Summary-statistics analysis guide
What users upload, what OmniGenet can run directly, and what still requires external reference data or software.
Shared GWAS input
Recommended columns are SNP, CHR, BP, A1, A2, BETA, SE, P, N, and optionally EAF. OmniGenet detects common aliases, inspects missingness and fields, and records a harmonized copy plus an input-QC manifest.
Analysis capabilities
| Analysis | Minimum user upload | Built-in output | Production additions |
|---|---|---|---|
| GWAS plots | P for Q-Q; P + chromosome + position for Manhattan | Manhattan, Q-Q with λGC, effect distribution, and a study-level forest plot when study/beta/SE rows are available | Review population structure, imputation/QC, genomic control, and meta-analysis heterogeneity before interpreting peaks |
| PRS | One discovery GWAS | Thresholded PLINK-compatible SNP-weight files, a weight manifest, and tuning/target readiness records | A tuning genotype/phenotype cohort for parameter selection and a separate target genotype/phenotype cohort for frozen-weight scoring and R²/AUC |
| TWAS | GWAS + expression-weight TSV | Weighted-Z gene associations, FDR, model coverage | LD/covariance and trained models for FUSION, PrediXcan, S-MultiXcan, UTMOST, or METRO |
| Fine-mapping | GWAS with chromosome, position, beta, and SE | Automatic lead locus, Wakefield ABF PIP, credible set | Locus LD matrix for SuSiE/FINEMAP and ancestry-aware engines |
| Colocalization | Two harmonizable GWAS files | PP0–PP4 and shared-variant posterior ranking | LD-aware multi-signal analysis such as coloc.susie when appropriate |
| MR | Exposure + outcome GWAS | IVW, weighted median, MR-Egger, confidence intervals, F statistics, heterogeneity and leave-one-out | Ancestry-matched LD clumping and specialist sensitivity packages for production inference |
PRSice-2 target and validation interface
The standard PRSice-2 workflow uses target_genotypes and target_phenotype with the discovery GWAS. A separate tuning_genotypes and tuning_phenotype pair is optional and is used only by the advanced independent-validation mode. PLINK BED/BIM/FAM companion files can be uploaded one at a time under the corresponding cohort role. Use POST /api/prs/cohorts/validate to check cohort readiness. Core execution calls the compiled prsice_bin directly; PRSice.R is optional. The result package includes prs_accuracy.tsv with the best threshold and target PRS R².
Reports and reproducibility
Each completed job preserves the submitted TSV request, YAML manifest, run log, QC metadata, method-eligibility table, analysis summary, checksums, result index, result tables, and an HTML report. The report is descriptive research output, not a clinical interpretation.
Accuracy boundary
Summary statistics alone can create SNP weights, gene associations, credible sets, colocalization probabilities, and MR estimates. They cannot measure target-cohort PRS accuracy. Predictive R², AUC, calibration, and clinical utility require an independent target genotype dataset, observed phenotype, covariates, and a prespecified validation design.
Data handling before public deployment
Use per-user storage, authentication, quotas, encrypted transport, automatic retention/deletion policies, malware and archive checks, and isolated worker processes. Do not expose uploaded datasets through the website static directory.
Built-in methods are transparent baselines. The method-readiness panel distinguishes ready analyses from missing inputs, resources, or external software.