Tophat2 chose a match with a deletion over one without

[samhokin]

I'm a bit mystified by Tophat's (v2.1.0) choice of alignment in a case where I've got two similar genomes placed end to end (B73 and W22 maize) and I'm trying to map reads from a mixed (or non-mixed in this case) sample to their originating genome.

So this case came up, where Tophat chose to assign a read to the B73 chromosome, even though it has an deletion in that case, while the same read maps to the corresponding W22 chromosome perfectly, without the deletion. Here's the BLAST comparison: "Chr2" is W22 chromosome 2, "2" is B73 chromosome 2. W22/Chr2 matches 100%, without gaps (e=1e-45); B73/2 has one gap (e=5e-44).

BLASTN 2.2.31+

Reference: Zheng Zhang, Scott Schwartz, Lukas Wagner, and Webb
Miller (2000), "A greedy algorithm for aligning DNA sequences", J
Comput Biol 2000; 7(1-2):203-14.

Database: B73v3.W22v3.genome.fa
           24 sequences; 4,458,629,333 total letters

Query= DB775P1:321:C4J4FACXX:7:1304:6219:16563

Length=101
                                                                      Score     E
Sequences producing significant alignments:                          (Bits)  Value

  Chr2                                                                187     1e-45
  2 dna:chromosome chromosome:AGPv3:2:1:237917468:1                   182     5e-44

> Chr2
Length=251269698

 Score = 187 bits (101),  Expect = 1e-45
 Identities = 101/101 (100%), Gaps = 0/101 (0%)
 Strand=Plus/Plus

Query  1          ACATAGTGCACCGATCACTGTTCTACTAGTATGGGCAGTAGAAGCAAAATGGTGGCGCGG  60
                  ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
Sbjct  250221315  ACATAGTGCACCGATCACTGTTCTACTAGTATGGGCAGTAGAAGCAAAATGGTGGCGCGG  250221374

Query  61         TCGCTTCTTTTCATGTTGACATCACATCCTGGAGTAGCCCT  101
                  |||||||||||||||||||||||||||||||||||||||||
Sbjct  250221375  TCGCTTCTTTTCATGTTGACATCACATCCTGGAGTAGCCCT  250221415


> 2 dna:chromosome chromosome:AGPv3:2:1:237917468:1
Length=237917468

 Score = 182 bits (98),  Expect = 5e-44
 Identities = 101/102 (99%), Gaps = 1/102 (1%)
 Strand=Plus/Plus

Query  1          ACATAGTGCACCGATCACTG-TTCTACTAGTATGGGCAGTAGAAGCAAAATGGTGGCGCG  59
                  |||||||||||||||||||| |||||||||||||||||||||||||||||||||||||||
Sbjct  236204183  ACATAGTGCACCGATCACTGTTTCTACTAGTATGGGCAGTAGAAGCAAAATGGTGGCGCG  236204242

Query  60         GTCGCTTCTTTTCATGTTGACATCACATCCTGGAGTAGCCCT  101
                  ||||||||||||||||||||||||||||||||||||||||||
Sbjct  236204243  GTCGCTTCTTTTCATGTTGACATCACATCCTGGAGTAGCCCT  236204284

Lambda      K        H
    1.33    0.621     1.12 

Gapped
Lambda      K        H
    1.28    0.460    0.850 

Effective search space used: 325479892253

  Database: B73v3.W22v3.genome.fa
    Posted date:  Apr 30, 2016  9:09 AM
  Number of letters in database: 4,458,629,333
  Number of sequences in database:  24

Matrix: blastn matrix 1 -2
Gap Penalties: Existence: 0, Extension: 2.5

The Tophat parameters are -g 1 -M -N 0 to enforce the choice of a single, perfect match. But Tophat should have chosen the W22 locus, not the B73 locus in this case.

Any ideas why it chose the non-perfect match???