# Cell type-specific contextualisation of the human phenome: towards the systematic treatment of all rare diseases.

- **DOI:** [10.1186/s13073-026-01692-0](https://doi.org/10.1186/s13073-026-01692-0)
- **Rating:** Limited (2.8 out of 5)
- **Verdict:** A valuable phenome-scale rare-disease resource whose therapeutic promise is constrained by unvalidated causal inference and internally inconsistent validation statistics.
- **Framework version:** v5.9
- **Pipeline version:** v0.4
- **Evaluation date:** 2026-10-08
- **Evaluation page:** https://www.nabu.science/paper/10.1186%2Fs13073-026-01692-0

## Key claims

1. We identified significant associations between 201 cell types and 9,575/11,028 (86.7%) phenotypes across 8,628 RDs, substantially expanding knowledge of phenotype-cell type links.
   - Support: Tentative
   - Type: Primary finding
   - Basis: broad association coverage with unreconciled denominators, moderate replication and no precision estimate for novel links
   - Location: Abstract, Results
2. All 7 HPO branches showed disproportionate associations with on-target cell types from their respective organ systems.
   - Support: Supported
   - Type: Secondary
   - Basis: branch-level ontology matching supports the reported enrichment pattern, though it is an internal positive-control analysis
   - Location: Results, Validation of expected phenotype-cell type relationships
3. This yielded putative therapeutic targets for 5,252 phenotypes across 4,819 diseases in 201 cell types and 3,148 genes.
   - Support: Supported
   - Type: Secondary
   - Basis: the reported pipeline directly produces these putative outputs, but sequential heuristic filters limit validation of their prioritisation
   - Location: Results, Therapeutic target identification
4. We recapitulated 87% (120 total) of all non-failed existing gene therapy targets, with significant enrichment (p=5.6×10⁻⁵, odds ratio=3.0).
   - Support: Tentative
   - Type: Secondary
   - Basis: retrospective gene-level overlap is reported, but recovery conflicts with sensitivity and uses a future-dated TTD release
   - Location: Results, Therapeutic target validation
5. Our pipeline implicates M2 macrophages (‘Alternatively activated macrophages’) as the only causal cell type
   - Support: Tentative
   - Type: Secondary
   - Basis: associative analysis of an incomplete reference atlas cannot establish an exclusive causal cell type
   - Location: Results, Selected example targets

## Strengths

- **Phenome-scale integrated rare-disease resource** The framework connects HPO and GenCC evidence to 201 cell types and reports associations across thousands of phenotypes and diseases, then exposes the results through software and a portal.
  - Dimension: contribution
  - Location: Abstract; Results, Phenotype-cell type associations; Methods, Rare Disease Celltyping Portal
- **Multiple complementary validation layers** The study uses organ-system matching, Monarch Knowledge Graph recall and comparisons across two atlases and developmental stages rather than relying on a single internal check.
  - Dimension: methodological rigour
  - Location: Results, Validation of expected phenotype-cell type relationships; Validation of inter- and intra-dataset consistency; Validation using biomedical knowledge graphs
- **Reusable software and access pathway** Three R packages, analysis scripts and an interactive portal support programmatic and non-programmatic reuse by rare-disease researchers and translational teams.
  - Dimension: application pathway
  - Location: Software and Data Availability; Methods, Novel R packages and Rare Disease Celltyping Portal

