403f9de122
The Problem ---------- Some read x haplotype pairs were getting very low likelihood when caching is on. Turning it off seemed to give the right result. Solution -------- The HaplotypeCaller only initializes the PairHMM once and then feed it with a set of reads and haplotypes. The PairHMM always caches the matrix when the previous haplotype length is the same as the current one. This is not true when the read has changed. This commit adds another condition to zero the haplotype start index when the read changes. Summarized Changes ------------------ * Added the recacheReadValue check to flush the matrix (hapStartIndex = 0) * Updated related MD5's Bamboo link: http://gsabamboo.broadinstitute.org/browse/GSAUNSTABLE-PARALLEL9 |
||
|---|---|---|
| .. | ||
| annotator | ||
| beagle | ||
| bqsr | ||
| compression/reducereads | ||
| diagnostics | ||
| diffengine | ||
| fasta | ||
| filters | ||
| genotyper | ||
| haplotypecaller | ||
| indels | ||
| phasing | ||
| validation | ||
| varianteval | ||
| variantrecalibration | ||
| variantutils | ||