Question on deconvolution using RCTD

[wzdrgi]

RCTD is a very wonderful piece of work and I have recently used RCTD to deconvolve my spatial transcriptomic data, but the inputs to the reference dataset have also raised a question for me. It is very difficult for the spatial transcriptomic data to appear identical to the reference dataset in terms of cell types, so I am curious to know how RCTD manages cell types that are present in the spatial data but absent in the reference dataset. Looking forward to your reply!