Automatic marker file creation?

[Eric Kernfeld]

My colleagues have generated scRNA data from some poorly understood transient embryonic states. We are not in a position to manually select markers from the literature -- often, there are no descriptions available near the correct time point. Even if there were, manual annotation defeats the purpose of an automatic annotation tool like this one. Does Garnett have functionality to automatically generate a marker file? If not, how would you recommend doing that? And if I write something up, should I post the code here?

Thanks!

Eric Kernfeld

[Hannah Pliner]

Hi Eric,

Sorry for the delayed reply. Couple of options here: first, Monocle3 has a mechanism for determining the top marker genes given a set of clustering that can then be automatically added to a marker file (https://cole-trapnell-lab.github.io/monocle3/docs/clustering/#garnett) - obviously you'll still need to name the cell types... second (if you're willing to wait a bit), I have recently written a modification to garnett that will generate a model from previously annotated data automatically that can then be applied to future datasets. Again, someone will have to have defined the cell types at some point on an earlier data set. This modification will hopefully come out in an alpha form in the next few weeks.

Hope this helps,

Hannah

[Hannah Pliner]

Hi Eric,

Heads up that I have just pushed the new marker-free version of garnett (in alpha and only in the monocle3 version) that allows you to build a classifier on a dataset with known cell types that you can then apply to new data.

See basic info here: https://cole-trapnell-lab.github.io/garnett/docs_m3/#1c-train-a-marker-free-classifier if you want to give it a try!

Hannah