Hello all,
We've just started using GATK in order to perform variant calling in a non-model teleost fish. The fish genome is highly repetitive (>65%), and also suffers from the whole genome duplication event common in teleosts (e.g. zebrafish). Additionally, the fish strain we use is highly inbred, which should result in a highly homogenous genome. We have generated a genome assembly and a de novo repeat library based on NGS data (manuscript submitted) before mapping the reads from four individuals (male and female) to the genome via bowtie2. Variants were called using UnifiedGenotyper.
We generally get a very good list of variants, but it seems that we're getting a number of false positives and negatives when calling variants. Some of these appear to be due to paralogues, but some seem to be errors in the actual genotype call. For example:
scaffold00001 1199020 . T G 44.35 . AC=1;AF=0.167;AN=6;BaseQRankSum=-7.420;DP=110;Dels=0.00;FS=152.859;HaplotypeScore=3.6965;MLEAC=1;MLEAF=0.167;MQ=42.00;MQ0=0;MQRankSum=-1.972;QD=1.53;ReadPosRankSum=-2.777;SB=-4.096e+00 GT:AD:DP:GQ:PL 0/1:20,9:29:79:79,0,588 0/0:16,7:23:12:0,12,447 0/0:39,18:57:65:0,65,1426 ./.
In this case, individual 3 has a homozygous reference genotype, despite having a 31% minor allele frequency. Individual 1 also has a 31% minor allele frequency, but is called heterozygous.Some of the bases used to call the G allele are of low quality (when looking more closely using IGV), but I would still expect the genotype to be heterozygous.
A reverse example:
scaffold00458 298207 . A G 64.81 . AC=2;AF=0.333;AN=6;BaseQRankSum=3.027;DP=64;Dels=0.00;FS=5.080;HaplotypeScore=0.0000;MLEAC=2;MLEAF=0.333;MQ=16.26;MQ0=0;MQRankSum=3.177;QD=1.16;ReadPosRankSum=-3.252;SB=0.439 GT:AD:DP:GQ:PL 0/0:8,0:8:21:0,21,207 0/1:20,1:21:13:13,0,152 0/1:31,4:35:90:90,0,102 ./.
Here, individual 2 is called heterozygous, but there is only a single read which supports the minor allele. Additionally, when looking at IGV, you can see that the read in question has a number of mismatches, suggesting it originates from another area of the genome.
I've also uploaded screenshots of IGV if that I hope will help clarify the problems we're having. We have used default parameters of GATK in almost all cases, and we did not used VQSR, as we did not have a list of high confidence SNPs at the time.
