nealgravindra / scnd Goto Github PK

• fig1

• update embedding plot for new cell types (gray-ish one for new hexcode)
• update violin plots for updated clusters (gray-ish one for new hexcode)
• update other plots for new cell type

• more hum data updates

• new ctype embeddings and violin plot
• dge
• rnavel

• fig4

• do latent time for new cell types, recalculate top dyn genes
• update heatmap for top-n dyn genes for the new ctypes
• re-run feature importance score plot for scGAT and potentially re-make the learned graph with connectivities vs. attn heads and make a tutorial so as to respond tot he reviewer comment

• meld
  visually & across rank correlation with ground truth,

• 3 versions of cell rank
  vs.
• [ ]3 versions of monocle
  vs.
• 3 versions of meld to show that ordering works better

• other small things

• dge comp
• scrublet doublet detection, rationale why simulation not good for cerebellum

• embeddings by ctype and other variables of interest
• dge per cell type
• rnavel

Priority

• RNA velocity for human
• get importance scores and regress out md vars
• time-series to incorporate new cell type

human data

• RNA velocity on OPC/OL cluster

mouse data

fig 4

• (fig 4a): jitterplots for pseudotime for ubc --> ubc + dcn
• (fig 4b): calc top dynamical genes for ubc+dcn and make heatmap
• (fig 4c): redo delta vs. pseutdotime lineplots for all ctypes
• (fig 4h): update heatmap for imp genes to ubc --> ubc + dcn
• regress out metadata vars from importance scores, then see which are important still
• (fig 4k) re-plot importance scores vs. time bin so ubc --> ubc + dcn