录是否原样应用、更新或重拟合;对治疗主张,另列真实结局关联响应数据。工作表用于在新队列分析前明确这些选择。

Evaluation checklist

Choose and report an analysis for a new external cancer cohort

Use one row per analysis, cohort, model version and endpoint. Record the output before selecting a method. Split subtype assignment, survival prognosis, network state and treatment response into their actual roles. NR means unresolved in this extraction, not that the original paper omitted a detail. For a new cohort, report the artifact, preprocessing, every adaptation/refit, independence, endpoint/time and available treatment labels; use the blank CSV to retain those decisions.

Download completed worksheet (CSV) · Download blank worksheet (CSV)

Choose and report an analysis for a new external cancer cohort
Reported analysisOutput and research taskApplication / refit statusTime reference and endpointEvaluation actually supportedImmunotherapy evidence roleClaim to reportSources
HCCcrds external subtype evaluationMolecular subtype assignment, not an individual survival-risk modelExternal expression/classification evaluation is reported. Whether every preprocessing/template element was frozen is NR here; do not relabel fresh clustering as unchanged classifier applicationExpression-defined CS-H/CS-L/CS-M labels; not measured physiological circadian phaseRetrospective cross-dataset subtype evaluation. Reuse the linked cohort-role worksheet rather than copying its accession inventorySeparate CIBERSORT/IPS analyses are inferred immune properties, not observed ICI responseReport external subtype assignment/reproducibility with the actual implementation status; keep subtype labels separate from survival prediction[1] [9]
LPAS retrospective LUAD survival-score evaluationPrognostic expression score associated with overall survivalRetrospective external dataset evaluation is reported. Applying an unchanged score and re-estimating coefficients/cutoffs in the new cohort are different analysesOverall survival; retain time origin, horizon, censoring and the exact metric definition in each analysis rowRetrospective external survival evaluation; no prospective LPAS-guided clinical decision validationTIDE/SubMap and heterogeneous treatment-cohort associations are separate from an observed response study in the intended new LUAD populationReport the actual score implementation and retrospective prognosis task. Do not rank its survival AUC against subtype accuracy or ICI-response AUC[4]
ISG15-related classifier in mixed-cancer ICI source dataRetrospective treated-response classification, distinct from TNBC molecular-state profilingThe reported evaluation is within a pooled source-data pipeline, not independent treated-TNBC external validationICI response label in the source treatment cohorts; label definition and pretreatment timing must be verified for reuseModest retrospective response discrimination in other-cancer source data; source cohorts being external to the TNBC atlas does not make the internal split independent TNBC validationResponse-labeled treated source data are a different evidence role from TIDE or a predicted-response group in an untreated/unclear-treatment cohortName the pooled/internal evaluation and its cancer context. Do not report a validated TNBC treatment-selection test[2]
ISG15 predicted-response groups in TNBC-related application analysesPredicted group, immune-signature or survival association; not observed ICI response in treated TNBCApplication/association layer; do not treat a new grouping or retrospective survival contrast as independent response-model validationPredicted-response group, immune estimate and overall survival are separate endpointsTarget-TNBC computational proxy analyses and a nonsignificant TCGA TNBC survival comparison; source-data response-model evaluation is recorded in its separate rowTIDE dysfunction/exclusion and predicted-response labels are computational proxies; they are not measured clinical responseState predicted-response/immune-association results at their actual scope, without claiming treatment benefit[2]
Thyroid DNB stage-associated network stateCross-sectional stage-associated network instability, not longitudinal individual transition predictionReconstructing a network/state calculation in new data is a distinct analysis from applying the separate miniPC subtype classifierReported Stage II instability peak across stage groups; no observed within-person time-to-transition clockStage-associated molecular evidence. It does not establish when a new patient's disease will deteriorateIPS/CMap are computational immune/drug hypotheses; no observed immunotherapy-response trial is establishedReport a stage-associated network-state analysis and its grouping assumptions; do not call it prospective dynamic-risk validation[3]
Thyroid miniPC external molecular-subtype classificationCompact classifier of consensus subtype labels, separate from the DNB-state analysis and TCPSLevel survival associationsExternally reported subtype-classifier evaluation. Training a new network or changing the label-generating clustering is a separate fitSubtype assignment; neither a treatment-response label nor an individual transition timeRetrospective expression-template/classifier evaluation; internal cluster stability is not external clinical validationSeparate IPS/CMap results remain hypotheses rather than measured patient responseReport subtype classification at its actual evaluation scope. Do not call miniPC a validated survival model[3]

