Strelka SomaticEVS filtering threshold about lcr-modules

[lkhilton]

Hey there,

We chose that threshold in part because a collaborator working on a similar consensus variant calling method found that it was an optimal balance of sensitivity and specificity. We accepted the tradeoff so that we could reduce false positives, especially in FFPE data. We found Strelka2 is a pretty lenient variant caller with default settings and we got a lot of noise with the default filtering in place.

Do you have examples of particular indels that are recurrently missing in the outputs, or is there no pattern to which indels drop out? Are they called robustly by SAGE?

Cheers,
Laura

[nuttynutmore]

Hi Laura,

Thank you for those insights- that's really helpful.

Our initial output returned no indels when we used the somaticEVS threshold of 10. When we remove the threshold filtering, we get indels in the output, some of which are also called by SAGE. We're currently troubleshooting to find the optimal threshold for our data.

Thanks again for your thoughts and for the great suite of modules.
Kind regards,
Natasha