Introduction to R: Using R for Statistics and Data Analysis. BaRC Hot Topics

Transcription

1 Introduction to R: Using R for Statistics and Data Analysis BaRC Hot Topics

2 Why use R? Perform inferential statistics (e.g., use a statistical test to calculate a p-value) Analyze numbers (vectors and matrices) Create custom figures Automate analysis routines (and make them more reproducible) RStudio: R markdown (Rmd) and knit (html) Reduce copying and pasting Some Unix commands may be easier ask us! Use up-to-date analysis algorithms Real statisticians use it, and it s free! 2 2

3 Why not use R? A spreadsheet application already works fine You re already using another statistics package Ex: Prism, MatLab It s hard to use at first You have to know what commands/syntax to use You don t know how to get started Irrelevant if you re here today 3 3

4 About R Originally written by Ross Ihaka and Robert Gentleman Written in mostly C R is accessible from other languages (Python/Perl) Packages many statisticians/developers have extended R by creating packages (libraries) containing a set of commands/code, data, documentation in a well-defined format: Comprehensive R Archive Network (CRAN) cran.r-project.org BioConductor bioconductor.org Install package: install.packages("packagename") 4

5 Class Object Method About R: object-oriented (OO) systems 5

6 Getting Started: Running R 1. Using Rstudio (tak.wi.mit.edu/rstudio) Enter your tak username and password 2. On the command line (on tak) Log into tak ssh USERNAME@tak -Y Start R by typing R 3. On your own computer Go to R ( Download base from CRAN and install it on your computer Open/install the program Install RStudio (optional) 6 6

7 Start of an R session 7

8 RStudio Interface Editor; R Script Workspace; History Console Output: Plots rstudio.org 8

9 Good practices Save all useful commands and rationale Add comments (starting with # ) Reproducibility - several approaches: Write commands in R and then paste into a text file, or By convention, we end files of R commands with.r Use a specific name for file (ex: compare_wt_ko_weights.r) Start file with #!/usr/bin/rscript to make it a R script In RStudio, save the file as R markdown (.Rmd) and knit (.html) Use the up-arrow to get to previous command Minimize typing, as this increases potential errors. To clear your R window, use Ctrl-L Names of variables should not begin with numbers or underscore, use letters (case sensitive) uppercase + lowercase helps (mywtmice) can include dots (my.wt.mice) 9 9

10 Getting help Use the Help menu Check out Manuals contributed documentation Use R s help and example?median [show info for cmd]??median [search docs] example(median) Search the web r-project median Our favorite book: Introductory Statistics with R (Peter Dalgard) html help 10

11 Useful Commands dir() #list the files in the directory getwd() #get working dir setwd ("/nfs/barc_public") #change working dir history(n) #command history (default n=25) savehistory(file="myfile.txt") #saves as.rhistory ls() #list objects sessioninfo() #print session info about R and loaded packages (useful for knowing version) quit() # quit R Save workspace (.RData file) Assigning value: = or <- X <- 2 X = 2 11

12 Data Types: Mode Mode: components of the same type Commonly used*: character (eg. "dog", "A") integer (eg. 3) numeric (eg. 3.15) logical (eg. TRUE, FALSE) Useful commands: is.numeric(x) #returns if x is TRUE or FALSE as.numeric(x) #converts x to numeric *incomplete 12

13 Data Types/Structure Vector: one-dimensional, most commonly used data structure of the same type chr_vector<-c("dna", "RNA","Protein") List: one-dimensional, elements can be of any type x_list<-list(1,2,"a",c(true,false)) Matrix: two-dimensional, elements of the same type x_matrix<-matrix(1:9, ncol=3,nrow=3) Data frame: two-dimensional, similar to a matrix but elements can be of different type df<-data.frame(x=1:4,y=c("a","b","c","d")) 13

14 Commands to Explore Data head(object) length(object) dim(object) names(object) mode(object) class(object) str(object) # see the top of your data # length of a vector # dimensions of a matrix or data frame # (get or set) names of an object # check if it s numeric, character, etc. #class or type of an object #structure of an object 14

15 Simple Workflow for a t-test # Number of tumors (from litter 2 on 11 July 2010) wt = c(5, 6, 7) ko = c(8, 9, 11)) # Try default t-test settings (Welch's 2-sample t-test) t.test(wt, ko) # Do standard 2-sample t-test t.test(wt, ko, var.equal=t) # Save the results as a variable wt.vs.ko = t.test(wt, ko, var.equal=t) # What are the different parts of this data frame? names(wt.vs.ko) # Just print the p-value wt.vs.ko$p.value # What commands did we use? history(max.show=inf) 15 15

16 Reading Files Take R to your preferred directory On R GUI, go to File > Change dir Check where you are (e.g., get your working directory) and see what files are there > getwd() [1] "X:/bell/Hot_Topics/Intro_to_R" > dir() [1] "compare_wt_ko_weights.r" 16 16

17 Reading Data Files Usually it s easiest to read data from a file Organize in Excel with one-word column names Save as tab-delimited text Check that file is there dir() Read file tumors = read.delim("tumors_wt_ko.txt", header=t) (Other Options: row.names=t, check.names=f) Check that it s OK > tumors wt ko

