Hacker News new | ask | show | jobs
by AlexCoventry 2772 days ago
"99,7% SNP precision and sensitivity." [0]

Sensitivity = true-positive-rate = 0.997.

Precision = 0.997 = #true-positives / (#true-positives + #false-positives) = true-positive-rate / (true-positive-rate + false-positive-rate) = 0.997 => true-positive-rate + false-positive-rate = 1 => false-positive-rate = 0.003. [1]

That seems like a very high error rate, about 10 million errors in the three-gigabase genome, and 100 thousand errors in the 30-megabase exome (protein-coding regions.) That might be an acceptable rate for population-level analysis if the errors are sufficiently uncorrelated, but I wouldn't want to be making decisions on the basis of it for personalized medicine. For comparison, here's a rough estimate that an individual human genome has 2-3 million SNPs [2].

I thought you could do better than that with 30x coverage, so I might be misinterpreting them, somehow. Or maybe they're using an unconventional sequencing technology which is cheaper but less accurate.

[0] https://us.dantelabs.com/products/whole-genome-sequencing-wg...

[1] Equations given here: https://en.wikipedia.org/wiki/Sensitivity_and_specificity

[2] https://biology.stackexchange.com/a/51315/37343