Negative strand option for liftover

[jigold]

@konradjk @bw2 Can you double check the test examples in test_liftover_negative_strand and give feedback on the interface?

[konradjk]

Did you all decide that mixed return types depending on parameters was ok? I'm fine with it, but I thought there was hesitation. In any case, interface looks fine to me. I'll let @bw2 comment on the exact examples

[konradjk]

Needs an Otherwise, ...

[bw2]

At first glance, the tests look fine to me. More thorough tests could be done using the data files here
https://github.com/broadinstitute/picard/blob/e0bb690d57f73fd2495fc5a77b497e9696f51f81/src/test/java/picard/util/LiftoverVcfTest.java#L65-L99

The interface also looks fine in terms of a non-breaking way to add strand info.
Is the plan to use this to implement a hl.liftover(mt) function that includes the checks from https://github.com/broadinstitute/picard/blob/master/src/main/java/picard/vcf/LiftoverVcf.java#L350-L397 ?

[jigold]

Thanks! I'll look into this later. The plan is to have a liftover_rows method that does the same thing as LiftoverVCF. However, this requires infrastructure to do a while loop, which needs to be built first.

[bw2]

oh, interesting.. I thought liftover processed each variant independently. Just out of curiosity, what's the purpose of the special while-loop infrastructure?
and will it be possible to avoid shuffling when sorting lifted-over variants?

[jigold]

The liftover function is just a mapping from old locus to new locus. The method I'm thinking of implementing will do the left aligning and strand flipping of the alleles as well. I haven't decided if we should be trying to flip the genotypes at all or some of the more complicated things that function does. The while loop is needed to left align the indels in the case the strand flipped. See this code: https://github.com/broadinstitute/picard/blob/dee16518d89838fac40c6f9c07f342430326cb16/src/main/java/picard/util/LiftoverUtils.java#L365-L428

[jigold]

I looked at the picard tests and realized they just made a fake chain file with one interval that mapped to the negative strand in the destination reference. My existing test covers this scenario -- one interval maps to the positive strand of the destination and the other to the negative strand. Since I'm only passing the strand information from Picard, I'm happy with the tests I have already.

[konradjk]

I trust that it's fine then (and if it's not, I'm sure we'll come by to complain at that point 😄). @patrick-schultz over to you