Looking for SCZ-associated NMFs

[shkwon17]

https://jhu-genomics.slack.com/archives/C02RMDJH0MS/p1709666227187879

....Rather than focusing on specificity, is it possible to check or evaluate the sensitivity - if that is the right word - or association of each NMF with SCZ? Attached is a screenshot of data from the McCarroll preprint, which they used to claim that only factor 4 was linked to schizophrenia, and I am considering something like this...that is, I’m curious if there’s anything special about the NMF5 or any other NMFs regarding SCZ...

[shkwon17]

For NMF5:

• we can characterize what layers contribute to NMF5
• we can try DE testing between NTC and SCZ, within the NMF5 domain?

[shkwon17]

https://jhu-genomics.slack.com/archives/C02RMDJH0MS/p1709830762257369?thread_ts=1709665264.747869&cid=C02RMDJH0MS

TODO

• List genes that have high ranks in NMF6 loading.

[shkwon17]

@boyiguo1

Hi Boyi - I promise that this would be my last sudden/demanding request :) ONLY IF we can leverage the established pipeline for DE testing, could you prioritize/try out running a DE testing on NMF6 between NTC and SCZ and pull out the genes at FDR < 5% and 10% or just top 50 - in case we get no DEGs with either of the FDR thresholds?

TODO:

• DE Analysis at the spot-level for each NMF, comparing NTC and SCZ
• DE Analysis at the pseudo-bulk level for each NMF, comparing NTC and SCZ

Please never mind if the pipeline needs to be optimized for this NMF analysis.

I have a strong hunch that NMF6 corresponds to L4 and many of PNN spots will be mapped to this layer. And, like you know this is the layer that we hypothesize is most vulnerable to SCZ. Regardless of the outcome, the results will be very informative/give us heads ups - not to mention that I prefer this NMF6 over the precast spatial domain for layer 4 ^^;;

I am sorry again for this sudden request. Please feel free to let me know what you think about this. I'd greatly appreciate it!

[boyiguo1]

Just FYI, This is not how NMF works. So this part is not possible. I rephrased and added todo items under it.

pull out the genes at FDR < 5% and 10% or just top 50 - in case we get no DEGs with either of the FDR thresholds

[boyiguo1]

@shkwon17 I've done the analysis, checking if NM factors are significant at the single cell level and at sample level. So far, the results is not negative. Neither Factor 5 or 6 are statistically significant.