nanostring-biostats / insitucor Goto Github PK

use withr::with_seed()

Specifically, environment scores seem to have contributions from multiple tissues.

remove it, and don't print out the message it currently gives in the attribution analysis

in build_conditionon_matrix

This vignette demonstrates the functionality of InSituCor using a precompiled example dataset. InSituCor can be run using the outputs of the FlatFiles export module [link?], specifically the counts file (_exprMat_file.csv) and the metadata file (_metadata_file.csv). To replicate the vignette with the flat files, the following steps should be followed:

#Load flat files
metadata <- read.csv("S0_metadata_file.csv")
counts <- read.csv("S0_exprMat_file.csv")

#prepare counts file
rownames(counts) <- counts$cell
counts <- counts[,-c(1,2,3)] #remove metadata columns
counts <- as.matrix(counts)

#prepare metadata
metadata$celltype <- metadata$RNA_nbclust_*_clusters #cell type column name will be long
conditionon <- metadata[, c("cell","celltype")]
rownames(conditionon) <- metadata$cell
conditionon$cell <- NULL

#prepare xy data
xy <- annot[,c("cell", "x", "y")] #the x and y coordinates

#cell type data
Celltype <- annot$celltype

• vignette from a bigger dataset, showing more of the post-hoc analysis choices to be made
• show major module names in the network plot, or show hub genes names
• ability to take seurat or tiledb as starting point
• Consider expanding the Hands-on workflow to include interpretations after each step. For example, explanations of the parameters a user can adjust at each step.

Attribution analysis errors when there's only one module. Probably a matrix has been reduced to a vector due to a lack of a drop = FALSE argument.
Fix: put in a sensible error message that advises a higher "res" parameter

These warnings cause slow downs and can turn into errors if the data is counts matrix is large enough.

Could find a way to rewrite these calculations using sparse matrix algebra, or maybe converting the non-zero values in the spaarse matrix to a data.table and doing the calculation on that.

I was able to run through sparc where the cell_IDs in counts and the cell_IDs in conditionon were mismatched.

I think there should be a requirement to check this if the rownames are provided.

Its an easy mistake to make, especially if your 'conditionon' dataframe is reordered after a merge.

use_cells <- met[grep("neuron|L", cell_type),cell_ID]
resmismatch_cell_ID <- 
SPARC::sparc(counts = Matrix::t(normed)[sample(use_cells),] ## counts have out of order cell ids
             ,conditionon = data.frame(met[match(use_cells,cell_ID)
                                           ,.(cell_type = as.character(cell_type)
                                              ,nCount_falsecode
                                              ,nCount_negprobes
                                              ,nCount_RNA)]
                                       ,row.names = use_cells
                                       )
             ,celltype = as.character(met[match(use_cells,cell_ID)][["cell_type"]])
             ,neighbors = NULL
             ,radius = 0.1
             ,xy=as.matrix(met[match(use_cells,cell_ID)][,.(sdimx,sdimy)]
                           ,rownames = use_cells)
             )

Add to readme

Many of the current vignette args aren't the recommended setting, so they can suggest the wrong approach to people.

A module had the probe for "HBA/HBB" in its name. Then attempts to print spatial plot .pngs with this name in the title errored, thinking the "/" was a new path.

Solution: module names should go through make.names()

1: an error that was fixed by forcing the conditionon argument to be a data frame, not a data table:

Error in [.data.table(df, , i) :
j (the 2nd argument inside [...]) is a single symbol but column name 'i' is not found. Perhaps you intended DT[, ..i]. This difference to data.frame is deliberate and explained in FAQ 1.1.

2: "Error in solve.default(cov_bb): system is computationally singluar."
... the matrix coming out of the build_conditionon function had an eigenvalue ~= 0. This shouldn't be happening. The existing countermeasures must not have caught an edge case. Unclear what the edge case is.
(Using bigger neighborhoods, from radius = 0.1 to K = 50, solved this.)

3. Warning about misaligned row names: part comes as a warning, the rest as a message. Make into one warning.

4. The vignette used a radius-based default - change it to K-nearest?