Non-binarization of matrix for non scATAC-seq data?

[simozhou]

Hello,

I'm working with scCutNTag data of H3K27me3 histone mark, and using 10k tiles to generate a count matrix.

That means that more than one fragment usually falls within each region, and I believe that this quantitative signal could be interesting to capture.

CisTopic is great for non-linear dimensionality reduction with really sparse data (which is my case too!), and in principle Latent Dirichlet Allocation could include the possibility of seeing one word in a document more than once.

Why is there no possibility of using non-binarised matrix when calculating LDA? Could we include the possibility of non-binarising the count matrix?

Thank you!