In Goralsth/FloodLight:

The below workflow recapitualtes part of SPOTlights validation workflow, but uses floodlight deconvolution instead.

#Import data
library(TabulaMurisSenisData)
sce <- TabulaMurisSenisDroplet(tissues = "Kidney")$Kidney

# Keep cells from 18m mice
sce <- sce[, sce$age == "18m"]
# Keep cells with clear cell type annotations
sce <- sce[, !sce$free_annotation %in% c("nan", "CD45")]


data<-sce@assays@data@listData[["counts"]]

cell_types<-sce$free_annotation

#Generate average expression profile 
cluster_centers<-average_expression(data, cell_types)

#generate Spots
spots_final2<- generate_spots(data, cell_types)

#deconvolute spots
decon_answer5<-deconvolute_using_NNLS(cluster_centers, spots_final2[["spots"]])

#Get your answer key
decon_answer_key5<-spots_final2[["cell_types"]]

#calculate mean squared error
mean(colSums(decon_answer_key5- decon_answer5)^2)

#get residual of known vs predicted
residual<- (decon_answer_key5-decon_answer5)

#get correlation matrix 
correlation<-cor(decon_answer_key5, decon_answer5)
plotCorrelationMatrix(correlation)

#Plot correlation matrix
heatmap(correlation)

 heatmap(t(residual))

ggplot()+geom_point(aes(x=vec8, y=vec7))

Goralsth/FloodLight documentation built on April 20, 2022, 12:42 a.m.