sPLS-DA and LASSO regression for feature selection against one dependent variable

Question

[Jonathan Yoou]

Hi all!

I'm working with microbiome data sampled from 5 people everyday for 5 days (i.e 25 samples; 5people per 5days). I also have several demographic metadata and faecal state (e.g pH, water content..) of sampled people.

What I want to know is which factors affect individual genus, and what I'm trying to use is sPLS-DA or LASSO for feature selection.

Here is the problem.

I can get classification error rate from sPLS-DA through 'perf' function, but I only get MSE/MAE from LASSO through 'assess.glmnet' function, and I have no idea which one is better at this dataset.

Is there specific threshold of MSE/MAE I can say which one is better at classification?

Thank you in advance!