Differential Expression

Raw Counts

Fix header for fc files:

python fixExonFC.py /project2/gilad/briana/Comparative_APA/Human/data/RNAseq/ExonCounts/RNAseqOrthoExon.fc /project2/gilad/briana/Comparative_APA/Human/data/RNAseq/ExonCounts/RNAseqOrthoExon.fixed.fc

python fixExonFC.py /project2/gilad/briana/Comparative_APA/Chimp/data/RNAseq/ExonCounts/RNAseqOrthoExon.fc /project2/gilad/briana/Comparative_APA/Chimp/data/RNAseq/ExonCounts/RNAseqOrthoExon.fixed.fc

HumanCounts=read.table("../Human/data/RNAseq/ExonCounts/RNAseqOrthoExon.fixed.fc", header = T, stringsAsFactors = F) %>% select(-Chr,-Start,-End,-Strand, -Length)

ChimpCounts=read.table("../Chimp/data/RNAseq/ExonCounts/RNAseqOrthoExon.fixed.fc", header = T, stringsAsFactors = F) %>% select(-Chr,-Start,-End,-Strand, -Length)


counts_genes=HumanCounts %>% inner_join(ChimpCounts,by="Geneid") %>% column_to_rownames(var="Geneid")

head(counts_genes)

                NA18504 NA18510 NA18523 NA18498 NA18499 NA18502 NAPT30
ENSG00000188976      24      50      31      69      34      58      2
ENSG00000188157     106      65     106      12     128      39      5
ENSG00000273443      40      43      54      26      48      11     21
ENSG00000217801      60      36     164      62      61      19     34
ENSG00000237330       0       1       0       0       1       1      0
ENSG00000223823       0       0       0       0       0       0      0
                NAPT91 NA3622 NA3659 NA4973 NA18358
ENSG00000188976      1      1      1      0       0
ENSG00000188157      7      7      9     34       6
ENSG00000273443      2      3     78     59      18
ENSG00000217801      8     19    139     68      31
ENSG00000237330      0      0      2      1       0
ENSG00000223823      0      0      0      0       0

# Load colors 

colors <- colorRampPalette(c(brewer.pal(9, "Blues")[1],brewer.pal(9, "Blues")[9]))(100)

pal <- c(brewer.pal(9, "Set1"), brewer.pal(8, "Set2"), brewer.pal(12, "Set3"))
labels <- paste(metaData$Species,metaData$Line, sep=" ")

#PCA function (original code from Julien Roux)
#Load in the plot_scores function
plot_scores <- function(pca, scores, n, m, cols, points=F, pchs =20, legend=F){
  xmin <- min(scores[,n]) - (max(scores[,n]) - min(scores[,n]))*0.05
  if (legend == T){ ## let some room (35%) for a legend                                                                                                                                                 
    xmax <- max(scores[,n]) + (max(scores[,n]) - min(scores[,n]))*0.50
  }
  else {
    xmax <- max(scores[,n]) + (max(scores[,n]) - min(scores[,n]))*0.05
  }
  ymin <- min(scores[,m]) - (max(scores[,m]) - min(scores[,m]))*0.05
  ymax <- max(scores[,m]) + (max(scores[,m]) - min(scores[,m]))*0.05
  plot(scores[,n], scores[,m], xlab=paste("PC", n, ": ", round(summary(pca)$importance[2,n],3)*100, "% variance explained", sep=""), ylab=paste("PC", m, ": ", round(summary(pca)$importance[2,m],3)*100, "% variance explained", sep=""), xlim=c(xmin, xmax), ylim=c(ymin, ymax), type="n")
  if (points == F){
    text(scores[,n],scores[,m], rownames(scores), col=cols, cex=1)
  }
  else {
    points(scores[,n],scores[,m], col=cols, pch=pchs, cex=1.3)
  }
}

# Clustering (original code from Julien Roux)
cors <- cor(counts_genes, method="spearman", use="pairwise.complete.obs")


heatmap.2( cors, scale="none", col = colors, margins = c(12, 12), trace='none', denscol="white", labCol=labels, ColSideColors=pal[as.integer(as.factor(metaData$Species))], RowSideColors=pal[as.integer(as.factor(metaData$Collection))+9], cexCol = 0.2 + 1/log10(15), cexRow = 0.2 + 1/log10(15))

