Markus Riester
Markus Riester
- [ ] gene-level summary not necessarily best way of doing it. check if we can safely extend to neighboring baits. maybe start 500kb left and right, iteratively shorten to...
Latest versions seem to have reduced artifacts quite a bit, but some are very difficult to get rid off. Flag genes with GC-content outliers, low mappability, small number of targets...
Probably wrapper around https://github.com/sztup/scarHRD
There should be an easy way to ignore recurrent artifactual segments directly in the segmentation instead of doing those things downstream.
Tri-allelic sites may be ignored in calculateMappingBiasVcf. This results in a pon.count of 0 and the default mapping bias prior.
Current heuristic simply uses neighboring SNPs. Would be better to train a regression model using database generating that takes distance, number of samples and coverage into account. Simple to benchmark...
This might be an improvement for difficult FFPE samples, besides the obvious advantage for small PoNs: https://www.biorxiv.org/content/10.1101/566505v1
When a segment contains mostly intervals with high variance in the PoN, it should be flagged. - [x] Flag in results$seg - [x] Flag in callAlterations - [ ] Flag...
It's becoming more common to have multiple biopsies from the same patient available. Together with join segmentation in processMultipleSamples, it would be nice to also support multiple biopsies in predictSomatic.