Incorporating additional metadata features in model

[ksilnoaa]

I'm finding your code and KernelBiome very useful for modeling with metabarcoding data. I'd like to also incorporate other metadata about the samples in the model, such as sampling site as a category and depth as a continuous variable. Can these variables be included as columns in the X dataframe, even though they are not part of the compositional data? How might this impact the model performance and interpretation of CFI?

[shimenghuang]

Hi @ksilnoaa , thank you for the question. One shall not include these variables as columns in the X dataframe directly with the compositional kernels, but it should be possible to use multi-kernel learning to jointly model them. I will look into this a bit further and follow up here. Best, Shimeng