Decisions from the comparison

  • Define the output and unit: Choose molecular subtype, prognostic risk, a group/network state or observed response before evaluating a new dataset. State the observation unit, reference labels and the particular analysis supported by each cohort. [1] [2] [3] [4]
  • Use stability and correspondence for their positive tasks: ConsensusClusterPlus supplies repeated-subsampling stability and membership evidence. SubMap tests correspondence between previously defined subclasses across independently generated datasets. The reporting inference is that a stability plot or molecular association matrix alone does not supply clinical outcomes that were not measured; the original SubMap paper also includes separate prediction and survival analyses. [5] [6]
  • Separate prediction-model application from updating: For a prognostic or response predictor, identify the original model/feature/weight artifact and document any recalibration, coefficient/cutoff refit or retraining, with its development and assessment data. This prediction-model guidance is not used to judge unsupervised clustering as though it were a fixed clinical risk model. [7] [4] [2]
  • Keep stage-associated states separate from patient time: The thyroid DNB comparison is cross-sectional and stage-associated. Reconstructing its network/state output and applying miniPC subtype labels are distinct analyses; neither supplies an observed within-patient transition time merely from stage membership. [3]
  • Identify the response evidence actually present: Record treatment, pretreatment sampling, cancer and measured response when available. Distinguish mixed-cancer source-data evaluation from a new cancer-specific response cohort. IPS/TIDE, predicted-response groups and SubMap molecular correspondence retain their actual computational roles. [2] [1] [3] [4] [6]
  • Reuse the existing HCC cohort-role table: For HCC Fig.5A subtype versus Fig.5B survival membership, follow the existing worked example and CSV. Do not recreate its dataset list in this resource. Keep cross-study metric ranking unset when tasks, endpoint/time definitions or implementations differ. [1] [9] [7]

How the selected papers contribute

Choose the analysis by its output, evaluation unit and source evidence
Paper / designPopulationEndpoint / horizonValidationReported evaluationOriginal source
Integrated-omics analysis defines subtypes of hepatocellular carcinoma based on circadian rhythm
Retrospective single-cell/bulk integration and HCC subtype evaluation
HCC expression cohorts, with separate circadian reference experimentsMolecular subtype assignment; subtype-survival comparisons in separately designated datasetsExternal expression/classification and retrospective outcome comparisons; use the linked cohort-role table to select the relevant analysis setSubtype and survival analyses have different cohort roles; do not add the overlapping Fig.5A/B sets into an independent-cohort total10.1016/j.joim.2025.06.003 [1] [9]
Resolving tumor microenvironment heterogeneity to forecast immunotherapy response in triple-negative breast cancer through multi-scale analysis
TNBC multi-modal association plus secondary mixed-cancer ICI-response classifier development
BRCA1-defined TNBC specimens and separately pooled treated cohorts from other cancersTME molecular states; response classification in the mixed-cancer source data; separate target-cohort predicted-group associationsWithin-pooled-source development/evaluation; the study explicitly lacks an independent treated-TNBC response cohortRetrospective KNN discrimination is distinct from TIDE estimates, predicted-response groups and survival associations10.3389/fonc.2025.1538574 [2]
Optimized Dynamic Network Biomarker Deciphers a High-Resolution Heterogeneity Within Thyroid Cancer Molecular Subtypes
Stage-associated network analysis, molecular subtyping and compact subtype-classifier evaluation
Public thyroid-cancer expression cohorts; single-cell and tissue/cell follow-up remain separate evidence layersDNB network instability, TCPSLevel and miniPC subtype assignment are distinct outputsCross-sectional stage-associated DNB evidence; separate retrospective external template/classifier evaluationA stage-associated instability peak is not an individual transition-time forecast; miniPC classification is not a survival-benefit evaluation10.1002/mdr2.70004 [3]
Deciphering lung adenocarcinoma evolution: Integrative single-cell genomics identifies the prognostic lung progression associated signature
Retrospective transcriptomic survival modeling with separate cell-functional and immune analyses
TCGA development and additional GEO LUAD survival series; other analytic layers are kept separateLPAS-associated overall survival; immune/treatment associations have their own definitionsAdditional retrospective survival-series evaluation; any new-cohort refitting must be reported separately from unchanged-score applicationUse horizon-specific survival evidence and the actual implementation; immune matching or cell perturbation is a different evaluation10.1111/jcmm.18408 [4] [7]

