add a flowchart detailing workflow

[smcclatchy]

I'm working on this and would like to incorporate some feedback about decision-making from one of the end-of-course surveys. Here's what it said:

I think it would be incredibly useful if you could add a checklist for each step, somewhat like a list of sanity checks to keep in mind at each step so you know you've considered the right assumptions for making decisions for your analysis and don't miss anything.

I like the idea of adding this into the workflow. Andreas Mueller did this for machine learning, and I found it helpful. See the following graphic.

[dmgatti]

I have a series of flowcharts highlighting each step in the Single Cell RNASeq lesson. See https://github.com/dmgatti/SingleCellRNAseq/blob/gh-pages/fig/single_cell_flowchart_1.png for one figure that highlights the first step. And https://github.com/dmgatti/SingleCellRNAseq/blob/gh-pages/fig/single_cell_flowchart_*.png for the whole series. These should be at the start of each episode or each major step.

[smcclatchy]

Here is a draft generic workflow

[bswhite]

I would remove '1. Reorder ...'. I think that is referring to aligning the H&E image and the expression data. But most preprocessing pipelines (in particular, Space Ranger) should handle that. Also, I think you wanted to avoid mention of specific technologies -- here, Seurat.

[bswhite]

You probably want a bifurcation point here (between clustering/DEGs and deconvolution). A typical analysis with reduce dimensionality (as you have) -> clustering -> DEGs across clusters. That may additionally be across samples. I will see if I can find a review with a schematic that could guide us.

I would probably move deconvolve cell types after feature selection (i.e., bifurcate following feature selection). That's a bit ambiguous -- it implies deconvolution would be applied to normalized data (earlier in the workflow). That isn't true of RCTD, for example, though it may be true for others. Still, I feel that logically we would normalize data first, just to get a handle on what the data look like. Even if, having normalized the data, we go back and apply deconvolution to the raw counts.

[Elaheh-Alizadeh]

This looks good. Just a minor thing, maybe you can flip the deconvolve part to start from right to left to remove the long arrow.

[Elaheh-Alizadeh]

Looks like each plot's note is on the arrow before the plot. I think the note on the arrow before pseudotime plot belongs to the plot before and does not explain pseudotime.

[dmgatti]

Is there a version of the plot that I can play with? Like an SVG? I feel like the layout could be better, but I'm not quite able to articulate it without playing around. I won't change the original.

[smcclatchy]

Is there a version of the plot that I can play with? Like an SVG? I feel like the layout could be better, but I'm not quite able to articulate it without playing around. I won't change the original.

@dmgatti It's in Biorender, which requires a license and has it's own data format. I can export the graphic as png, pdf or jpg. Could jpg be uploaded to some other software for editing? @dmgatti here is a pdf for upload into some other software spatial-transcriptomics-workflow.pdf

[smcclatchy]

Looks like each plot's note is on the arrow before the plot. I think the note on the arrow before pseudotime plot belongs to the plot before and does not explain pseudotime.

@Elaheh-Alizadeh where is pseudotime on the graphic?