In lihuamei/LinDeconSeq: Cell Type Deconvolution using gene expression data for bulk samples.

Desctiption

LinDeconSeq, which consists of 1) identifying marker genes using specificity scoring and mutual linearity strategies across any number of cell types, and 2) predicting cellular fractions of bulk samples using weighted robust linear regression with the marker genes identified in the first stage.

This method is described in the publication from Li et al., 2020 available at https://bmcgenomics.biomedcentral.com/articles/10.1186/s12864-020-06888-1.

Identifying markers for each cell type and build signature matrix.

library(LinDeconSeq)
library(dplyr)
pures <- shen_orr$data[, rowSums(shen_orr$annotation$pure) != 0 ] %>% t
head(pures)[, 1 : 6] %>% print

phes <- shen_orr$phenotype
head(phes) %>% print
markerRes <- findMarkers(pures[, colnames(shen_orr$phenotype)], shen_orr$phenotype, QN = TRUE)
heatmap(markerRes$sigMatrix$sig.mat %>% as.matrix)

Deconvolution for mixture samples.

library(ggplot2)
library(ggpubr)
fractions <- deconSeq(shen_orr$data %>% t, markerRes$sigMatrix$sig.mat, verbose = TRUE)
true.prop <- shen_orr$annotation$mixture[rownames(fractions), colnames(fractions)]

plot.df <- cbind.data.frame(Pred = unlist(fractions), Actual = unlist(data.frame(true.prop)), Sample = rownames(fractions))
ggplot(plot.df, aes(x = Pred, y = Actual)) + geom_point(size = 3, shape = 21, color = 'red') + theme_classic(base_size = 14) + theme(legend.position = 'none') + geom_smooth(method = lm) + stat_cor(method = "pearson")

sessionInfo()

lihuamei/LinDeconSeq documentation built on Nov. 29, 2024, 2:59 p.m.