Use these rows for applicability and reporting, rather than to rank unlike performance measures. The original methods establish resampling stability and cross-dataset subclass correspondence as positive capabilities. Requiring measured outcome-linked evidence for a clinical or response claim is this resource's task-based inference, not a statement that those method authors reject clinical applications. External-validation guidance is applied here to prediction models, not as a universal clustering criterion. [5] [6] [7] [1] [2] [3] [4]

Research applications

Sources and locators

  1. Integrated-omics analysis defines subtypes of hepatocellular carcinoma based on circadian rhythm · DOI: 10.1016/j.joim.2025.06.003 — Author-deposited Results 3.3-3.7, Figures 4-6. Detailed figure locators use the ahead-of-print copy; final typeset figure equivalence has not been rechecked.
  2. Resolving tumor microenvironment heterogeneity to forecast immunotherapy response in triple-negative breast cancer through multi-scale analysis · DOI: 10.3389/fonc.2025.1538574 — PMC12401915 Methods s2_10, Results s3_9, Figure 8D-K and Discussion s4.
  3. Optimized Dynamic Network Biomarker Deciphers a High-Resolution Heterogeneity Within Thyroid Cancer Molecular Subtypes · DOI: 10.1002/mdr2.70004 — Methods 2.2/2.6-2.8; Results 3.1/3.3-3.4; Figures 1/3-4. Immune hypotheses: Methods 2.10/2.20, Results 3.5/3.9.
  4. Deciphering lung adenocarcinoma evolution: Integrative single-cell genomics identifies the prognostic lung progression associated signature · DOI: 10.1111/jcmm.18408 — PMC11149493 Methods 2.8 (jcmm18408-sec-0010), Results 3.3-3.4 (sec-0026/0027), Figures 4-5; immune/treatment analyses Results 3.5/3.7/3.9, Figures 6/8/10.
  5. ConsensusClusterPlus: a class discovery tool with confidence assessments and item tracking · DOI: 10.1093/bioinformatics/btq170 — Introduction SEC1/p[1]; Software Features SEC2/p[1]; Example Application SEC3/p[1]. Repeated subsampling, consensus stability and item tracking; same-input software comparison is described as an integrity check.
  6. Subclass Mapping: Identifying Common Subtypes in Independent Disease Data Sets · DOI: 10.1371/journal.pone.0001195 — Materials and Methods s4b1 candidate subclasses; s4b2-s4b4 mutual marker enrichment and subclass-association matrix. The paper separately describes prediction (s4c) and survival (s4d) analyses.
  7. Evaluation of clinical prediction models (part 2): how to undertake an external validation study · DOI: 10.1136/bmj-2023-074820 — What do we mean by external validation?; Steps 2-4; Table 1 and performance/calibration sections. Applied here to clinical prediction-model implementation and updating, not as a rule for unsupervised clustering.
  8. Omics-based molecular classifications empowering in precision oncology · DOI: 10.1007/s13402-023-00912-8 — Sections 3.1.4-3.1.5, 5.1-5.2 and 6. Narrative molecular-classification and translation context; later primary examples in this author-maintained resource are separately sourced.
  9. Worked example: select HCC cohorts by analysis purpose · DOI: 10.1016/j.joim.2025.06.003 — Author-maintained cohort-role worksheet and linked data/hcc-cohort-roles.csv, derived from HCCcrds Results 3.5/Figure 5A-B for that same study.
  10. Artificial Intelligence Hybrid Survival Assessment System for Robot-Assisted Proctectomy: A Retrospective Cohort Study · DOI: 10.1200/po.24.00089 — Results: Extension to External Cohort From the RAP-AIscore; Figure 5A-J. Independent retrospective SYEMH one-, two- and three-year overall-survival evaluation, as recorded in the reviewed study evidence.
  11. Analysis of Local Recurrence After Robotic-Assisted Total Mesorectal Excision (ALRITE): An International, Multicentre, Retrospective Cohort · DOI: 10.3390/cancers17060992 — Methods 2.1 and 2.5; Results 3.4; Discussion. Three-year local recurrence, approximately 10% within-dataset holdout and the reported 0.76 AUC with oversampling in training/testing and validation.
  12. Can Surgical Approach and Postoperative Factors Impact Survival in Rectal Cancer? Robotic Versus Laparoscopic Insights · DOI: 10.1002/cam4.71453 — Official abstract: retrospective robot-assisted versus laparoscopic rectal-cancer surgery and multivariable Cox overall/disease-free survival associations. This comparison uses the abstract, not the unread original full text.
Different research tasks and evidence types
Research taskReviewed designReviewed evidence type
HCC molecular-subtype assignment and cohort-specific outcome comparisonsPublic single-cell and bulk transcriptomic integration with retrospective subtype evaluationHuman molecular association and cross-dataset classification analysis
TNBC TME associations and cross-cancer ICI-response modelingRetrospective single-cell TNBC group comparison, spatial reference mapping, bulk expression/network prioritization, paired in-house tissue proteomics and secondary immunotherapy-response classifier development.Human multi-modal observational molecular association with a mixed-cancer retrospective treatment-response model; no experimental causal gene perturbation.
Thyroid stage-associated DNB states and separate miniPC subtype classificationPublic bulk/single-cell analysis with human tissue assessment and cell experimentsRetrospective molecular subtyping with experimental candidate follow-up
LUAD LPAS survival-score evaluation and separate immune/treatment associationsSecondary integration of human LUAD single-cell and bulk expression cohorts with retrospective survival modeling, computational immune/drug-response inference and in-vitro PSMB1 perturbation.Human transcriptomic association, retrospective model evaluation and LUAD cell-line functional assays.

