6 Applications

6.1 Principal component analysis

The wCV values directly affect to AUC in the restricted MCD when the fitted regression lines are very different per sample. Since AUC is used to get SQI, we can check how PC scores of wCV are related to SQI. After handling missing values, let’s keep genes having positive variance within samples.

wCV.mat2 <- wCV.mat
wCV.mat2[is.na(wCV.mat2)] <- 0
wCV.mat2 <- wCV.mat2[apply(wCV.mat2, 1, var)!=0, ]

stats::prcomp function performs PCA, and its result can be applied to calculate pc1_contributions to figure out contributions of each gene to the PC1.

pca_result <- prcomp(t(wCV.mat2), scale=TRUE)
pc1_contributions <- abs(pca_result$rotation[, 1])
top_genes <- order(pc1_contributions, decreasing=TRUE)

6.2 window CV heatmap

A heatmap from ComplexHeatmap helps to visualize patterns of the wCV matrix with genes and samples properties. To construct row and column annotations, additional variables can be added by matching genes and samples:

• Robust CV from CVs per level
• AUC, PD, and SQI from sample quality index
• PC1, PC2, and PC1 contributions from principal component analysis

The columns are split as 2 clusters by dendrogram, which is corresponding to most of bad quality samples.

6.3 Table

table1::table1 function displays descriptive statistics by a categorical variable. We can simply review mean and range of sample properties by sample quality for general comparison.

library(table1)
table1(~ AUC + PD + RINs + RatioIntron + robustCV | SQI, 
       data=SplAnnomat,
       render.continuous=function(x) {sprintf("%.2f [%.2f, %.2f]", mean(x, na.rm=TRUE), min(x, na.rm=TRUE), max(x, na.rm=TRUE))}
       )

	Bad (N=19)	Good (N=152)	Overall (N=171)
AUC	0.83 [0.67, 1.01]	0.48 [0.42, 0.59]	0.52 [0.42, 1.01]
PD	6.79 [3.74, 10.34]	-0.23 [-2.25, 2.13]	0.55 [-2.25, 10.34]
RINs	2.79 [1.10, 4.60]	5.77 [1.30, 9.20]	5.44 [1.10, 9.20]
Missing	1 (5.3%)	3 (2.0%)	4 (2.3%)
RatioIntron	1.32 [0.57, 2.80]	1.94 [0.46, 4.07]	1.87 [0.46, 4.07]
robustCV	0.55 [0.48, 0.60]	0.36 [0.29, 0.44]	0.38 [0.29, 0.60]