A clinical NGS pipeline moves a sample through extraction, library prep, sequencing, alignment, QC filtering, variant calling, interpretation and reporting. Each read covering the variant locus is either "reference" or "alternate", and each carries a per-base Phred quality score. Reads below the QC threshold are discarded before calling — this is exactly what the flowing particles show: grey unsequenced material becomes ref/alt-coloured reads at the sequencer, and low-quality reads are dropped (and visually fall away) at the QC station.
Variant confidence is computed the way real callers estimate statistical significance: assuming a background sequencing error rate ε, the probability that the observed alt-read count k arose from noise alone follows a Poisson tail, converted to a Phred-scaled quality score.
λ = n · ε (n = depth, ε = 0.005 error rate)
P(error) = P(X ≥ k | X~Poisson(λ))
QUAL = −10·log10(P(error)) [Phred scale]
call = PASS if QUAL ≥ threshold, else FAIL
- Sequencing depth — reads generated at the sequencer station; higher depth = more statistical power to detect a variant.
- Variant allele fraction — true fraction of alt reads sampled via a binomial draw at each batch.
- QC / call threshold — minimum Phred quality a read needs to survive QC, and the QUAL score a variant needs to be called PASS.
- Inject low-VAF variant — sets VAF to 3%, mimicking a low-frequency somatic/mosaic variant near the detection limit.
This is why clinical labs specify minimum depth (e.g. ≥30–50× for germline, ≥500× for cancer hotspot panels) and accredited QC thresholds before a variant can be reported: low depth or low VAF pushes QUAL below the confidence needed for a validated clinical call.