Categorical covariates with more than two levels

[Luming-L]

Hi,

My input file contains several categorical covariates with more than two levels. For example, the covariate smoker has three levels: "non-smoker", "past smoker" and "current smoker". When running DeepNull, I got this error:

Cast string to float is not supported

After a long search, I realised that only numbers are accepted in regression. It seems that converting strings to numbers has not been embedded in DeepNull yet. Therefore, my questions are:

• If I want to include categorical covariates with more than two levels, should I recode these categorical covariates before running DeepNull?
• If yes, which encoding type is more proper for unordered categorical variable, one-hot encoding, dummy encoding or anything else?

Thanks!

[fhormoz]

Hi Luming,

Thank for you interested in DeepNull.

It is true that DeepNull can not deal with string categorical covariates similar to most GWAS pipeline methods (EMMA, BOLT-LMM, GEMMA, REGENIE, etc).

I would use the same encoding that you will use for your GWAS analysis. I personally would prefer to use dummy encoding.

Thanks,
Farhad

[Luming-L]

Hi Farhad,

Thanks for the clarification! Dummy encoding also makes sense to my analysis.

Cheers,
Luming