search

lareaulab / psix

14 stars 7 forks source link

Getting stuck in psix_object.run_psix #3

Closed nonesuch1936 closed 3 years ago

nonesuch1936 commented 3 years ago

Hi, I have been following your code example on Starsolo. The first step of calling junctions2psi worked well, in the second step "run_psix", the cell.cell metric is successfully calculated, however the Psix score calculation seems to be stuck. I have filtered my dataset to work only on ~1800 interesting cells, and within this the junctions2psi was able to identify 7 exons. With this small number I am not sure where is the hangup. Also I have another question related to the STARsolo approach. I realise in your example you have left the TPM ="", does this mean that for STARsolo I need not have to give any mrna output i.e StarGene/matrix.mtx is not needed? Thanks.

cfbuenabadn commented 3 years ago

Hello,

The TPM ="" in that particular example is a mistake. I was testing something different and forgot to remove that part of the notebook before committing. The notebook cell right above it is the correct one, and it specifies the TPM file. I corrected the notebook to remove the incorrect code. Apologies for the confusion. A TPM file is required for smart-seq2 data. UMI data does not require a TPM file.

Regarding run_psix getting stuck, would you mind sharing which parameters you are using? And what is the last message that you get; is it "Successfully computed cell-cell metric"?

Thanks,

nonesuch1936 commented 3 years ago

Hi, Figured out what I was doing wrong and managed to calculate the psix_scores. These however come out to be negative and the pvals are all 1 ???

cfbuenabadn commented 3 years ago

Could you please let me know what was the issue? It would be helpful if I add an error message, instead of having Psix just hanging.

A negative psix score means that the exon is likely not cell-state associated (i.e., it does not present meaningful biological variation across the cell metric). All p-values being 1 is a bit odd, but not impossible, specially with just a small handful of exons with no cell-state association (it is an empirical p-value obtained by shuffling the data). The q-values being all 1 is not surprising, if nothing is significant. I would suggest that you plot the exons in your low-dimensional representation of the cell state and see if the results make sense (i.e., if the exons do not look cell-state associated). Please let me know if you have any concerns. If there are some exons that look clearly cell-state associated, and are not being recognized as such, I'd probably have to follow up.

May I ask, are you using 10x data?