Design a study

Design a study answers, before a sample is sequenced, which mutational signatures a planned study could see. Choose a tumor type, an assay (whole genome, exome, panel) and the mutation count a sample is expected to have in that territory; the table lists every signature PCAWG tumors of the type carry, how many carry it and at what median share, how often the refit rule finds it when present at that count, how often it calls it when absent, how often such a call is right, and the count at which the rule finds it 80% of the time.

A signature is detectable when the rule finds it at least 80% of the time, calls it in at most 5% of samples without it, and a call is right at least 80% of the time; missed more often than not when the rule finds it in fewer than half the samples that carry it; unreliable in between; not seen when no PCAWG tumor carries it at that assay and count, so the grid has no case. The summary line gives the count to plan for so that every signature the grid can find reaches 80%.

The rates are the calibration grid behind Refit confidence: synthetic samples from half of PCAWG's tumors, projected onto each assay's territory and drawn at 5 to 5,000 mutations. Nothing is fitted here; the function is a lookup, needs no data, and opens from the Start board. A share link carries the three choices.

Absence is not evidence

The same lookup sits under every refit confidence report as Absence is not evidence for: the signatures the type (or PCAWG at that assay and count) carries that the rule would miss more often than not, each with the count at which it becomes findable and, for an exome or panel, the rate a genome at the same count would reach. A verdict with a caveat (a trade-off, a possible false positive, an uncertain call) now says what would settle it: the count on the same assay at which the grid says the call holds, or that no count in the grid does and the sample explorer, MuTopia or a second method would.