Hi! 

I am trying to run the 'res.multisplit' function from the hierGWAS package on some preliminary genotype data I have. Upon running the line ' res.multisplit <- multisplit(genomatrix,phenotypevector,covar = covarsmatrix)', the program completes with 21 identical warnings, all reading 'In lognet(x, is.sparse, ix, jx, y, weights, offset, alpha,  ... : one multinomial or binomial class has fewer than 8  observations; dangerous ground'. I can't seem to figure out why I am receiving this error? 

'genomatrix' is my genotype matrix of 36 x 15084, all coded with 0, 1, and 2

'phenotypevector' is my phenotype data vector of 36 length, coded with 0, and 1

'covarsmatrix' is 2 covariates I have in a matrix of 36 x 2 (these are continuous)

sessionInfo()

R version 3.4.1 (2017-06-30)

Platform: x86_64-pc-linux-gnu (64-bit)

Running under: Ubuntu 16.04.4 LTS

Any help is much appreciated, thanks!

The idea behind (multi) sample-splitting is that one splits the data randomly into two halves. The first half is used for variable selection and the second half is used for testing. One then repeats this procedure multiple times, i.e. by default 50 times. The warnings are from the variable selection step where Lasso (implemented in the package glmnet) is applied. Your response is binary and you have only very few observations. Consider one split, i.e. half of the observations, then you have an even smaller number of cases and controls. That’s why the package glmnet / Lasso throws a warning.

You can still try to run it but it might be difficult to find significant variables or clusters because of the very small number of observations.