Validation of filtration of low-quality cells on the inDrop Mouse PCs dataset

Load data

d_folders <- c('SRR3879617/est_11_22/', 'SRR3879618/est_11_22/', 'SRR3879619/est_11_29/')
rds_paths <- paste0(kDataPath, d_folders, 'cell.counts.rds')
holders <- mclapply(rds_paths, readRDS, mc.cores=3)

cms <- lapply(holders, `[[`, 'cm')
common_genes <- Reduce(intersect, lapply(cms, rownames))
cms <- lapply(cms, function(x) x[common_genes,])

cm_union <- LibrariesUnion(cms, colnames, `colnames<-`, function(x) Reduce(cbind, x))
scores <- mclapply(holders, ScorePipelineCells, mit.chromosome.name='chrM', 
                   mc.cores=length(holders)) %>% LibrariesUnion()

mit_fracs <- lapply(holders, `[[`, 'reads_per_chr_per_cells') %>% lapply(`[[`, 'Exon') %>%
  lapply(GetChromosomeFraction, 'chrM') %>% LibrariesUnion()

umis_per_cb <- sort(Matrix::colSums(cm_union), decreasing=T)

published_csv <- url(kCsvLink) %>% gzcon() %>% readLines() %>% paste0(collapse='\n') %>% 
  data.table::fread(data.table=F)

published_csv <- published_csv[, 1:3] %>%
  mutate(LibId = substr(V1, 11, 11) %>% as.integer(),
         Barcode = gsub("-", "", barcode) %>% paste0('-', LibId),
         Cluster = gsub("_", " ", assigned_cluster)) %>%
  dplyr::select(-V1, -barcode, -LibId, -assigned_cluster)

cluster_by_barcodes <- setNames(published_csv$Cluster, published_csv$Barcode)
dim(published_csv)

[1] 1064    2

r_cm <- cm_union[, order(Matrix::colSums(cm_union), decreasing=T)]

cell_mask <- setNames(rep(FALSE, ncol(r_cm)), colnames(r_cm))
cell_mask[names(scores)[scores > kRescuedScoreThreshold]] <- TRUE
cell_mask[intersect(names(cluster_by_barcodes), names(cell_mask))] <- TRUE
r_cm <- r_cm[, cell_mask]

r_cm <- r_cm[Matrix::rowSums(r_cm) > 20, ]

cluster_by_barcodes <- cluster_by_barcodes[names(cluster_by_barcodes) %>% 
                                             intersect(colnames(r_cm))]

r <- GetPagoda(r_cm, n.cores=10)

1167 cells, 14393 genes; normalizing ... using plain model winsorizing ... log scale ... done.
calculating variance fit ... using gam 333 overdispersed genes ... 333 persisting ... done.
running PCA using 1000 OD genes .... done
calculating distance ... pearson ...running tSNE using 10 cores:

PlotPagodaEmbeding(r, colors=mit_fracs, show.legend=T, title='Mitochondrial fraction',
                   alpha=0.9, size=0.5, font.size=4.5) +
  scale_color_continuous(name='Fraction') +
  theme_pdf(legend.pos=c(1, 1))

Number of cells

umi_counts_union_full <- sort(colSums(cm_union), decreasing=T)
umi_counts_union <- umi_counts_union_full[1:5000]

gg1 <- PlotCellsNumberHist(umi_counts_union_full, breaks=80, estimate.cells.number=T) +
  scale_x_continuous(name='#UMI in cell', labels=round(10^(2:4)), breaks=2:4) + 
  theme_pdf()
gg2 <- PlotCellsNumberLine(umi_counts_union_full, breaks=80, estimate.cells.number=T) + 
  theme_pdf() + theme(legend.position="none")
gg3 <- PlotCellsNumberLogLog(umi_counts_union_full, estimate.cells.number=T) + 
  theme_pdf() + theme(legend.position="none")

gg_cell_num <- cowplot::plot_grid(gg1, gg2, gg3, ncol=1, nrow=3, labels="AUTO")

ggsave(paste0(kOutputFolder, 'figures/supp_cell_num_srr3.pdf'), gg_cell_num, 
       width=5, height=7)

gg_cell_num

Rescued cells

filtered_cbs <- intersect(names(cluster_by_barcodes), 
                          names(scores)[scores <= kFilteredScoreThreshold])
rescued_cbs <- setdiff(names(scores)[scores > kRescuedScoreThreshold], 
                       names(cluster_by_barcodes))
intersect_cbs <- intersect(names(r$clusters$PCA$infomap), names(cluster_by_barcodes))

notannotated_cells <- setdiff(colnames(r_cm), names(cluster_by_barcodes))
clusters_annotated_resc <- AnnotateClustersByGraph(r$graphs$PCA, cluster_by_barcodes, 
                                            notannotated_cells,
                                            max.iter=100, mc.cores=10)
rescued_clusters <- clusters_annotated_resc[rescued_cbs]

plot_clusters <- cluster_by_barcodes[names(scores)[scores > kFilteredScoreThreshold]]
plot_rescued_clusters <- rescued_clusters

for (type in c("immune other", "activated stellate", "quiescent stellate")) {
  plot_clusters[plot_clusters == type] <- gsub(" ", "\n", type)
  plot_rescued_clusters[plot_rescued_clusters == type] <- gsub(" ", "\n", type)
}

