I've been experimenting with using BWA to assemble 454 (1kb reads), comparing previous assemblies performed with the proprietary Roche assembler and I've noticed that BWA seems to be able to handle SNPs fairly well (both the BWASW and MEM algorithms), however some shorter indels that occur in tandem repeat sections only show up with the Roche software.
The section of the indel is: GGGGGC, and the roche software will find a deletion GGGG-C. However BWA won't find this.
I'm not sure what to make of this, is there a way to relax the BWA settings to capture this? Or on the other hand, should indels in tandem repeats not be trusted. The coverage on these indels is >300, while the frequency is <20%.