Analysis Methods
Integrated Trait–Gene Analysis Methods
Understand what TWAS, SMR, colocalization, and EWAS test and how PlantQTLdb filters their results.
Methods and reported support
- TWAS · FUSION
- Tests genetically predicted expression, splicing or APA usage against a trait. The workflow evaluates top1, LASSO and elastic-net models with five-fold cross-validation, using cis variants within 200 kb of the feature interval and at least five variants for training. A molecular feature is not necessarily a whole gene.
- SMR and HEIDI
- SMR combines molecular QTL and GWAS summary statistics. HEIDI separately tests heterogeneity: P_HEIDI < 0.05 is flagged Fail, P_HEIDI ≥ 0.05 is Pass, and a missing value remains NA. Passing HEIDI is not proof of mediation or causality.
- Colocalization · coloc
- Estimates posterior support for a shared causal variant within a locus. H0: neither trait associated; H1/H2: only one trait associated; H3: distinct causal variants; H4: a shared causal variant. This workflow uses coloc.abf, which assumes at most one causal variant per trait in the tested region.
- EWAS · OSCA
- Tests measured expression, splicing or APA features against a trait. The actual model is MOMENT or MOA. OSCA can fall back from MOMENT to MOA; the website labels the final model from the run log rather than inferring it from the output filename.
Thresholds and result availability
Window, model-training and prior settings above describe the accompanying scripts, not a guarantee that every historical dataset used identical parameters. Read the selected result's source metadata and stored thresholds. Under the single-causal-variant-per-trait assumption of coloc.abf, locus definition, priors, allele matching and LD context affect interpretation.
- TWAS and SMR
- The supplied summary scripts use P ≤ 0.05 / N tested molecular features within the selected dataset, trait and molecular layer. N counts features with observed test P values, not plotted points or unique genes. The website retains the threshold stored with the imported result.
- SMR versus HEIDI
- SMR significance and HEIDI status are reported separately. A significant SMR association can have Fail or NA for HEIDI; do not count every SMR hit as a result that passes both tests.
- EWAS
- P ≤ 0.001 is an unadjusted exploratory cutoff, not a Bonferroni or FDR threshold. A completed MOMENT or MOA result must also pass result-level QC. QC excluded means that output exists but fails validation, for example because of invalid estimates or degenerate P values. A failed or nonconverged run is not evidence of no association.
- Colocalization support
- PP.H4 < 0.5 is Weak; 0.5 ≤ PP.H4 < 0.8 is Suggestive; PP.H4 ≥ 0.8 is Strong. The website retains valid imported results in all three groups. The pipeline's supported (≥ 0.5) and strong (≥ 0.8) output subsets are distinct from this full table.
- Colocalization eligibility
- Features are selected through GWAS lead-variant windows of ±200 kb, then tested using their shared cis variants. Defaults require at least 50 shared variants and use priors p1 = 10⁻⁴, p2 = 10⁻⁴ and p12 = 10⁻⁵. No lead variants or insufficient shared variants means the feature was not eligible for this test.
Association plot display
In Trait–Gene Evidence, TWAS, SMR, and EWAS overview plots retain all significant results and, when needed, subsample non-significant results to at most 2,000 background points. This display step does not change the original P values or significance thresholds, which remain those of the complete analysis. Background point density therefore does not represent the full result distribution. The accompanying significant-result tables retain all significant results, rather than all tested features.
TWAS panels show molecular feature–trait associations, not a molecular QTL scan. EWAS uses molecular-feature rank on the horizontal axis. Colocalization reports posterior probabilities rather than P values; its Strong label uses PP.H4 ≥ 0.8.
Interpretation limits
These methods answer related but non-identical questions. Their results should be compared within the same dataset, trait, and molecular layer and interpreted together with locus structure and biological knowledge.
Workflow source code
Reviewed code version: f2bde42 (8 November 2026). Use the workflow and its linked preparation and summary scripts together when reproducing an analysis.