Past versions of unnamed-chunk-11-1.png

select <- counts_genes
summary(apply(select, 1, var) == 0) 

   Mode   FALSE    TRUE 
logical   32975   11150 

# Perform PCA

pca_genes <- prcomp(t(counts_genes), scale = F)
scores <- pca_genes$x


#Make PCA plots with the factors colored by species

### PCs 1 and 2 Raw Data
for (n in 1:1){
  col.v <- pal[as.integer(metaData$Species)]
  plot_scores(pca_genes, scores, n, n+1, col.v)
}

Past versions of unnamed-chunk-13-1.png

### PCs 3 and 4 Raw Data

for (n in 3:3){
  col.v <- pal[as.integer(metaData$Species)]
  plot_scores(pca_genes, scores, n, n+1, col.v)
}

Past versions of unnamed-chunk-14-1.png

Plot density for raw data:

density_plot_18504 <- ggplot(counts_genes, aes(x = NA18504)) + geom_density() + labs(title = "Density plot of raw gene counts of NA18504") + labs(x = "Raw counts for each gene")

density_plot_18504

Past versions of unnamed-chunk-15-1.png

Convert to log2

log_counts_genes <- as.data.frame(log2(counts_genes))
head(log_counts_genes)

                 NA18504  NA18510  NA18523  NA18498  NA18499  NA18502
ENSG00000188976 4.584963 5.643856 4.954196 6.108524 5.087463 5.857981
ENSG00000188157 6.727920 6.022368 6.727920 3.584963 7.000000 5.285402
ENSG00000273443 5.321928 5.426265 5.754888 4.700440 5.584963 3.459432
ENSG00000217801 5.906891 5.169925 7.357552 5.954196 5.930737 4.247928
ENSG00000237330     -Inf 0.000000     -Inf     -Inf 0.000000 0.000000
ENSG00000223823     -Inf     -Inf     -Inf     -Inf     -Inf     -Inf
                  NAPT30   NAPT91   NA3622   NA3659   NA4973  NA18358
ENSG00000188976 1.000000 0.000000 0.000000 0.000000     -Inf     -Inf
ENSG00000188157 2.321928 2.807355 2.807355 3.169925 5.087463 2.584963
ENSG00000273443 4.392317 1.000000 1.584963 6.285402 5.882643 4.169925
ENSG00000217801 5.087463 3.000000 4.247928 7.118941 6.087463 4.954196
ENSG00000237330     -Inf     -Inf     -Inf 1.000000 0.000000     -Inf
ENSG00000223823     -Inf     -Inf     -Inf     -Inf     -Inf     -Inf

density_plot_18504 <- ggplot(log_counts_genes, aes(x = 18504)) + geom_density()

density_plot_18504 + labs(title = "Density plot of log2 counts of 18504") + labs(x = "Log2 counts for each gene") + geom_vline(xintercept = 1) 

Past versions of unnamed-chunk-17-1.png

plotDensities(log_counts_genes, col=pal[as.numeric(metaData$Species)], legend="topright")

Past versions of unnamed-chunk-18-1.png

Convert to CPM

cpm <- cpm(counts_genes, log=TRUE)
head(cpm)

                   NA18504   NA18510   NA18523    NA18498   NA18499
ENSG00000188976  0.9947425  1.625971  1.213485  1.9873949  1.075299
ENSG00000188157  3.0661207  1.990970  2.931252 -0.3353280  2.923740
ENSG00000273443  1.6951397  1.417837  1.980752  0.6524534  1.547681
ENSG00000217801  2.2614645  1.174529  3.552519  1.8381797  1.880283
ENSG00000237330 -3.0036033 -2.442815 -3.003603 -3.0036033 -2.450115
ENSG00000223823 -3.0036033 -3.003603 -3.003603 -3.0036033 -3.003603
                   NA18502     NAPT30     NAPT91     NA3622     NA3659
ENSG00000188976  1.8489000 -2.0183629 -2.3993239 -2.4294571 -2.4761762
ENSG00000188157  1.3004469 -1.2173800 -0.7890014 -0.8590437 -0.6897168
ENSG00000273443 -0.3506575  0.4928914 -1.9747241 -1.7012034  2.1431252
ENSG00000217801  0.3378719  1.1383145 -0.6357420  0.3591647  2.9586805
ENSG00000237330 -2.4371643 -3.0036033 -3.0036033 -3.0036033 -2.0907904
ENSG00000223823 -3.0036033 -3.0036033 -3.0036033 -3.0036033 -3.0036033
                   NA4973    NA18358
ENSG00000188976 -3.003603 -3.0036033
ENSG00000188157  1.082327 -0.8045812
ENSG00000273443  1.841053  0.5540567
ENSG00000217801  2.039212  1.2860108
ENSG00000237330 -2.447731 -3.0036033
ENSG00000223823 -3.003603 -3.0036033