HCC molecular-subtype assignment and cohort-specific outcome comparisons

Integrated-omics analysis defines subtypes of hepatocellular carcinoma based on circadian rhythm

Authors: Xiao-jie Li; Le Chang; Yang Mi; Ge Zhang; Shan-shan Zhu; Yue-xiao Zhang; Hao-yu Wang; Yi-shuang Lu; Ye-xuan Ping; Peng-yuan Zheng; Xia Xue

Year: 2025 · Journal: Journal of Integrative Medicine · Publication type: Article

Author evidence page · Formal publication DOI · BibTeX · RIS · CSL JSON

Which HCC subtypes were associated with clinical outcomes?

TCGA-LIHC analysis identified CS-H, CS-L and CS-M, with poorer outcomes associated with CS-H. TCGA-LIHC n: 369; Subtypes: CS-H; CS-L; CS-M. Association does not establish that circadian disruption causes liver cancer or constitute a validated clinical decision rule.

  • KF3 — Author-deposited full text: Results 3.3-3.4; Fig. 4

Were external classification and survival analyses performed in identical cohorts?

External subtype and survival analyses used overlapping but different cohort sets. Fig. 5A datasets: ICGC_LIRI; GSE14520; GSE54236; GSE104580; Fig. 5B datasets: ICGC_LIRI; GSE14520; GSE54236; GSE27150. Do not claim that all five cohorts validated survival, or add the two four-cohort analyses as eight independent cohorts.

  • KF4 — Author-deposited full text: Results 3.5; Fig. 5 legend

Do the immune analyses demonstrate immunotherapy benefit?

Subtypes differed in proliferation/metabolic programs and inferred immune characteristics. Analyses: Expression enrichment; CIBERSORT; IPS. IPS and immune deconvolution are not an observed immunotherapy-response trial.

