I'm always unsure whether to use VQSR or hard filtering when I do site filtering after joint-calling using GATK. One reasonable criterion I can think of is to inspect the structure of the VQSR model (that 2D heatmap produced when running VQSR). I'm not sure if I'm on the right track. Any suggestions are welcome! Thanks!
For an example, here I attached the report of the model fitted for INDELS from a ~150X WES of ~200 samples. Based on the result, will you recommend VQSR or hard filtering? And why?