flo-compbio / moana-tutorials Goto Github PK

Hi,
I've read the tutorial documents and re run it at local, but I didn't see any information about how to create the classifier. If the script or tutorial about producing PBMC classifier is available, could you update it to repository? Thank you!

Hi,

I am trying to apply multiple cell assignment methods and trying to understand which works best with my data set. I was wondering if there is any way to create a custom training data for classification using moana?

Thanks!

Hi.

Your prediction notebook raises suspicion about your work and may discourage exploration or adoption. The classifier could be failing for highly similar cell types, but I'd recommend investigating a more likely suspect first: by going down to 20 PCs, especially after smoothing, you've probably removed all the variation there was left to find. I would try open/fast-tSNE or UMAP without PCA, maybe ivis, maybe 100 PCs.

Minor quibble, but for the same reason: the same notebook loads the data, does t-SNE, loads the data again, does t-SNE again. Why not save the t-SNE coordinates the first time, predict, add the predicted classes to the points, and plot again with color?

Hi. Nice work, and thanks for making the notebooks available. It seems that you have a couple of dependencies.

You may want to have a section in each notebook for file and directory names. I had your notebooks below my home directory, so I had to fix the shifts from ~/data/... to data/... within and between notebooks. Also, the ribosome gene files were installed (using pip install) in ~/bin/moana/moana/data/gene_lists while the notebooks look for them in ~/anaconda3/lib/python3.7/site-packages/moana/data/gene_lists.