Closed hsiaoyi0504 closed 8 months ago
Hi @hsiaoyi0504,
It's a little difficult to debug without seeing the input file. It seems like the algorithm is calling K=0
meaning no latent factor passes the active_thresh
during factorization. You could tinker with that threshold by lowering --active_thresh
, but this normally means something is off about the input file. Additionally - try setting the input K0
to say ~50? We generally set this to at least 2x the expected number of "signatures."
Other questions: does the input matrix contain continuous data and is zero-centered? How many samples do you have (I see you have 4735 input features)?
@shankara-a I shared the data with the email. I will give it a try of the suggestions you told.
The input matrix is continuous and median-centered. There are 103 samples.
Sorry to open up this old thread, but I'm encountering the exact same issue.
I've tried running using the following code :
signatureanalyzer -n 10 -t matrix --objective gaussian --max_iter 30000 --prior_on_H L1 --prior_on_W L1 matrix.tsv
I've tried with the following input matrices - top 10% of the most variable genes from DESeq2 normalized matrix, and log2 transformation of 1 and I end up having the same error. My data is 1847 genes x 61 samples.
Thanks