01_exploration.Rmd

---
title: 'Tutorial 1: Exploratory analysis of pharmacogenomic data'
author: "Alejandro Reyes"
output:
  html_document: default
  pdf_document: default
---

```{r echo=FALSE}
knitr::opts_chunk$set(cache=FALSE)
```

## Introduction

Probably the most important step of analyzing datasets is to actually understand the data. This process is crucial to know what kind of
questions we can answer with it. This tutorial has code that will help
guiding you through this process. Make sure you understand well the
experimental design of the two studies and try to link each variable
to this experimental design. Also, make sure you understand what each
*R* command is doing. Feel free to hack the code!

If you have any question about the code, ask! Also remember that
[google](www.google.com) is one of the most important tool for data
science.

Let's start by installing the package that contains the data.
Now I am about to modify this line of code


```{r readRaw}

source("http://bioconductor.org/biocLite.R")
biocLite("devtools")

if( !require( PharmPlotter ) ){
  devtools:::install_github("areyesq89/PharmPlotter")
}

library(PharmPlotter)
data("rawData", package="PharmPlotter")
pharmacoData <- rawData
head( pharmacoData )
sessionInfo()
```

What kind of variables are in the data? Are these variables numerical and/or categorical? What does each column represent?

```{r quest2}
head( pharmacoData )
str( pharmacoData )
```

How many drugs are contained in these data?

```{r quest3}
length( levels( pharmacoData$drug ) )
```

How many drug concentrations were used in each study?

```{r quest4}
tapply( pharmacoData$doseID, pharmacoData$study, function(x){
        length( unique( x ) )
    })
```

One of the first things data scientists do when digging into new data
is to explore their distributions.  Histograms visualize the data
distributions and can also point us towards appropiate statistical
models. The code belowtransforms the data into a logarithmic scale and
plots a histogram separately for each study. Based on these plots,
which study would you say has the most consistent experimental
protocol?

```{r quest6, warning=FALSE}
library(ggplot2)
library(cowplot)
ggplot( pharmacoData, aes( log2(concentration) ) ) +
    geom_histogram(fill = "white", colour="black") +
    facet_wrap(~study)
```

Viability scores are the percentage
of cells that survive upon exposure to a certain drug.
Below, we will explore the range of the data and calculate how
many data points are below 0 and above 100.

```{r}
range( pharmacoData$viability )
sum( pharmacoData$viability < 0 )
sum( pharmacoData$viability > 100 )
```
We can also compare the distribution of viability scores between 
the two studies using density plots. Based on the distribution of 
the viability scores, would you say there are obvious differences 
between the two studies?

```{r}
ggplot( pharmacoData, aes( viability, group=study, colour=study) ) +
    geom_density(fill="white", lwd=2, alpha=0.1) + xlim(0, 170)
```

The code below plots the viability scores as box-plots for each drug,
stratified by the two studies. Can you tell something about the toxic
properties of the different drugs? Are these properties consistent
across studies?

```{r}
ggplot( pharmacoData, aes( y=viability, x=drug, fill=study) ) +
    geom_boxplot() +
    theme(axis.text.x = element_text(angle = 90, hjust = 1, vjust=.5)) +
    ylim(0, 200)
```