Open snurk opened 4 years ago
In particular, posing a stricter initial threshold can help filtering out potential placements in bubble contig analysis.
This task includes reconsideration (and likely removal) of all the 'stricter' threshold that we might have experimentally introduced in individual procedures.
Related is possible increase of minOvlLength to 1K
Some overlaps seem to indeed be low quality (due microsatellite repeats). Options seem to be:
Currently we are considering overlaps with error-rate up to 1%, which might be an overkill, considering identity distribution of the reads post-compression! Decreasing this threshold might be important for bubble/confusion detection, because we can not easily pose stricter thresholds within individual procedures (for the overlaps, which are expected not to originate from the very same location/haplotype). Indeed, if we ignore a 0.7% overlap of size 7k, while trying to use a 0.5% threshold, we might miss the fact that it actually had a 'suboverlap' of 0.3% of size 5k.