Negative correlation between the traits, MTAG?

[jcarmar]

Hello!

I am performing multitrait analysis in an exploratory way after selecting traits based on the significant genetic correlation using LDSC. I performed MTAG analysis for those pairs of traits with a positive genetic correlation, but I was wondering what can I do with those with a negative correlation. Not sure if I can just flip the sign of the Z and analyze it as "positive correlated traits".

What would you suggest?

Thanks!

[paturley]

MTAG can account for negatively genetically correlated phenotypes. You shouldn't need to edit your summary statistics at all (beyond the usual QC). In the original MTAG paper, for example, we use depressive symptoms and life satisfaction.

Best, Patrick

[jcarmar]

Thank you so much for your reply!

[carbocation]

A bit of a late follow-up question, but I imagine that for traits that are near-perfect inverses of one another this will not be the case, and the sumstats need to be munged first. Does that seem right? (E.g., imagine something silly like a radar tracking a runner's velocity from the start of the track and another from the end of the track - they should report the same value but reversed sign. So if you use their readouts as quantitative traits, they'll be nearly perfectly negatively correlated.)

[paturley]

If the phenotypes are exactly negative inverse of each other (or if they have a phenotypic correlation of -1) and there is perfect sample overlap, MTAG should still work, though MTAG will just spit out results that are equal to the original sets of summary statistics. The information that would be gained by the strong negative correlation is offset by the inability of MTAG to distinguish between phenotypic correlation and genetic correlation when there is sample overlap.