  • KF5 — Author-deposited full text: Results 3.6-3.7; Fig. 6

Reviewed study profile

profile type
multicohort_omics
study design
Public single-cell and bulk transcriptomic integration with retrospective subtype evaluation
evidence type
Human molecular association and cross-dataset classification analysis
population
HCC expression cohorts, with separate circadian reference experiments
primary endpoint
HCCcrds-associated molecular subtype structure
secondary endpoint
Retrospective outcome associations and computational biological annotations
data modalities
  • single-cell RNA sequencing
  • bulk transcriptomics
  • co-expression analysis
  • LRA clustering/NTP
  • immune deconvolution
external validation
true
scale metrics
  • label
    TCGA-LIHC
    value
    369 samples in the reported analysis
  • label
    GSE149614 analyzed cells
    value
    11,383 cells
  • label
    External datasets
    value
    5 datasets with different analytical roles
counting note
Cell counts, candidate genes, HCC cohort samples and separate circadian reference experiments are different units. No complete unique-participant total is inferred. The five external datasets have different analytical uses: the four Fig. 5A and four Fig. 5B sets overlap but are not identical and must not be added as eight independent cohorts. External validation means retrospective cross-dataset expression/classification evaluation, not prospective clinical utility or direct physiological rhythm measurement in HCC patients.

Complete reviewed evidence scope

supports
  • Single-cell modules supplied the HCCcrds gene set.
  • The score was evaluated in separate clock, feeding and sleep-reference data.
  • TCGA-LIHC analysis identified CS-H, CS-L and CS-M, with poorer outcomes associated with CS-H.
  • External subtype and survival analyses used overlapping but different cohort sets.
  • Subtypes differed in proliferation/metabolic programs and inferred immune characteristics.
  • NNMT and MARCKSL1 were among subtype-associated candidate genes.
does not establish
  • Prospective diagnostic or prognostic clinical utility.
  • Direct individual measurement of physiological circadian phase or amplitude.
  • Circadian disruption as a demonstrated cause of cancer.
  • Clinical benefit from score-guided treatment.
  • IPS as measured response in patients receiving immunotherapy.
  • Therapeutic validation of the candidate genes.
  • Completed page-by-page equivalence checking between the author-deposited text and final typeset version.
  • Unprovided AUCs, hazard ratios, participant totals or additional omics modalities.

Complete reviewed limitations

TNBC TME associations and cross-cancer ICI-response modeling

Resolving tumor microenvironment heterogeneity to forecast immunotherapy response in triple-negative breast cancer through multi-scale analysis

Authors: Shihao Sun; Shuang Chen; Kaiyuan Li; Ge Zhang; Nan Wang; Yijia Xu; Xinxing Wang; Jiangrui Chi; Lin Li; Yi Sun

Year: 2025 · Journal: Frontiers in Oncology · Publication type: Article

Author evidence page · Formal publication DOI · BibTeX · RIS · CSL JSON

How well did the ICI response model discriminate?

The selected KNN classifier reported AUC 0.60 (95% CI 0.51–0.69) for validation and 0.61 (95% CI 0.51–0.71) for testing in the external-treatment source data.

  • KF4 — PMC12401915: Methods Reconstruction of the ICI response predictor (s2_10); Results Construction of an ISG15-driven predictive system (s3_9); Figure 8D–G. Paper text inconsistency: Methods mentions 13 cohorts (n=829); Results mentions 12.

Was the ICI model validated in treated TNBC patients?

No. The Discussion explicitly identifies the lack of validation in ICI-treated TNBC cohorts; it is inappropriate to claim established TNBC treatment selection.

  • KF4 — PMC12401915: Methods Reconstruction of the ICI response predictor (s2_10); Results Construction of an ISG15-driven predictive system (s3_9); Figure 8D–G. Paper text inconsistency: Methods mentions 13 cohorts (n=829); Results mentions 12.
  • KF5 — PMC12401915: Results Construction of an ISG15-driven predictive system for ICI response (s3_9); Figure 8H–K; Discussion (s4), paragraph explicitly stating no ICI-treated TNBC validation.

How many ICI cohorts did the authors use?

The article is inconsistent: Methods describes 13 pretreatment ICI cohorts (829 records) whereas Results mentions 12 cohorts. Four cohorts pooled as 661 records for the principal development split can be identified; the other cohort count is not silently harmonized.

  • KF4 — PMC12401915: Methods Reconstruction of the ICI response predictor (s2_10); Results Construction of an ISG15-driven predictive system (s3_9); Figure 8D–G. Paper text inconsistency: Methods mentions 13 cohorts (n=829); Results mentions 12.

