Last updated: 2018-08-29

workflowr checks: (Click a bullet for more information)

Read data.

library(data.table)
tmp = setDF(fread('data/unnecessary_in_building/GSM1626794_P14Retina_2.digital_expression.txt'))
orig = tmp[,-1]; rownames(orig) = tmp[,1]; rm(tmp)
genenames = sapply(strsplit(rownames(orig), ":"), function(x) x[3])
gc(verbose=FALSE);
            used (Mb) gc trigger   (Mb) limit (Mb)  max used  (Mb)
Ncells   2789945  149    4252298  227.1         NA   4252298 227.1
Vcells 106685452  814  155388100 1185.6      16384 107643487 821.3

Quality Control and Cell Filter

nGene = colSums(orig > 0)
hist(nGene)

summaryX = cellFilter(X = orig, 
                      genenames = genenames,
                      minGene = 500, 
                      maxGene = 2000, 
                      maxMitoProp = 0.1)
tmpX = summaryX$X
nUMI = summaryX$nUMI
nGene = summaryX$nGene
percent.mito = summaryX$percent.mito
det.rate = summaryX$det.rate

par(mfrow = c(1,4))
boxplot(nUMI, main='nUMI'); 
boxplot(nGene, main='nGene'); 
boxplot(percent.mito, main='mitochondrial gene', ylim=c(0,0.5)); 
boxplot(det.rate, main='detection rate', ylim=c(0,0.1))

Gene filtering

X = tmpX[rowSums(tmpX) > 0, ]
genenames = genenames[rowSums(tmpX) > 0]

#gene filter by dispersion
disp = dispersion(X, bins = 20)
plot(disp$z ~ disp$genemeans,
     xlab = "mean expression",
     ylab = "normalized dispersion")

select = which(abs(disp$z) > 1)
X = X[select, ]
genenames = genenames[select]

UMI Normalization

Use quantile-normalization to make the distribution of each cell the same.

nX = quantile_normalize(as.matrix(X))

Correct Detection Rate

After normalization, the linear relationship between the first PC and the detection rate usually disappears. The plots show that even without correction, the two are not heavily correlated. So we do not regress out the detection rate.

#take log
logX = as.matrix(log(nX + 1))

#check dependency
out = correct_detection_rate(logX, det.rate)

#regress out
# log.cpm = out$residual
log.cpm = logX

Dimension reduction on the data for visualization.

pc = irlba(log.cpm, 20)
plot(pc$d, ylab = "singular values")

tsne = Rtsne(pc$v[,1:5], dims=2, perplexity = 100, pca=FALSE)
plot(tsne$Y, xlab = 'tsne1', ylab = 'tsne2', cex = 0.5)

rm(pc)

Run SLSL on the log.cpm matrix.

# out = SLSL(log.cpm, log=FALSE,
#                 filter = FALSE,
#                 correct_detection_rate = FALSE,
#                 klist = c(300,350,400),
#                 sigmalist = c(1,1.5,2),
#                 kernel_type = "pearson",
#                 verbose=FALSE)
# save(out, file = 'dropseq_slsl1.Rdata')
load('analysis/dropseq_slsl1.Rdata')
tab = table(out$result)
plot(tsne$Y, col=rainbow(length(tab))[out$result],
     xlab = 'tsne1', ylab='tsne2', main="SLSL", cex = 0.5)

Using SC3 with our gene and cell filter procedure and disabling the biology and filter feature of the function leads to the following result. Using SC3’s default methods lead to the error “distribution of gene expression in cells is too skewed towards 0”.

library(SingleCellExperiment)
library(SC3)
# colnames(X) = paste0("C", 1:ncol(X))
# sce = SingleCellExperiment(
#   assays = list(
#     counts = X,
#     logcounts = as.matrix(log(X+1))
#   ),
#   colData = colnames(X)
# )
# rowData(sce)$feature_symbol = genenames
# sce = sc3_prepare(sce, kmeans_nstart = 50)
# sce = sc3_estimate_k(sce)
# k = metadata(sce)$sc3$k_estimation
# N = ncol(X)
# sce = sc3(sce, ks=k, biology = FALSE, gene_filter=FALSE,
#          kmeans_nstart=10)
# col_data_selected = colData(sce)
# save(col_data_selected, file  = "analysis/dropseq_sc3_result.Rdata")
load('analysis/dropseq_sc3_result.Rdata')
plot(tsne$Y, col=rainbow(k)[col_data_selected$sc3_6_clusters],
     xlab = 'tsne1', ylab='tsne2', main="SC3", cex = 0.5)
ind = which(genenames %in% c('Chat', 'Gad1', 'Gad2', 'Slc17a8', 'Slc6a9', 'Gjd2'))
df = data.frame(tsne1 = tsne$Y[,1], tsne2 = tsne$Y[,2],
                Slc17a8 = log.cpm[ind[1], ],
                Gad2 = log.cpm[ind[2], ],
                Gad1 = log.cpm[ind[3],]
                )
g1 = ggplot(df, aes(x=tsne1, y=tsne2, col = Slc17a8)) + geom_point(size=0.1) +
  scale_colour_gradient(low="pink", high="black") + guides(color = FALSE) + ggtitle('Slc17a8')
g2 = ggplot(df, aes(x=tsne1, y=tsne2, col = Gad2)) + geom_point(size=0.1) +
  scale_colour_gradient(low="pink", high="black") + guides(color = FALSE) + ggtitle('Gad2')
g3 = ggplot(df, aes(x=tsne1, y=tsne2, col = Gad1)) + geom_point(size=0.1) +
  scale_colour_gradient(low="pink", high="black") + guides(color = FALSE) + ggtitle('Gad1')
grid.arrange(g1,g2,g3, nrow=1)

