Fix indel position offset in paired-end reverse reads

[colinhercus]

For read_2 (reverse), the reference_segment is constructed starting padding bases before self.position to allow room for deletions after reverse_complement. apply_mutations was computing the intra-segment index using self.position, which caused mutations to be placed padding (= read_len // 5) bases too early in the segment. After reverse_complement this manifested as indels appearing at a different reference coordinate in read_2 than in read_1.

This results in one false positive indel call for each true positive.

Fix: add segment_start attribute to Read, set to the actual reference coordinate of reference_segment[0]. apply_mutations now subtracts segment_start instead of position when computing the index.

Image shows indel in two positions and at one loci all in the +strand reads and in the other all in the - strand reads.

Correction done with assistance from Claude

[colinhercus]

While this pull request fixes the issue a more robust fix would do:

1. Generate fragment length and extract
2. Apply mutations
3. Generate read1 & 2 from either end of the fragment.

This would require no padding and be less prone to bugs

[joshfactorial]

The code looks good. We need some regression tests. I added some on this branch: https://github.com/ncsa/NEAT/tree/colinhercus-fix-paired-end-indel-position - you can check and add those or others to this for the PR.

[colinhercus]

Thanks for developing NEAT, it's is an excellent read simulator with it's working for us now. We get excellent precision and recall of indels and other variants on all our tests.

Sorry but I don't really have time to do anymore at the moment.

[colinhercus]

Thanks!