Reviewed study profile

profile type
multicohort_omics
study design
Retrospective single-cell TNBC group comparison, spatial reference mapping, bulk expression/network prioritization, paired in-house tissue proteomics and secondary immunotherapy-response classifier development.
evidence type
Human multi-modal observational molecular association with a mixed-cancer retrospective treatment-response model; no experimental causal gene perturbation.
population
Eight BRCA1-defined TNBC single-cell specimens; four BRCA1-WT TNBC spatial samples; independently collected 9 paired breast tumor/adjacent tissues; mixed-cancer ICI-response expression cohorts for model training/testing.
primary endpoint
Differences in single-cell and spatially inferred tumor immune/stromal programs across BRCA1 status.
secondary endpoint
ISG15/myCAF association and retrospective response-model discrimination in cross-cancer ICI cohorts.
data modalities
  • TNBC single-cell RNA sequencing
  • Spatial RNA expression reference mapping
  • Bulk expression, coexpression and pathway analysis
  • Paired tissue label-free proteomics
  • Cross-cancer ICI transcriptomic response modeling
external validation
false
scale metrics
  • label
    TNBC single-cell donors
    value
    8 specimens: 4 BRCA1-mutant and 4 BRCA1 wild-type
  • label
    Cells after QC
    value
    32,386; 18,182 BRCA1-mutant and 14,204 BRCA1 wild-type
  • label
    Spatial specimens
    value
    4 BRCA1-WT TNBC spatial samples; no matched BRCA1-MT spatial series in this analysis
  • label
    Paired proteomics
    value
    9 tumor samples and 9 corresponding para-tumor specimens; nine pairs, not 18 independent patients
  • label
    ICI model pooled subset
    value
    Four external treatment cohorts pooled to 661 records and divided 80:20 internally
  • label
    KNN classifier discrimination
    value
    Reported validation AUC 0.60 (95% CI 0.51–0.69); testing AUC 0.61 (95% CI 0.51–0.71)
counting note
Cell profiles, donor tumor specimens, spatial samples, paired proteomic tissues and external cross-cancer treated cohorts are distinct counting units and may involve overlaps or subsets. The article inconsistently describes 13 ICI pre-treatment cohorts (829 records) in Methods versus 12 in Results; no unsupported sum or cohort count is promoted.

Complete reviewed evidence scope

supports
  • Cell-state and CAF-associated transcriptomic differences between the eight sampled BRCA1 groups.
  • Spatially associated ISG15 and myCAF signals within the four BRCA1-WT spatial specimens.
  • Retrospective mixed-cancer ICI response model discrimination near AUC 0.60–0.61 under the described analysis.
does not establish
  • Causal BRCA1 control of CAF differentiation or a verified ISG15 immune-response mechanism from the multi-omics associations.
  • Independently validated or clinically useful ICI prediction specifically for treated TNBC patients.
  • Excellent predictive discrimination, proven chemotherapy/ICI treatment efficacy or a validated treatment-selection cut-off.
  • A statistically significant TCGA TNBC overall-survival contrast (reported P=0.182).
  • A verified universal count of distinct participants across cell, spatial, paired protein and ICI cohorts.

Complete reviewed limitations

Thyroid stage-associated DNB states and separate miniPC subtype classification

Optimized Dynamic Network Biomarker Deciphers a High-Resolution Heterogeneity Within Thyroid Cancer Molecular Subtypes

Authors: Haonan Zhang; Ge Zhang; Pengyuan Xu; Fengyi Yu; Liwen Li; Runzhi Huang; Pengpeng Zhang; Kaisaierjiang Kadier; Yuanyuan Wang; Qiqi Gu; Yalei Ding; Tianshu Gu; Hao Chi; Shiqian Zhang; Ruhao Wu; Yudi Xu; Shanshan Zhu; Honglin Zheng; Taiqi Zhao; Qi He; Xinguang Qiu

Year: 2025 · Journal: Med Research · Publication type: Article

Author evidence page · Formal publication DOI · BibTeX · RIS · CSL JSON

What does the Stage II DNB signal establish?

Stage II had the highest reported network instability signal. DNB genes: 36. Cross-sectional stage comparisons do not establish longitudinal prediction of each patient's future deterioration time.

  • KF1 — Methods 2.2; Results 3.1; Fig. 1

How were the thyroid molecular subtypes evaluated externally?