Session information

sessionInfo()
R version 3.5.1 (2018-07-02)
Platform: x86_64-apple-darwin15.6.0 (64-bit)
Running under: macOS Sierra 10.12.5

Matrix products: default
BLAS: /Library/Frameworks/R.framework/Versions/3.5/Resources/lib/libRblas.0.dylib
LAPACK: /Library/Frameworks/R.framework/Versions/3.5/Resources/lib/libRlapack.dylib

locale:
[1] en_US.UTF-8/en_US.UTF-8/en_US.UTF-8/C/en_US.UTF-8/en_US.UTF-8

attached base packages:
[1] parallel  stats     graphics  grDevices utils     datasets  methods  
[8] base     

other attached packages:
 [1] bindrcpp_0.2.2          data.table_1.11.4      
 [3] gridExtra_2.3           gdata_2.18.0           
 [5] stargazer_5.2.2         abind_1.4-5            
 [7] broom_0.5.0             gplots_3.0.1           
 [9] diceR_0.5.1             Rtsne_0.13             
[11] igraph_1.2.2            scatterplot3d_0.3-41   
[13] pracma_2.1.4            fossil_0.3.7           
[15] shapefiles_0.7          foreign_0.8-71         
[17] maps_3.3.0              sp_1.3-1               
[19] caret_6.0-80            lattice_0.20-35        
[21] reshape_0.8.7           dplyr_0.7.6            
[23] ggplot2_3.0.0           irlba_2.3.2            
[25] Matrix_1.2-14           quadprog_1.5-5         
[27] inline_0.3.15           matrixStats_0.54.0     
[29] SCNoisyClustering_0.1.0

loaded via a namespace (and not attached):
 [1] nlme_3.1-137              bitops_1.0-6             
 [3] lubridate_1.7.4           dimRed_0.1.0             
 [5] rprojroot_1.3-2           tools_3.5.1              
 [7] backports_1.1.2           R6_2.2.2                 
 [9] KernSmooth_2.23-15        rpart_4.1-13             
[11] lazyeval_0.2.1            colorspace_1.3-2         
[13] nnet_7.3-12               withr_2.1.2              
[15] tidyselect_0.2.4          compiler_3.5.1           
[17] git2r_0.23.0              labeling_0.3             
[19] caTools_1.17.1.1          scales_0.5.0             
[21] sfsmisc_1.1-2             DEoptimR_1.0-8           
[23] robustbase_0.93-2         stringr_1.3.1            
[25] digest_0.6.15             rmarkdown_1.10           
[27] R.utils_2.6.0             pkgconfig_2.0.1          
[29] htmltools_0.3.6           rlang_0.2.1              
[31] ddalpha_1.3.4             bindr_0.1.1              
[33] gtools_3.8.1              mclust_5.4.1             
[35] ModelMetrics_1.1.0        R.oo_1.22.0              
[37] magrittr_1.5              Rcpp_0.12.18             
[39] munsell_0.5.0             R.methodsS3_1.7.1        
[41] stringi_1.2.4             whisker_0.3-2            
[43] yaml_2.2.0                MASS_7.3-50              
[45] plyr_1.8.4                recipes_0.1.3            
[47] grid_3.5.1                pls_2.6-0                
[49] crayon_1.3.4              splines_3.5.1            
[51] knitr_1.20                pillar_1.3.0             
[53] reshape2_1.4.3            codetools_0.2-15         
[55] stats4_3.5.1              CVST_0.2-2               
[57] magic_1.5-8               glue_1.3.0               
[59] evaluate_0.11             RcppArmadillo_0.8.600.0.0
[61] foreach_1.4.4             gtable_0.2.0             
[63] purrr_0.2.5               tidyr_0.8.1              
[65] kernlab_0.9-26            assertthat_0.2.0         
[67] DRR_0.0.3                 gower_0.1.2              
[69] prodlim_2018.04.18        class_7.3-14             
[71] survival_2.42-6           geometry_0.3-6           
[73] timeDate_3043.102         RcppRoll_0.3.0           
[75] tibble_1.4.2              iterators_1.0.10         
[77] workflowr_1.1.1           lava_1.6.2               
[79] ipred_0.9-6

This reproducible R Markdown analysis was created with workflowr 1.1.1