Interactive research supplement

FusionSieve candidate evidence explorer

Load the private analysis file locally to examine every sample-level hypothesis marked REVIEW. These are computational candidates for human review, not confirmed fusions.

← Return to the scientific report

What was detected?

The corrected analysis generated 5,994 hypotheses. It rejected 5,952 and retained 42 sample-level hypotheses in 15 breakpoint clusters for human review. None reached PASS or HIGH, and none has an orthogonal truth label.

42REVIEW hypotheses
15breakpoint clusters
21sample aliases represented
0confirmed positive calls

Load the detailed results without publishing them

The public page contains aggregate counts only. On the analysis computer, choose the private candidate file generated with this report. The browser reads it into this tab and reveals original sample names, exact breakpoints, and evidence counts locally. The file is not uploaded, cached, or stored by the website.

No sample-level data are loaded. Aggregate public results are shown above.

Filter, hover, and click a hypothesis

Hover or focus a row for a rapid evidence preview. Click the fusion name or row for full read-support details, score components, filters, recurrence, and exact GRCh37/hg19 loci for IGV.

Choose the private candidate file above to inspect sample-level results.

Fusion Sample Evidence score Conservative support Unique split Unique spanning REVIEW samples in cluster

How the evidence score is calculated

Points added

  • Unique split-read evidence: up to 35 points.
  • Unique spanning-pair evidence: up to 25 points.
  • Fusion support fraction: up to 10 points.
  • Weaker-arm wild-type depth: up to 10 points.
  • Split-read uniqueness and arm-depth balance: up to 10 points each.

Warnings subtracted

  • Placement-ambiguity hard filters: 20 points; sequence-artifact hard filters: 10 points.
  • No mate support: 8 points; anchor imbalance: 6 points.
  • Low weaker-arm depth, mismatch-rich evidence, and uniform offsets: up to 6 points each.
  • Exact recurrence across all detected samples: up to 15 points.
  • Coordinate-family disagreement and pre/post duplicate sensitivity: up to 10 points each.

The known-fusion catalogue flag, the caller’s unvalidated model score, and prior-only p/q values add no points. All scores are capped at 69; stricter caps apply to split-only, no-mate, recurrent, hard-filtered, or duplicate-sensitive evidence.

Open the exact breakpoints in IGV

  1. Select a candidate and choose Open both loci in IGV-Web.
  2. In IGV-Web, use Tracks → Local File to choose that sample’s original BAM and matching BAI.
  3. Keep the BAM-loaded IGV session open. For another candidate, use Copy IGV loci here and paste both loci into IGV’s search field.
  4. Inspect soft clipping, split alignment, mate orientation, mapping quality, duplicate flags, start-position diversity, nearby mismapping, and coverage on both arms.

IGV-Web user guide

Candidate