Organism other than Human or Mouse

[payal banerjee]

Hello,

I have pig brain data. I have clusters and top markers from Monocle3. Can I use Garnett to annotate the clusters?

Thanks,

Payal

[payal banerjee]

Also, does Garnett require pre-annotated file to do the simulations? Like I have a marker file from Monocle3, but the clusters are not identified yet. There are just top genes associated with each cluster.

[Robert Alpin]

You can use whatever species you like so long as your marker file is written using the same annotation as your data set with db = 'none' (as the tutorial's troubleshooting section makes note of), or have the correct org.__.eg.db installed and slotted into the right places. In the case of pigs, org.Ss.eg.db. The use of the database package is used so your marker file doesn't nessisarily need to be in the same annotation style as your dataset.

Monocle doesn't generate a marker file for Garnett. Top markers is generally used for cell type classification inside of Monocle. The marker file fed into Garnett it's something you generally make by hand by taking the canonical marker genes for known cell types from existing papers, or by running Garnett on a known dataset. In either case, it seems to be considered best practices to train your classifier on a known dataset before running it on your 'new' dataset. Unfortunately the Garnett tutorial doesn't make this entirely clear, so it's easy to miss this step.

[payal banerjee]

Thank you Robert. Garnett doesn't have a classifier for pig data. However, this is brain data, can I use their mouse brain data or should I train my data first? How to train my own data?

[Robert Alpin]

You need to make your own. I personally would take another published pig brain paper, download their dataset, and re-create their main clustering figure as best as I could to use Garnett's automatic marker file compilation, and use this to train the classifier on my own data. I do not know how well cross-species classifiers work, or if Garnett will allow it at all. Considering there is only one db= argument, I doubt it. Refer to the tutorial for training.

[payal banerjee]

Thanks! Yes I am going over their tutorials. I just saw a new feature "Train a marker-free classifier" still in beta.

Currently working on annotation!