discrete/dichotomous Es

[Zhengtong-Liu]

We are interested in your method MonsterLM to estimate GxE heritability. It stands out in the field for its innovative approach and its consideration of noise heterogeneity, demonstrating scalability to large-scale biobank datasets. However, we ran into some problems when using the code shared in the GitHub repository. Specifically, the MonsterLM paper suggests applying separate transformations for dichotomous exposures (E) and outcomes (Y) prior to estimating GxE heritability. However, the code available on Github appears to be applicable only to the setting with continuous E and Y variables. We hypothesize that this was the reason for some of the high false positive rates in some of our simulations with discrete E variables. To address this issue, we made four specific modifications to the CPU implementation (outlined below) to incorporate the handling of discrete/dichotomous Es, as described in the paper. Our simulations indicate that these adjustments led to significantly lower false positives. Could you please review the changes we made for accuracy?

Thanks for your consideration.

[discipim]

Hi Zhengtong,

This is an excellent user-friendly update to enable the specific transformations required for a continuous vs dichotomous E. The new input and suggested edits here have been updated to the CPULS script.