AMR/vignettes/PCA.Rmd

93 lines
2.5 KiB
Plaintext
Executable File

---
title: "How to conduct principal component analysis (PCA) for AMR"
author: "Matthijs S. Berends"
date: '`r format(Sys.Date(), "%d %B %Y")`'
output:
rmarkdown::html_vignette:
toc: true
toc_depth: 3
vignette: >
%\VignetteIndexEntry{Benchmarks}
%\VignetteEncoding{UTF-8}
%\VignetteEngine{knitr::rmarkdown}
editor_options:
chunk_output_type: console
---
```{r setup, include = FALSE, results = 'markup'}
knitr::opts_chunk$set(
collapse = TRUE,
comment = "#",
fig.width = 7.5,
fig.height = 4.5,
dpi = 100
)
```
**NOTE: This page will be updated soon, as the pca() function is currently being developed.**
# Introduction
# Transforming
For PCA, we need to transform our AMR data first. This is what the `example_isolates` data set in this package looks like:
```{r, message = FALSE}
library(AMR)
library(dplyr)
glimpse(example_isolates)
```
Now to transform this to a data set with only resistance percentages per taxonomic order and genus:
```{r, warning = FALSE}
resistance_data <- example_isolates %>%
group_by(order = mo_order(mo), # group on anything, like order
genus = mo_genus(mo)) %>% # and genus as we do here
summarise_if(is.rsi, resistance) %>% # then get resistance of all drugs
select(order, genus, AMC, CXM, CTX,
CAZ, GEN, TOB, TMP, SXT) # and select only relevant columns
head(resistance_data)
```
# Perform principal component analysis
The new `pca()` function will automatically filter on rows that contain numeric values in all selected variables, so we now only need to do:
```{r pca}
pca_result <- pca(resistance_data)
```
The result can be reviewed with the good old `summary()` function:
```{r}
summary(pca_result)
```
```{r, echo = FALSE}
proportion_of_variance <- summary(pca_result)$importance[2, ]
```
Good news. The first two components explain a total of `r cleaner::percentage(sum(proportion_of_variance[1:2]))` of the variance (see the PC1 and PC2 values of the *Proportion of Variance*. We can create a so-called biplot with the base R `biplot()` function, to see which antimicrobial resistance per drug explain the difference per microorganism.
# Plotting the results
```{r}
biplot(pca_result)
```
But we can't see the explanation of the points. Perhaps this works better with our new `ggplot_pca()` function, that automatically adds the right labels and even groups:
```{r}
ggplot_pca(pca_result)
```
You can also print an ellipse per group, and edit the appearance:
```{r}
ggplot_pca(pca_result, ellipse = TRUE) +
ggplot2::labs(title = "An AMR/PCA biplot!")
```