Difference: DanielsJournal (109 vs. 110)

09/01/10

Finished removing SAM generation from all mappers, and made SAM generation an optional post process:

• As mentioned yesterday, all mappers now return positions
• To generate SAM entries, the following is done:
  • Using the position, a subsequence from the refernece is extracted
    • The length of this subsequence is typically slightly longer than the read itself, to allow for indels, if present
  • An alignment is conducted, and the aligned read and refernece (with dashes) is returned
  • The dashed alignments are used to generate SAM and MD
• Some major problems:
  • LocalAligner has a tendency to lop off mismatches at the end of reads to generate a higher score
    • This is highly evident when aligning with mismatch_test.fq
  • Setting the mismatch penalty to 0 can prevent the removal of mismatches at one end, but not both...
    • When this is done, the CIGAR breaks... by inserting a lot of S (for softly clipped, which are "free bases" in the alignment)
  • The indel preference of the SW algorithm is more similar to BWA, as described on 08/06/10, and thus, most of the solutions provided in indel_test.fq are incorrect...
  • Fixing these requires some changes to all aligners, and also the CIGAR and MD generation functions inside SamEntry

Other changes to mapper:

• LFPair and LFAll are used in all possible places
• bowtie and quality mappers work completely in numberspace
  • bwa still works in letter space... and converts when required

And assuming that last infinite loop in getPosition isn't a mapper problem, I guess that's it