Proposed analysis workflow

[kriemo]

I'd like to open an issue to map out ideas about the general workflow for the analysis. Please edit or comment if you have suggestions. A general goal is to move away from using Seurat, if possible, and replace the workflow used in barnyard_data, with our own approach. To do this, we will need to implement the following:

1) Load in the cell-associated 10x mRNA data and the cell associated and background haircut data into a single R object.

• [X] For now use the MultiAssayExperiment class from BioC. (See create_haircut)

2) Generate QC plots to examine haircut signals across each hairpin from cell-associated and background barcodes.

• [x] add QC plots (reuse plotting code from @mandylr)

3) Implement filtering function to exclude signal from background droplets (see #2)

• [ ] add filtering function

4) Implement functionality to generate 2D cell projections via uMAP (see uzot for a Rcpp implementation) or tSNE and clustering via simple k-means for now.

• [x] Add normalization method for RNA (simple colSums approach should be fine)
• [x] Add normalization method for haircut signal (not sure what's best?, perhaps the centered log ratio method used in CITE-Seq ? )
• [x] Add a scaling and PCA function
• [x] Add a kmeans wrapper
• [x] Add a uMAP/tSNE wrapper

5) Plotting function to visualize cells in uMAP/PCA/tSNE plot. Needs to color cells by mRNA, haircut, and other categorical data (e.g. sample name).

• [x] plotting function (likely can easily modify plot_feature from lung scRNA project)

6) Add in a statistical test to test for differences in haircut/mRNA signal between clusters. Wilcox.test works pretty well.

• [ ] function to call wilcox.test

[jayhesselberth]

Looks likd https://github.com/tkonopka/umap is on CRAN and may be a better choice for UMAP.

[kriemo]

I've added additional functionality to enable basic analysis (see scrunchy vignette).

I'll next work on the #10 to improve handling for more than just a single functional dataset.