gg_tsne <- PlotFiltrationResults(r, clusters=plot_clusters, 
                                 rescued.clusters=plot_rescued_clusters,
                                 filtered.cbs=filtered_cbs,
                                 unchanged.alpha=1, rescued.alpha=1, lineheight=0.7,
                                 raster.width=7 / 1.75, raster.height=6 / 1.3) +
  theme_pdf(legend.pos=c(0, 1), show.ticks=F)

gg_tsne

rescued_table <- TableOfRescuedCells(clusters_annotated_resc[c(intersect_cbs, rescued_cbs)], 
                                     rescued_cbs)
write.csv(rescued_table, paste0(kOutputFolder, "tables/rescued_cbc_srr3.csv"), row.names=F)
rescued_table

srt <- CreateSeuratObject(raw.data=r_cm, min.cells=10, min.genes=0, display.progress=F)
srt <- NormalizeData(object = srt, normalization.method = "LogNormalize", 
                     scale.factor = 10000, display.progress=F)
srt <- FindVariableGenes(object=srt, mean.function=ExpMean, dispersion.function=LogVMR,
                         x.low.cutoff = 0.0125, x.high.cutoff = 3, y.cutoff = 1, 
                         do.plot=F, display.progress=F)
srt <- ScaleData(object = srt, vars.to.regress = "nUMI", display.progress=F)

srt@ident <- as.factor(clusters_annotated_resc[colnames(srt@raw.data)])
names(srt@ident) <- colnames(srt@raw.data)
compared_clusters <- c('beta', 'delta', 'alpha', 'gamma')
cluster_markers <- mclapply(compared_clusters, function(i) 
  mclapply(setdiff(compared_clusters, i), FindClusterMarkers, i, srt, mc.cores=3), 
  mc.cores=4)
overexpressed_genes <- GetOverexpressedGenes(srt, compared_clusters, cluster_markers, 
                                             genes.from.cluster=50, 
                                             expression.threshold=0.3)
length(overexpressed_genes)

[1] 174

Heatmaps

clusts_after_filt <- cluster_by_barcodes[setdiff(names(cluster_by_barcodes), filtered_cbs)]
real_clusts <- c(clusts_after_filt, rescued_clusters)

selected_clusters <- compared_clusters

tested_clusts <- sort(real_clusts[real_clusts %in% selected_clusters])
separation <- c(setNames(rep('rescued', length(rescued_cbs)), rescued_cbs),
                setNames(rep('real', length(clusts_after_filt)), names(clusts_after_filt)))

umis_per_cb_subset <- log10(Matrix::colSums(r_cm[, names(tested_clusts)]))
tested_clusts <- tested_clusts[order(tested_clusts, -umis_per_cb_subset)]
de_genes <- intersect(overexpressed_genes, colnames(r$counts))

plot_df <- ExpressionMatrixToDataFrame(r$counts[names(tested_clusts), de_genes], 
                                       umis_per_cb_subset, tested_clusts, rescued_cbs)
# plot_df <- plot_df %>% filter(UmisPerCb < 3.5)
plot_dfs <- split(plot_df, plot_df$IsRescued) %>% setNames(c('real', 'rescued'))

ggs <- lapply(plot_dfs, HeatmapAnnotGG) %>%
  lapply(lapply, `+`, theme(plot.margin=margin()))

legend_guides <- list(HeatmapLegendGuide('Expression'),
                      HeatmapLegendGuide('Cell type', guide=guide_legend, nrow=4),
                      HeatmapLegendGuide('log10(#molecules)'))
gg_legends <- mapply(`+`, ggs$real, legend_guides, SIMPLIFY=F) %>%
  lapply(`+`, theme(legend.margin=margin(l=4, r=4, unit='pt'))) %>% lapply(get_legend)

ggs$real$heatmap <- ggs$real$heatmap + rremove('xlab') + ylab('Cells')
ggs$rescued$heatmap <- ggs$rescued$heatmap + labs(x = 'Genes', y = 'Cells')
ggs_annot <- lapply(ggs, function(gg) plot_grid(plotlist=lapply(gg, `+`, rremove('legend')), 
                                                nrow=1, rel_widths=c(1.5, 0.12, 0.12),
                                                align='h'))

Figure

gg_legends_plot <- plot_grid(plotlist=gg_legends[c(1, 3)], nrow=2, align='v') %>%
  plot_grid(gg_legends[[2]], ncol=2, rel_widths=c(2, 1))

gg_left <- plot_grid(ggs_annot$real, ggs_annot$rescued, nrow=2, labels='AUTO')
gg_right <- plot_grid(gg_tsne + theme(plot.margin=margin(l=0.1, unit='in')), gg_legends_plot,
                      nrow=2, rel_heights=c(1, 0.3), align='v', labels=c('C'))

gg_fig <- plot_grid(gg_left, gg_right, rel_widths=c(0.8, 1)) +
  theme(plot.margin=margin(1, 1, 1, 1))

gg_fig

ggsave(paste0(kOutputFolder, 'figures/supp_lq_srr3.pdf'), gg_fig, width=7, height=6)

Session information