## Limitations

- **Validation statistics do not reconcile** The consistency analyses report permutation p-values unattainable with 1,000 permutations, and therapeutic-target recovery, sensitivity and failed-target interpretations conflict between the Results and Fig. 8. These discrepancies make exact validation strength unreliable.
  - Dimension: methodological rigour
  - Location: Methods and Results, Validation of inter- and intra-dataset consistency; Results, Therapeutic target validation; Fig. 8
- **Causal language exceeds associative design** Regression of curated gene evidence against reference-atlas expression specificity can nominate associations but cannot establish causal or exclusive cell types. Claims such as “only causal cell type” are especially difficult to reconcile with acknowledged atlas omissions.
  - Dimension: reporting
  - Location: Methods, Phenotype-cell type associations; Results, Selected example targets; Discussion, Current limitations
- **Benchmarks miss decisive performance questions** The MKG analysis reports recall without precision, the therapeutic benchmark is gene-level rather than cell-type-specific, and sequential target filters are not ablated. The claimed EWCE advantage is also deferred to an external issue rather than demonstrated in the article.
  - Dimension: methodological rigour
  - Location: Methods, Phenotype-cell type associations and Therapeutic target identification; Results, Validation using biomedical knowledge graphs and Therapeutic target validation

## Reliability checks

Caveats (4 of 4 checks).

- **Results check:** Noted
  - We identified significant associations between 201 cell types and 9,575/11,028 (86.7%) phenotypes across 8,628 RDs, substantially expanding knowledge of phenotype-cell type links: "We identified significant associations between 201 cell types and 9,575/11,028 (86.7%) phenotypes across 8,628 RDs, substantially expanding knowledge of phenotype-cell type links." (Abstract, Results) vs "Across 201 cell types and 11,047 phenotypes, we found >46,514 significant phenotype–cell type relationships" (Discussion). The Abstract fraction gives 9,575/11,028 × 100 = 86.824%, whereas using the Discussion denominator gives 9,575/11,047 × 100 = 86.675%, so the reported 86.7% matches the latter denominator after rounding. → Conclusion holds
  - We recapitulated 87% (120 total) of all non-failed existing gene therapy targets, with significant enrichment (p=5.6×10⁻⁵, odds ratio=3.0): "\begin{document}$$87$$\end{document}% (120 total) of all non-failed existing gene therapy targets" (Results, Therapeutic target validation) vs "\text {sensitivity}$$\end{document}=\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$0.83$$\end{document}" (Results, Therapeutic target validation, non-failed gene-therapy enrichment metrics). Recovery of 87% corresponds to sensitivity 0.87 rather than 0.83 if both quantities use the stated non-failed gene-therapy target set, but the underlying contingency counts needed to determine which figure is correct are not supplied. → Effect on result unclear
  - Minor reporting notes (2):
    - Descriptive: "of the non-failed therapeutic targets and 0% of the terminated/withdrawn therapeutic targets" (Results, Therapeutic target validation, all therapeutic modalities) vs "\text {sensitivity}$$\end{document}=\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$0.27$$\end{document}" (Results, Therapeutic target validation, failed-therapeutic depletion metrics). Zero recovery of terminated/withdrawn targets would imply sensitivity zero if failed targets define the positive class, whereas the accompanying sensitivity is 0.27, but the text supplies neither the contingency table nor a different definition reconciling the metrics. → Effect on result unclear
    - Descriptive: "In all cases, we confirmed the robustness of these results by iteratively comparing the observed correlations against permuted baseline correlations (1,000 permutations, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$p<$$\end{document} \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$2.2 \times 10^{-308}$$\end{document}" (Results, More specific phenotypes are associated with fewer genes and cell types). An empirical proportion from 1,000 permutations has increments of 1/1,000 = 0.001, so an estimated zero exceedance proportion cannot support the reported extreme precision and would conventionally be reported at approximately the 0.001 resolution. → Conclusion holds
- **Reporting and transparency:** Noted. The selected-example introduction announces four examples but lists three.
  - Data availability: declared · Code availability: declared
- **Reference integrity:** Clean
  - 79 references in manuscript
  - 76 of 76 checkable references found in an index
  - 3 have no canonical index record — counted, but not index-checkable
  - 1 reference confirmed by manual review
  - References not verified automatically are resolved manually within 24 hours.

## Post-publication record

- **Status:** Clean
- **Summary:** No retraction or expression of concern found in Crossref (includes Retraction Watch).
- **Last checked:** 2026-10-17
