MPs driven primarily by one gene

[sandcell10]

Hello,

Thank you for creating and maintaining this awesome package! I was running GeneNMF on some data and noticed some MPs that were driven primarily by one gene, such as MP9 and MP10 pictured below. These MPs were considered high quality based on their metaprogram metrics.

I was wondering how these MP’s should be interpreted? Are the main driver genes still meaningfully associated with the other genes in the MP even if those other genes have very small weights? Additionally, would these top driver genes be considered key markers or hub genes of latent expression? Any insight you could provide would be greatly appreciated and thank you for your consideration!

[mass-a]

Hello @sandcell10, thanks for the question.
I can see that some of your MPs have one or few genes with higher weight than the others. This probably means that these genes explain a sizable portion of the data variability when trying to "compress" the data in low-dimensional space. To construct gene signatures for the MPS, our usual approach is to include genes until they together contribute to x% (e.g. 50% or 70%) of the total weight of their MP; so even for MP9 where your top gene VGF has a weight of 0.3, you will have to include several other genes to reach 0.5 or 0.7 of the total weight.
It's difficult to comment on the biological interpretation, I would say that the dominant genes are those that best explain the internal data variability, and that do so consistently across multiple samples. I hope that is useful.

[sandcell10]

Hi @mass-a,

Thank you for the reply! That makes sense. In the graphs, I show only the top 10 genes per MP, though the actual MPs I use contain more genes. The approach I took was to filter out genes from MPs that had a weight less than 0.01. Does this approach make sense? I also like your approach of including genes until they together contribute x%. Will try that out!