pca.Rmd

---
title: "Principal Deaths Analysis (PDA)"
author: "The Team / Leo"
date: "`r Sys.Date()`"
output:
  pdf_document
---


```{r setup, echo=FALSE, message=FALSE, warning=FALSE}
knitr::opts_chunk$set(echo = TRUE)
library(devtools)
library(gridExtra)
library(phyloseq)
library(dplyr)
library(reshape2)
library(knitr)
library(plotROC)
library(pROC)
library(ggROC)
library(tidyr)
library(microbiome)
library(cowplot)
library(ggplot2)
require(bookdown)
knitr::opts_chunk$set(fig.path = "figure_manuscript/", dev="CairoPNG", cache=FALSE)
knitr::opts_chunk$set(fig.align = "center")
# http://derekogle.com/fishR/2015-09-17-Figure-Table-Captions-in-Markdown
library(captioner)
tbls <- captioner(prefix="Table")
figs <- captioner(prefix="Fig.")

subtbls <- captioner(prefix="Supplementary Table")
subfigs <- captioner(prefix="Supplementary Fig.")

qth <- 0.1
```


The 2nd PCA axis on CLR-transformed abundance data seems to be associated with death rate.


```{r pca, echo=FALSE, message=FALSE, warning=FALSE, fig.width=14, fig.height=10}
library(spdep)
# Compare to PCoa and rotated PCoA
#pca$scores <- Rotation(phyla.df[, c("pcoa1", "pcoa2")], .2 * pi)
#pca$scores <- Rotation(phyla.df[, c("pcoa1", "pcoa2")], 0 * pi)
#pca$cores
# phyla.df[, c("pcoa1", "pcoa2")]
# t(abundances(transform(phfinrisk_genus, trans))[c("Prevotella (Bacteria)", "Bacteroides (Bacteria)"), ])

# Visualize PCA
p <- plot_landscape(pca$scores[, 1:2]) +
       labs(x = "PCA1",
            y = "PCA2",
	    title = paste("PCA on genus-level data (", trans, "/", distmeth, ")"))

print(p)
```


```{r pcacomp, echo=FALSE, message=FALSE, warning=FALSE, fig.show="hold", out.width="200px", fig.width=6, fig.height=5}
for (k in 1:2) {
  plot(density(pca$scores[,k]), main = paste("Component", k, "; 10% quantile indicated"))
  abline(v = quantile(pca$scores[,k], qth), lty = 2)
}
```



```{r drivers, echo=FALSE, message=FALSE, warning=FALSE, fig.show="hold", out.width="200px", fig.width=8, fig.height=7}
for (k in 1:2) {
  s <- sort(pca$loadings[,k])
  s <- s[abs(s) > quantile(abs(s), 0.99)]
  par(mar = c(3, 15, 1, 1))
  barplot(sort(s), main = paste("Drivers on component", k), horiz = T, las = 1)
}
```


```{r deathbar, echo=FALSE, message=FALSE, warning=FALSE, fig.show="hold", out.width="200px", fig.width=8, fig.height=6}
library(knitr)
for (k in 1:2) {

 # Define the lowest 10% quantile on the given PCA axis
 df <- meta(phfinrisk_genus)
 df$quantile <- rep("High", nrow(df))
 df$quantile[pca$scores[,k] < quantile(pca$scores[,k], qth)] <- "Low"
 df$quantile <- factor(df$quantile, levels = c("Low", "High"))

 pval <- fisher.test(table(df$DEATH, df$quantile))$p.value

 tab <- prop.table(table(Death = df$DEATH, Quantile = df$quantile), 2)
 kable(tab, digits = 2)

 dfm <- subset(melt(tab), Death == TRUE)
 p <- ggplot(dfm, aes(x = Quantile, y = value)) +
        geom_bar(stat = "identity", position = "dodge") +
	scale_y_continuous(labels = scales::percent) +
	labs(x = "Low PCA 2",
	     y = "Deaths (%)",
	     title = paste("Deaths on PCA", k, "(low vs. high; p=", round(pval, 3), ")"))
 print(p)
}
```


ROC/AUC for selected phyla on PCA2.

```{r ROC, echo=FALSE, message=FALSE, warning=FALSE, fig.show="hold", out.width="200px", fig.width=5, fig.height=3}
library(pROC)
k <- 2
for (phyla in selected.phyla[1:4]) {
  r <- roc(gsub(" \\(Bacteria\\)", "", names(pca$loadings[,k])) %in% tt.m[which(tt.m[, "Phylum"] %in% phyla), "Genus"], pca$loadings[,k])
  plot(r, main = paste(phyla, "; AUC=", round(r$auc, 2)))
}
```