The ensemble analysis yielded three molecular subtypes. Candidate combinations: 20; Selected combination: CV-PAM; Reported external NTP datasets: 6. Methods 2.6 prints GSE333630 whereas other locations use GSE33630. These are not silently rewritten as a fully verified accession inventory.

  • KF3 — Methods 2.6-2.7; Results 3.3; Fig. 3

Does miniPC classify subtypes or directly prove improved survival?

miniPC classified the consensus subtype labels using a compact gene set. Candidate genes: 296; Final genes: 12; Final classifier: Neural network; External classifier cohorts: 5. This is subtype classification, not prospective individual survival-benefit validation. Unverified figure-only accuracy or AUC values are not reconstructed.

  • KF4 — Methods 2.8; Results 3.4; Fig. 4

Were CMap-nominated compounds shown to benefit patients?

IPS and CMap supplied candidate immune/drug-response hypotheses. Evidence: Computational scoring and signature matching. Predicted compounds are not proven effective treatments. IPS is not an observed outcome in patients receiving immunotherapy.

  • KF7 — Methods 2.10 and 2.20; Results 3.5 and 3.9; Figure 7G-I; Discussion

Reviewed study profile

profile type
multicohort_omics
study design
Public bulk/single-cell analysis with human tissue assessment and cell experiments
evidence type
Retrospective molecular subtyping with experimental candidate follow-up
population
Thyroid-cancer datasets and tissue specimens, with separate cultured-cell experiments
primary endpoint
DNB-associated expression structure and molecular subtype assignment
secondary endpoint
Retrospective outcome associations and ASPH-related experimental readouts
data modalities
  • bulk transcriptomics
  • single-cell RNA sequencing
  • network/classifier analysis
  • human immunohistochemistry
  • in-vitro gene perturbation
external validation
true
scale metrics
  • label
    GEO microarray cohorts
    value
    21 cohorts, plus TCGA
  • label
    TCGA count in Methods
    value
    544
  • label
    TCGA count in GRN analysis
    value
    484
  • label
    scRNA cells in Methods
    value
    17,709
  • label
    scRNA cells in Results
    value
    17,518
  • label
    Reported single-cell donors
    value
    1
  • label
    Human tissue specimens
    value
    48 tumor and 48 adjacent specimens
counting note
TCGA counts of 544 in Methods and 484 in GRN analysis describe different reported analytical stages; no reason for excluding the numerical difference is inferred. Methods reports 17,709 cells and Results 17,518, without a reconstructed QC-exclusion process. One reported donor is not a large participant cohort; 48 tumor and 48 adjacent specimens do not mean 96 independent patients. Cohorts, cells, paired specimens and cultured-cell experiments are not added into a unique-participant total. External validation means retrospective expression-template/classifier evaluation, not prospective clinical utility.

Complete reviewed evidence scope

supports
  • Stage II had the highest reported network instability signal.
  • TCPSLevel was associated with clinical and outcome differences.
  • The ensemble analysis yielded three molecular subtypes.
  • miniPC classified the consensus subtype labels using a compact gene set.
  • Single-cell analysis localized expression patterns across cellular compartments.
  • ASPH was evaluated in human tissues and through TPC-1 silencing.
  • IPS and CMap supplied candidate immune/drug-response hypotheses.
does not establish
  • Longitudinal prediction of an individual patient's transition time.
  • Clinical intervention benefit.
  • miniPC as an established survival model rather than a subtype classifier.
  • Cell counts as patient counts or paired-specimen counts as independent participants.
  • Complete demonstration of direct ASCL2 regulation or binding.
  • Clinical efficacy of CMap-nominated compounds.
  • Classification of this study as a physiological circadian-rhythm experiment.
  • Resolution of all Methods/Results counts or accession differences.

Complete reviewed limitations

LUAD LPAS survival-score evaluation and separate immune/treatment associations

Deciphering lung adenocarcinoma evolution: Integrative single-cell genomics identifies the prognostic lung progression associated signature

Authors: Pengpeng Zhang; Zijun Yang; Zuo Liu; Ge Zhang; Lianmin Zhang; Zhenfa Zhang; Jun Fan

Year: 2024 · Journal: Journal of Cellular and Molecular Medicine · Publication type: Article

Author evidence page · Formal publication DOI · BibTeX · RIS · CSL JSON

What is the nine-gene LPAS model?