plotDensities(cpm, col=pal[as.numeric(metaData$Species)], legend="topright")

Past versions of unnamed-chunk-20-1.png

Log2 CPM

TMM/log2(CPM)

## Create edgeR object (dge) to calculate TMM normalization  
dge_original <- DGEList(counts=as.matrix(counts_genes), genes=rownames(counts_genes), group = as.character(t(labels)))
dge_original <- calcNormFactors(dge_original)

tmm_cpm <- cpm(dge_original, normalized.lib.sizes=TRUE, log=TRUE, prior.count = 0.25)
head(cpm)

                   NA18504   NA18510   NA18523    NA18498   NA18499
ENSG00000188976  0.9947425  1.625971  1.213485  1.9873949  1.075299
ENSG00000188157  3.0661207  1.990970  2.931252 -0.3353280  2.923740
ENSG00000273443  1.6951397  1.417837  1.980752  0.6524534  1.547681
ENSG00000217801  2.2614645  1.174529  3.552519  1.8381797  1.880283
ENSG00000237330 -3.0036033 -2.442815 -3.003603 -3.0036033 -2.450115
ENSG00000223823 -3.0036033 -3.003603 -3.003603 -3.0036033 -3.003603
                   NA18502     NAPT30     NAPT91     NA3622     NA3659
ENSG00000188976  1.8489000 -2.0183629 -2.3993239 -2.4294571 -2.4761762
ENSG00000188157  1.3004469 -1.2173800 -0.7890014 -0.8590437 -0.6897168
ENSG00000273443 -0.3506575  0.4928914 -1.9747241 -1.7012034  2.1431252
ENSG00000217801  0.3378719  1.1383145 -0.6357420  0.3591647  2.9586805
ENSG00000237330 -2.4371643 -3.0036033 -3.0036033 -3.0036033 -2.0907904
ENSG00000223823 -3.0036033 -3.0036033 -3.0036033 -3.0036033 -3.0036033
                   NA4973    NA18358
ENSG00000188976 -3.003603 -3.0036033
ENSG00000188157  1.082327 -0.8045812
ENSG00000273443  1.841053  0.5540567
ENSG00000217801  2.039212  1.2860108
ENSG00000237330 -2.447731 -3.0036033
ENSG00000223823 -3.003603 -3.0036033

pca_genes <- prcomp(t(tmm_cpm), scale = F)
scores <- pca_genes$x

for (n in 1:2){
  col.v <- pal[as.integer(metaData$Species)]
  plot_scores(pca_genes, scores, n, n+1, col.v)
}

Past versions of unnamed-chunk-22-1.png

Past versions of unnamed-chunk-22-2.png

# Plot library size

boxplot_library_size <- ggplot(dge_original$samples, aes(x=metaData$Species, y = dge_original$samples$lib.size, fill = metaData$Species)) + geom_boxplot()
 
boxplot_library_size + labs(title = "Library size by Species") + labs(y = "Library size") + labs(x = "Species") + guides(fill=guide_legend(title="Species"))

Past versions of unnamed-chunk-23-1.png

plotDensities(tmm_cpm, col=pal[as.numeric(metaData$Species)], legend="topright")

Past versions of unnamed-chunk-24-1.png

Filter low expressed gene

Filter based on log2 cpm

filter log2(cpm >1) in at least 10 of the samples (2/3)

