When I compared the result of HaplotypeCaller and Unifiedgenotyper, I found some locus that only be calling variation by UG. However, the origin bam file and bamout file produced by reassembly both contain this variation(see the figure 1). The number of reads supporting reference and alter allele are 456(87%) and 69(13%) respectively in raw bam. Meanwhile they are 439(86%) and 69(14%) in bamout file. I have got confused with this problem.