Confirm the input data type running cNMF

[skye187]

hi, @dylkot
I’d like to clarify the exact data type of the input. The tutorial repeatedly mentions using “count” data, so I assumed that I should prepare the raw read counts from a scRNA-seq dataset (as in Seurat, where slot = "counts"). Is that correct?

However, in the tutorial’s output, there’s a file called spectra_tpm, described as “contains the GEP spectra in units of TPM, and top_genes contains an ordered list of the top 100 associated genes for each program.”

This is confusing to me. Does that mean the output is automatically converted into TPM units? In other words, does the workflow include normalization as part of the NMF process?

Thanks for your clarification.

[dylkot]

Hi, yes, the input data is expected to be integer transcript counts.

One of the outputs is gene expression program spectra in units of TPM. As one part of the computation, the cNMF algorithm normalizes the input data to TPM and refits the spectra in these units.

Does that help?

[skye187]

Hi, yes, the input data is expected to be integer transcript counts.

One of the outputs is gene expression program spectra in units of TPM. As one part of the computation, the cNMF algorithm normalizes the input data to TPM and refits the spectra in these units.

Does that help?

Thank you for answering! It really helped!