Clair3 very low F1 INDEL score

Question

[Satrio]

Hello everyone,

I am testing out epi2me-labs/wf-human-snp which utilize Clair3 using this HG002 data. The result then compared to latest truth set using illumina/hap.py. I found the F1 score for INDEL is incredibly low (Precision 35.875%, Recall 3.18%, F1 5.84%), but for SNP it is quite all right (Precision 93.76, Recall 94.79%, F1 94.27%). Previously I have read this article about Clair3 INDEL performance, and it mentions about Clair3 low recall score. My question is this expected with Clair3? Should I use cuteSV instead for INDEL?

Currently running variant calling using epi2me-labs/wf-human-sv.

Thank you

Answer 1

[yussab]

Dear Satrio,

I'm having a similar issue. How did you managed INDEL calls from Clair3? The problem is that I have it also with shorter indels from my comparison Did you proceed with cuteSV?

Thanks in advance, best regards! AY