glennhickey
commented
5 years ago 100% agreed to make everyone's output explicit and do all comparisons with --normalize. I think that's simplest and fairest and shouldn't be that hard even if it requires a new script or two.
On Wed, Mar 13, 2019 at 3:32 AM Jean Monlong notifications@github.com wrote:
In toilvg, we used to normalize the truth and call sets before evaluating the calls. Then we discussed not normalizing anything to be fair with the calls from methods that produced symbolic VCFs that can't be directly normalized. This affects SVTyper.
On the other hand, BayesTyper was run on normalized variants. And I think there is a graph normalization step during the graph construction (right @glennhickey https://github.com/glennhickey?), that might also affects the variants positions.
In the end the easiest (and still fair) way might be to convert the output from SVTyper to an explicit VCF and run the usual normalize+evaluation. Jonas told me BayesTyper has a script to convert to explicit VCF.
— You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub https://github.com/jmonlong/manu-vgsv/issues/39, or mute the thread https://github.com/notifications/unsubscribe-auth/AA2_7gFPxwKUtGDLzXpDacsLJXcOjxAbks5vWKmAgaJpZM4bskfl .
In toilvg, we used to normalize the truth and call sets before evaluating the calls. Then we discussed not normalizing anything to be fair with the calls from methods that produced symbolic VCFs that can't be directly normalized. This affects SVTyper.
On the other hand, BayesTyper was run on normalized variants. And I think there is a graph normalization step during the graph construction (right @glennhickey?), that might also affects the variants positions.
In the end the easiest (and still fair) way might be to convert the output from SVTyper to an explicit VCF and run the usual normalize+evaluation. Jonas told me BayesTyper has a script to convert to explicit VCF.