Read more about it here!
HaplotypeCaller in GATK4 vs GATK3
I know that HaplotypeCaller in GATK4 is still in development, but I want to make sure this issue is known We're seeing very different behaviour between the HC in GATK3 and GATK4 when using targeted resequencing data. By design, and in contrast to exome seq., we have many reads that start in the same place (and so set --maxReadsPerAlignmentStart 0), but also many positions that are covered only by reads from one strand and not the other. HC in GATK3 was assembling these and handling them fine in most cases (except when the SNP was very close to read end), but GATK4 seems to not assemble those portions in many cases, even when the SNP is ~20 nt away from the end of the read. Is this known? We can provide sample data for reproducing this problem if needed.