#filter counts
keep.exprs=rowSums(tmm_cpm>1) >8

counts_filtered= counts_genes[keep.exprs,]




plotDensities(counts_filtered, col=pal[as.numeric(metaData$Species)], legend="topright")

Past versions of unnamed-chunk-25-1.png

labels <- paste(metaData$Species, metaData$Line, sep=" ")
dge_in_cutoff <- DGEList(counts=as.matrix(counts_filtered), genes=rownames(counts_filtered), group = as.character(t(labels)))
dge_in_cutoff <- calcNormFactors(dge_in_cutoff)

cpm_in_cutoff <- cpm(dge_in_cutoff, normalized.lib.sizes=TRUE, log=TRUE, prior.count = 0.25)
head(cpm_in_cutoff)

                 NA18504    NA18510  NA18523  NA18498  NA18499   NA18502
ENSG00000217801 2.243163  1.0859263 3.672233 1.802011 1.859359 0.2120175
ENSG00000186891 5.022046  5.0400892 4.902343 6.682104 5.710517 3.3639006
ENSG00000186827 3.102236  4.6574228 1.712312 1.802011 3.968056 1.9042287
ENSG00000078808 6.868747  7.0337697 7.475369 7.036034 6.625443 6.7359729
ENSG00000176022 4.762642  4.6839390 4.801525 5.267970 4.613121 4.7028975
ENSG00000184163 1.427504 -0.1617807 1.712312 1.404540 1.062680 0.3544851
                   NAPT30     NAPT91    NA3622   NA3659   NA4973  NA18358
ENSG00000217801 1.1675009 -0.7849597 0.2795297 3.003758 2.101032 1.307542
ENSG00000186891 4.6403916  3.1655019 4.7307807 6.452972 6.478017 6.488794
ENSG00000186827 0.2591792  2.0129015 3.1544019 0.651737 5.254278 2.181924
ENSG00000078808 6.6946199  6.7706146 6.7797936 6.789003 7.211007 7.067711
ENSG00000176022 4.8386000  5.5562504 4.8490853 5.281245 4.790317 5.568370
ENSG00000184163 2.8929268  2.2016666 1.4133401 1.771404 1.743030 3.229013

hist(cpm_in_cutoff, xlab = "Log2(CPM)", main = "Log2(CPM) values for genes meeting the filtering criteria", breaks = 100 )

Past versions of unnamed-chunk-27-1.png

Voom transformation:

Species <- factor(metaData$Species)
design <- model.matrix(~ 0 + Species)
colnames(design) <- gsub("Species", "", dput(colnames(design)))

c("SpeciesChimp", "SpeciesHuman")

# Voom with individual as a random variable

cpm.voom<- voom(counts_filtered, design, normalize.method="quantile", plot=T)

Past versions of unnamed-chunk-29-1.png

boxplot(cpm.voom$E, col = pal[as.numeric(metaData$Species)],las=2)

Past versions of unnamed-chunk-30-1.png

plotDensities(cpm.voom, col =  pal[as.numeric(metaData$Species)], legend = "topleft") 

Past versions of unnamed-chunk-31-1.png

Looks like i still have a skew on the lower side of the distribution.

# PCA 

pca_genes <- prcomp(t(cpm.voom$E), scale = F)
scores <- pca_genes$x


eigsGene <- pca_genes$sdev^2
proportionG = eigsGene/sum(eigsGene)

plot(proportionG)

Past versions of unnamed-chunk-32-1.png

for (n in 1:2){
  col.v <- pal[as.integer(metaData$Species)]
  plot_scores(pca_genes, scores, n, n+1, col.v)
}

Past versions of unnamed-chunk-32-2.png

Past versions of unnamed-chunk-32-3.png

#Clustering (original code from Julien Roux)
cors <- cor(cpm.voom$E, method="spearman", use="pairwise.complete.obs")


heatmap.2( cors, scale="none", col = colors, margins = c(12, 12), trace='none', denscol="white", labCol=labels, ColSideColors=pal[as.integer(as.factor(metaData$Species))], RowSideColors=pal[as.integer(as.factor(metaData$Species))+9], cexCol = 0.2 + 1/log10(15), cexRow = 0.2 + 1/log10(15))

Past versions of unnamed-chunk-33-1.png

This is wierd. Normalization moves 2 samples to opposite species clusters but the samples that separate in the correlation are not those samples. 4973 and 18498 are the samples that looked funny on the original 3’ data. This may be a sample swap at the RNA stage. These samples were in the same extraction batch. It could have happened then. I will look into this more.

