Improve criteria for when to stop remap iterations

[donkirkby]

As an example, sample 88067A_HCV from the 29-May-2017.M01841 run has a drop in coverage at the start of the E2 region in HCV-2. That drop wasn't there when it was analysed with version 7.6; it appeared with version 7.7. When I investigated, I found that version 7.6 ran two remap iterations, and 7.7 only ran one. After the first iteration, version 7.6 still had the gap, so the second iteration was very useful.

Version 7.6 ran the second iteration, because one of the HIV seeds got dropped. It just happened to also help the HCV coverage. In 7.7, the HIV had already been mapped by the G2P step, and we already had over 95% mapped reads, so it didn't do a second remap iteration.

Ideas for better criteria:

• Always remap at least twice. (Too wasteful?)
• Don't consider reads with a lot of soft clipping as mapped.

[donkirkby]

Another example: I was comparing the 7.7 results with a 7.8 release candidate (03410f7). Sample 73053A-HCV_S20 went from 4 remaps in 7.7 down to 3, and a big gap got worse. Perhaps, we should always do at least two remaps after the last reference has been dropped.

[donkirkby]

After a strategy discussion, we decided to look at de novo assembly before trying to improve the remapping strategy.