# Co-expression-based models improve eQTL predictions for transcriptome-wide association studies and highlight new schizophrenia-associated genes.

- **DOI:** [10.1038/s41588-026-02646-3](https://doi.org/10.1038/s41588-026-02646-3)
- **Rating:** Sound (3.0 out of 5)
- **Verdict:** Coexpression-informed trans models broaden schizophrenia TWAS coverage, but evaluation-cohort-dependent filtering makes the reported predictive gains materially uncertain.
- **Framework version:** v5.9
- **Pipeline version:** v0.4
- **Evaluation date:** 2026-10-08
- **Evaluation page:** https://www.nabu.science/paper/10.1038%2Fs41588-026-02646-3

## Key claims

1. Applying this framework to Psychiatric Genomics Consortium wave 3 genotypes identified 766 genes associated with schizophrenia
   - Support: Tentative
   - Type: Primary finding
   - Basis: association design lacks independent schizophrenia replication and inherits evaluation-dependent gene selection
   - Location: Abstract; Results, “coTWAS identifies trait associations in PGC3 cohorts”
2. Integrating these models with conventional cis-based predictors improved gene expression imputation for 18,744 genes across regions
   - Support: Tentative
   - Type: Secondary
   - Basis: the pooled count uses evaluation-dependent filtering and exceeds the per-region cis–trans enhancement counts
   - Location: Abstract; Results, Figure 4a
3. incorporating distal (trans) regulatory effects improves the prediction of gene expression and the identification of disease-associated genes
   - Support: Tentative
   - Type: Secondary
   - Basis: coverage increases, but trans models have lower external per-gene accuracy and disease associations remain unvalidated
   - Location: Abstract

## Strengths

- **Broad trans-aware TWAS methodological extension** INGENE and MODULE provide distinct coexpression-informed routes to trans prediction and are integrated with CIS and EpiXcan. The study also compares its approach with BGW-TWAS and MOSTWAS in DLPFC.
  - Dimension: contribution
  - Location: Introduction; Results, “Evaluation of cis and trans model training performance”
- **Multi-cohort predictive evaluation design** Models are examined across LIBD, GTEx, and CMC brain datasets, and the combined predictor is evaluated in CMC after fitting in GTEx. The results transparently show that cis models retain higher per-gene external adjusted R².
  - Dimension: methodological rigour
  - Location: Results, Figures 3 and 4c
- **Large schizophrenia association application** The integrated framework is applied across 62 PGC3 cohorts comprising 102,613 individuals, with conditional analysis and leave-site-out consistency assessment. This creates a substantial hypothesis-generation resource for psychiatric genetics.
  - Dimension: application pathway
  - Location: Results, “coTWAS identifies trait associations in PGC3 cohorts”; Table 2; Supplementary Figure 18

## Limitations

- **Independent testing is materially compromised** Trans genes are retained using agreement with models trained in GTEx or CMC before those cohorts are described as independent evaluation datasets. This asymmetric, evaluation-cohort-dependent selection can inflate the reported coverage of trans models.
  - Dimension: methodological rigour
  - Location: Methods, “Cross-dataset training of INGENE and MODULE”; Results, “Evaluation of cis and trans models on a testing dataset”
- **Headline improvement count lacks precision** The abstract associates improved imputation with 18,744 genes, whereas Figure 4a reports only 847–1,362 genes per region for which the cis–trans model was best. Uncorrected per-gene selection at α=0.05 further complicates interpretation of the pooled count.
  - Dimension: reporting
  - Location: Abstract; Methods, “ANOVA analysis in GTEx”; Results, Figure 4a
- **Mechanistic claims exceed association design** The paper acknowledges that imputed-expression associations do not establish causality and that bulk tissue obscures cell types, yet concludes that network interactions contribute substantially to schizophrenia risk. The 641 newly reported associations also lack PGC-independent or functional validation.
  - Dimension: positioning
  - Location: Discussion, limitations paragraph and final paragraph; Results, Figure 6

## Reliability checks

Caveats (4 of 4 checks).

- **Results check:** Noted
  - Integrating these models with conventional cis-based predictors improved gene expression imputation for 18,744 genes across regions: "Integrating these models with conventional cis-based predictors improved gene expression imputation (maximum likelihood estimation, α = 0.05) for 18,744 genes across regions." (Abstract) vs "Best-fitting model counts per region: amygdala (trans, 7,169; cis, 2,717; cis–trans, 847), CN (trans, 5,763; cis, 3,674; cis–trans, 889), dACC (trans, 8,393; cis, 2,822; cis–trans, 949), DLPFC (trans, 8,148; cis, 4,475; cis–trans, 1,362), HP (trans, 5,790; cis, 2,654; cis–trans, 881), sACC (trans, 8,521; cis, 3,443; cis–trans, 1,232)." (Figure 4a caption). The cis–trans best-model counts sum to 847 + 889 + 949 + 1,362 + 881 + 1,232 = 6,160 gene–region pairs, so they cannot establish improvement from combining cis and trans predictors for 18,744 distinct genes. → Not supported by the paper's own data
  - Minor reporting notes (1):
    - Descriptive: "Mapping coTWAS-significant genes to PGC3 loci5 showed that 158 genes (27.4% of all tissue-specific associations) fell within significant genome-wide regions, while 608 (72.6%) lay outside" (Results, coTWAS identifies trait associations in PGC3 cohorts) vs "Across 96,535 tests, we identified 1,162 significant associations (975 non-MHC), corresponding to 766 independent genes" (Results, coTWAS identifies trait associations in PGC3 cohorts). The mapping counts sum to 766 genes, for which 158/766 × 100 = 20.6% and 608/766 × 100 = 79.4%, whereas using the stated 1,162 tissue-specific associations would yield 13.6% and 52.3%, respectively. → Changes the size materially
- **Reporting and transparency:** Noted. The ancestry-exclusion sentence conflicts with the stated European-ancestry inclusion criterion, but the supplied text cannot establish which threshold was implemented or recalculate any affected result.
  - Ethics approval: declared
- **Reference integrity:** Clean
  - 122 references in manuscript
  - 118 of 118 checkable references found in an index
  - 4 have no canonical index record — counted, but not index-checkable
  - 4 references 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