One thing I can do is look at the correlation between the PCs and other factors in the data.

# PCA 

pca_genes <- prcomp(t(cpm.voom$E), scale = F)
scores <- pca_genes$x

for (n in 1:2){
  col.v <- pal[as.integer(metaData$Collection)]
  plot_scores(pca_genes, scores, n, n+1, col.v)
}

Past versions of unnamed-chunk-34-1.png

Past versions of unnamed-chunk-34-2.png

metaData$Extraction=as.factor(metaData$Extraction)

for (n in 1:2){
  col.v <- pal[as.integer(metaData$Extraction)]
  plot_scores(pca_genes, scores, n, n+1, col.v)
}

Past versions of unnamed-chunk-35-1.png

Past versions of unnamed-chunk-35-2.png

It does not look like batch (who collected or extraction date batch)

cols = brewer.pal(9, "Blues")
palC = colorRampPalette(cols)


metaData$UndilutedAverageorder = findInterval(metaData$UndilutedAverage, sort(metaData$UndilutedAverage))
for (n in 1:2){
  col.v <- palC(nrow(metaData))[metaData$UndilutedAverageorder]
  plot_scores(pca_genes, scores, n, n+1, col.v)
}

Past versions of unnamed-chunk-36-1.png

Past versions of unnamed-chunk-36-2.png

metaData$BioAConcorder = findInterval(metaData$BioAConc, sort(metaData$BioAConc))
for (n in 1:2){
  col.v <- palC(nrow(metaData))[metaData$BioAConcorder]
  plot_scores(pca_genes, scores, n, n+1, col.v)
}

Past versions of unnamed-chunk-37-1.png

Past versions of unnamed-chunk-37-2.png

metaData$RinConcorder = findInterval(metaData$Rin, sort(metaData$Rin))
for (n in 1:2){
  col.v <- palC(nrow(metaData))[metaData$RinConcorder]
  plot_scores(pca_genes, scores, n, n+1, col.v)
}

Past versions of unnamed-chunk-38-1.png

Past versions of unnamed-chunk-38-2.png

The samples do not cluster by collection concentration, RNA rin score or RNA concentration.

metaData$AssignedOrthoorder = findInterval(metaData$AssignedOrtho, sort(metaData$AssignedOrtho))
for (n in 1:2){
  col.v <- palC(nrow(metaData))[metaData$AssignedOrthoorder]
  plot_scores(pca_genes, scores, n, n+1, col.v)
}

Past versions of unnamed-chunk-39-1.png

Past versions of unnamed-chunk-39-2.png

They also do not cluster by number of reads mapping to ortho exons.

I should look at the correlation between PCs and these factors.

Try to run the pca the opposite way:

pca_line=prcomp(cpm.voom$E, center=T,scale=T)

pca_line_df=as.data.frame(pca_line$rotation) %>% rownames_to_column(var="Line") %>% select(1:11)


eigs <- pca_line$sdev^2
proportion = eigs/sum(eigs)

plot(proportion)

Past versions of unnamed-chunk-40-1.png

metaData_order=metaData %>% arrange(Line)
PCA_order=pca_line_df %>% arrange(Line)
Pc1Spec <- summary(lm(PCA_order$PC1 ~ metaData_order$Species))$adj.r.squared
Pc2Spec <- summary(lm(PCA_order$PC2 ~ metaData_order$Species))$adj.r.squared
Pc3Spec <- summary(lm(PCA_order$PC3 ~ metaData_order$Species))$adj.r.squared
Pc1Spec

[1] -0.09882106

Pc2Spec

[1] 0.6285914

Pc3Spec

[1] -0.0714836

Expand this to full heatmap.

plotpca

col.v <- pal[as.integer(metaData$Species)]
plot_scores(pca_line, scores, n, n+1,cols = col.v )

Past versions of unnamed-chunk-42-1.png

Question: what is the difference between these PCAs???