DRAGEN variant detection benchmarking summary
DRAGEN v4.0.5b was compared to DRAGEN v4.0.5 and minimal changes were identified. As a result, DRAGEN v4.0.5b is now used in the pipeline: report.
Previously, validation of DRAGEN v4 software was undertaken to assess performance of variant detection for use in the rare disease bioinformatics pipeline for the NHS GMS. Validation assessed variant types that are currently included in the NHS GMS Rare Disease pipeline small variants, copy number variants and short tandem repeats (NB. short tandem repeats, mitochondrial variants and de novo variants are not currently assessed as part of the Newborn Screening Pipeline). The validation showed that DRAGEN v4 software achieves similar or improved performance for variant detection in comparison with the previously used DRAGEN v3.2.22 (NHS GMS Rare Disease pipeline). An overall summary of variant detection assessments is shown in the table below, with changes in comparison to the current ISO accredited pipeline (NHS GMS Rare Disease pipeline) indicated in parentheses. The performance of DRAGEN v4 is equivalent or improved in all areas included in the variant detection assessment.
| Variant Type | Validation Category | Specific DRAGEN version(s) assessed | Sensitivity (%) | Precision (%) | Specificity (%) |
|---|---|---|---|---|---|
| Small variants | Benchmarking - SNVs* | v4.0.5 | 99.78 (+0.39) | 99.95 (+0.21) | - |
| Benchmarking - Indels* | v4.0.5 | 99.79 (+0.26) | 99.91 (+0.21) | - | |
| GenQA sample - SNVs | v4.0.3 | 99.85 (+0.05) | 99.93 (+0.06) | - | |
| GenQA sample - Indels | v4.0.3 | 99.64 (+0.05) | 99.57 (+0.27) | - | |
| de novo detection | v4.0.4 , v4.0.5 | 99.4 (0.00) | - | - | |
| Mitochondrial variants | v4.0.5 | 100 (0.00) | - | - | |
| Copy Number Variants | Detection** | v4.0.3 | 97.95 (0.00) | - | - |
| Short Tandem Repeats | Detection*** | v4.0.3 | 100 (+4.2) | - | 99.7 (+0.10) |
* GIAB truthset metrics for high confidence regions defined in NISTv4.2.1 identified through singleton analysis are presented in the summary. For more details and results from family-detection see section 5.1. ** Exact sensitivity for events that are partial or completely overlapping events from orthogonal testing *** Exact metrics for STRs that are green in PanelApp, excluding expansions at the FMR1 locus. More details can be found in the validation reports (Genomics England employees only).