Identifying mutations from Paired-End Sequencing data

Hello! I'm trying to get mutations from paired-end sequenced reads aligned with BWA  using SamTools. Coverage is about 16,000. Generally it works fine, but in one fragment (TGGGC) i see that in reads sequenced from left to right there is deletion of G (TGGC) in 12,000 out of 14000 cases, but in reads sequenced from right to left there is no this mutation at all. So is there deletion in heterozygote or there is just problem with sequencing (was carried out with Ion Torrent PGM), or there is problem with alignment?