LPAS is a retrospective expression-based LUAD survival score containing nine selected genes, including PSMB1. It was developed from TCGA data and evaluated in five additional GEO survival series.

  • KF3 — PMC11149493: Methods §2.8 (jcmm18408-sec-0010); Results §§3.3–3.4 (jcmm18408-sec-0026, jcmm18408-sec-0027); Figures 4B–G and 5A–F.

Was LPAS externally evaluated?

Five additional GEO LUAD survival datasets were evaluated, with differing reported AUCs and follow-up horizons. This is retrospective external dataset evaluation, not a prospective treatment-impact trial.

  • KF3 — PMC11149493: Methods §2.8 (jcmm18408-sec-0010); Results §§3.3–3.4 (jcmm18408-sec-0026, jcmm18408-sec-0027); Figures 4B–G and 5A–F.

Can low LPAS identify patients who will benefit from immunotherapy?

The article reports in-silico immune scores and SubMap/TIDE comparisons plus treatment-cohort associations, but no prospective LPAS-guided treatment study establishes patient-level clinical benefit.

  • KF4 — PMC11149493: Results §§3.5, 3.7, 3.9 (jcmm18408-sec-0028, jcmm18408-sec-0030, jcmm18408-sec-0032); Figures 6A–C, 8D–G, 10A–L.

Reviewed study profile

profile type
multicohort_omics
study design
Secondary integration of human LUAD single-cell and bulk expression cohorts with retrospective survival modeling, computational immune/drug-response inference and in-vitro PSMB1 perturbation.
evidence type
Human transcriptomic association, retrospective model evaluation and LUAD cell-line functional assays.
population
Early pathological LUAD tissue samples across AIS, MIA and IAC; independent retrospective GEO/TCGA LUAD survival cohorts and additional tumor-treatment series from heterogeneous cancers.
primary endpoint
Association of the LPAS expression score with overall survival in retrospective LUAD series.
secondary endpoint
Single-cell state contrasts across pathological stages; immune/stromal signature associations; computational immunotherapy/drug-response hypotheses; PSMB1 knockdown cell phenotypes.
data modalities
  • Two publicly available LUAD scRNA-seq series
  • TCGA and GEO transcriptomic and clinical survival data
  • Exploratory external treatment-response expression datasets
  • Local LUAD RNA sequencing
  • A549 and H1299 siRNA perturbation assays
external validation
true
scale metrics
  • label
    Single-cell specimen scale
    value
    GSE189357 and GSE150938: 4 AIS, 8 MIA and 9 IAC specimens (21 reported across pathology groups)
  • label
    Independent survival-series list
    value
    GSE13213 (119), GSE29016 (39), GSE30219 (86), GSE31210 (227) and GSE68465 (443) as reported analyzed series
  • label
    Model discovery
    value
    TCGA-LUAD; cohort n not inferred from unspecified filtering
  • label
    Local stage-associated RNA sequencing
    value
    14 LUAD surgical specimens (reported)
  • label
    LPAS signature
    value
    9 genes
counting note
The 21 scRNA-seq specimens, 14 local RNA-seq specimens, TCGA cases, five survival cohorts and mixed-cancer immunotherapy datasets represent distinct analytic layers; neither cross-dataset independence nor a synthetic unique-patient total is inferred.

Complete reviewed evidence scope

supports
  • Cluster 6 and LPAS show reproducible associations in the reported transcriptomic cohorts.
  • Higher LPAS corresponds to less favorable retrospective LUAD survival in the evaluated cohorts.
  • PSMB1 silencing reduces selected proliferation and movement phenotypes in A549/H1299 assays.
does not establish
  • A directly traced longitudinal path from an AIS cell to an IAC cell or a causal origin of Cluster 6.
  • Independent prospective diagnostic, treatment selection, or outcome improvement from LPAS.
  • Clinical benefit from PSMB1 inhibition, LPAS-guided immunotherapy or computationally prioritized compounds.
  • Unique participant counts aggregated across partially overlapping or heterogeneous public datasets.

Complete reviewed limitations