18 Accessing Data > tumors$wt # Use the column name [1] > tumors[1:3,1] # [rows, columns] [1] > tumors[,1] # missing row or column => all [1] > tumors[1:2,1:2] # select a submatrix wt ko > tumors wt ko > t.test(tumors$wt, tumors$ko) # t-test as before 18 18

19 Creating an output table Most analyses involve several outputs You may want to create a matrix to hold it all Create an empty matrix name rows and columns pvals.out = matrix(data=na, ncol=2, nrow=2) colnames(pvals.out) = c( two.tail", one.tail") rownames(pvals.out) = c("welch", "Wilcoxon") pvals.out two.tail one.tail Welch NA NA Wilcoxon NA NA 19 19

20 Filling the output table (matrix) Do the stats # Welch s test (t-test with pooled variance) pvals.out[1,1] = t.test(tumors$wt, tumors$ko)$p.value pvals.out[1,2] = t.test(tumors$wt, tumors$ko, alt="less")$p.value # Wilcoxon rank sum test (non-parametric alternative to t- test) pvals.out[2,1] = wilcox.test(tumors$wt, tumors$ko)$p.value pvals.out[2,2] = wilcox.test(tumors$wt, tumors$ko, alt="less")$p.value pvals.out two.tail one.tail Welch Wilcoxon

21 Printing the output table We may want to round the p-values pvals.out.rounded = signif(pvals.out, 3) Print the matrix (table) write.table(pvals.out.rounded, file="tumor_pvals.txt", quote=f, sep="\t") Warning: output column names are shifted by 1 when read in Excel 21 21

22 Running a series of commands 1. Copy and paste commands into R session, or 2. Run a script in R, or source("compare_wt_ko_weights.r") [but not so useful in this case, since we aren t creating any files] 3. Run a script from Unix terminal i) method 1 Print output on screen R --vanilla < compare_wt_ko_weights.r Print output in file R --vanilla < compare_wt_ko_weights.r > R_out.txt ii) method 2./compare_WT_KO_weights.R but script must have this as the first line: #!/usr/bin/rscript 22 22

23 User-defined functions Easily perform repetitive task, useful within a script. Syntax myfunction <- function (arg1, arg2, ){ body return(object) } Example #Given a vector, calculate percent for each element calcpercent <- function(x){ percent<- 100*x/sum(x) return(percent) } x<-c(7,5,8,10) calcpercent(x) [1]

24 Introduction to figures R is very powerful and very flexible with its figure generation Any aspect of a figure should be modifiable Some figures aren t available in spreadsheets Boxplot example boxplot(tumors) # Simplest case # Add some more details boxplot(tumors, col=c("gray", "red"), main="mfg appears to be a tumor suppressor", ylab="number of tumors") 24 24

25 Boxplot description RStudio Any points beyond the whiskers are defined as outliers IQR <= 1.5 x IQR 75 th percentile median 25 th percentile Right-click to save figure (for R GUI) 25 25

26 Figure formats and sizes By default, figures on tak are saved as Rplots.pdf Helpful figure names can be included in code To select name and size (in inches) of pdf file pdf( tumor_boxplot.pdf, w=11, h=8.5) boxplot(tumors) # can have >1 page dev.off() # tell R that we re done with this figure To create another format (with size in pixels) png( tumor_boxplot.png, w=1800, h=1200) boxplot(tumors) dev.off() 26 26

27 Useful Packages I: tidyverse Originally written by Hadley Wickham Packages for data analysis dplyr tidyr readr ggplot2 others 27

28 Tidyverse: readr and tidyr readr read in tabular data read_tsv or read_csv data<-read_tsv("normalizedcounts_subset.txt") tidyr "tidy" your data messy_data<-data.frame( gene=c("gapdh", "ACTA1","ZEB1"), heart=c(100,140,450),liver=c(241,10,20) ) gene heart liver 1 GAPDH ACTA ZEB tidy_data<-gather(messy_data,tissue,count,heart:liver) gene tissue count 1 GAPDH heart ACTA1 heart ZEB1 heart GAPDH liver ACTA1 liver 10 6 ZEB1 liver 20 28

29 Tidyverse: dplyr Verb filter select arrange mutate summarise group_by Description Select rows based on criteria Select columns by name Reorder (rows) Create new column Summarise values Group operations 29

30 Useful Packages II: BioConductor Packages/tools for analysis of highthroughput genomic data To see all packages: library() For searchable listing of packages: All require the package to be installed AND explicitly called (or in Rstudio, checked), for example, library(limma) Install what you need on your computer or, for tak, ask the IT group to install packages via

31 Sampling of Popular BioConductor limma biomart Rsamtools edger DESeq/DESeq2 affy topgo ShortRead Packages* *Top75 listing from based on download 31

32 Other Useful Commands library() mean() round(x, n) dir() median() min() length() sd() max() dim() rbind() paste() nrow() cbind() x[x>0] ncol() sort() x[c(1,3,5)] unique() rev() seq(from, to, by) t() log(x, base) commandargs() 32 32

33 R Resources R/Bioconductor short course R scripts (and commands) for Bioinformatics We re glad to share commands and/or scripts to get you started BaRC scripts in \\BaRC_Public\BaRC_code\R Rstudio Cheat Sheets: