MCB 585: Multidisciplinary/Quantitative Approaches to Solving Biological Problems: Basic Operation

R is the programming language; RStudio is a user friendly environment for interacting with R.

Using RStudio can make programming in R much more productive.

Consider what working directory you are in when sourcing a script and loading data.

Functions are the basic tool used to manipulate data and other objects in R.

Functions take the form: output <- function(argument1, argument2, ...).

Arguments can be passed to functions by matching based on name, by position, or by omitting them (in which case the default value is used).

Use # to add comments to programs.

Annotate your code!

Name and style code consistently.

Break code into small, discrete pieces.

Factor out common operations rather than repeating them.

Keep all of the source files for a project in one directory and use relative paths to access them.

Keep track of session information in your project folder.

Start each program with a description of what it does.

Load all required packages at the beginning of a script.

Use #---- comments to mark off sections of code.

When working with large datasets, removing variables that you no longer need and releasing memory can improve performance.

Writing functions for sets of commands that you will be running repeatedly will make your data processing more efficient.

Use install.packages() to install new packages and library() to load and use installed packages in your current workspace.

Use help() and ? to get basic information about functions.

Online forums have the answers to almost any R-related question you can come up with. Know how to Google your question!

The University of Arizona offers serveral resources to assist with both biostatistics and programming.

The most commonly encountered data types in R are character, numeric, and logical.

R’s basic data structures are vectors and matrices.

Objects may have attributes, such as name, dimension, and class.

Use object[x] and object[x, y] to select a single element from a 2- and 3-dimensional data structure, respectively.

Use from:to to specify a sequence that includes the indices from from to to.

R’s basic data types are character, numeric, integer, complex, and logical.

R’s data structures include the vector, list, matrix, data frame, and factors. Some of these structures require that all members be of the same data type (e.g. vectors, matrices) while others permit multiple data types (e.g. lists, data frames).

Factors are used to represent categorical data.

Factors can be ordered or unordered.

Some R functions have special methods for handling factors.

The function dim gives the dimensions of a data structure.

All the indexing and subsetting that works on matrices also works on data frames.

Use object[x, y] to select a single element from a data frame.

Each column of a data frame can be directly addressed by specifying the column name using the $ operator (e.g. mydataframe$column1).

Data in data structures can by accessed by specifying the appropriate index, by logical vector, or by column/row name (if defined).

Import data from a .csv or .txt file using the read.table(...), read.csv(...), and read.delim(...) functions.

Write data to a new .csv or .txt file using the write.table(...) and write.csv(...) functions.

Lists are a standard data structure in R in which each element can contain any other R object.

Lists can contain elements of different classes, unlike vectors.

Data frames are a specific type of list in which all elements are vectors of the same length. Each vector can contain data of different classes.

Use object[[x]] to select a single element from a list.

Each element of a list can be assigned a name that can be addressed using the $ operator (e.g. mylist$element1).

Different indexing methods can be combined to efficiently extract desired data subsets for further analysis.

Use mean, max, min, and sd to calculate simple statistics.

Use plot to create simple visualizations.

Display simple graphs.

Save a plot in a pdf file using pdf(...) and stop writing to the pdf file with dev.off().

Use apply(), lapply(), and sapply() to calculate statistics across the rows or columns of a data frame.

There are more advanced tools for complex data manipulation, including the dplyr and data.table packages.

Use aggregate() to calculate statistics based on the structure of a dataset.

Use if (condition) {take action} to start a conditional statement, else if (condition) {take alternative action} to provide additional tests, and else {take alternative action} to provide a default alternative.

Use == to test for equality.

X && Y is only true if both X and Y are true.

X || Y is true if either X, Y, or both, are true.

Use for (variable in collection) {take repeated action} to process the elements of a collection one at a time.

Use regular expressions (regex) to perform pattern searches in textual data.

There are a number of ways to use logical operators, relational operators, and various functions to ask complex and specific questions in the form of if statements.

Two commone ways to index the loop variable in for loops are to use the form for(variable in collection) to pull the elements (“variable”) from a vector (“collection”) one at at time, or to use the form for(i in 1:length(collection)) or for(i in seq_along(collection)) to instead pull each index (i.e. location) in a vector one at a time.

Use functions such as apply instead of for loops to conduct repeated operates on values contained within defined subsets of a data structure.

Use list.files(path = "path", pattern = "pattern", full.names = TRUE) to create a list of files whose names match a pattern.

Samples are sets of observations drawn from a population.

The population distribution describes the characteristics of the observed phenotype in the population of interest.

The sampling distribution describes the characteristics of all possible samples of a given size.

Use hist(), density(), dnorm(), qqnorm(), and qqline() to visually assess whether a sample is normally distributed.

Use shapiro.test() to perform at Shapiro-Wilk test of normality. However, the power of the test is such that a negative result can eitehr indicate insufficient sample size or lack of normality.

Simple manipulations such as taking the logrithm or the square root of a sample can reversibly transform the observations into a more normal distribution.

Understand basic model development and testing.

All statistical tests make assumptions about your sample and population. Understanding these assumptions is critical to running a valid test.

Running multiple comparisons increases you chance of making a Type I Error.

Different multiple test correction strategies correct for different types of errors (Type I vs. Type II) using different strategies.

The basic outcome of a multiple test correction is to lower the P-value threshold (\(lpha\)) below which you reject the null hypothesis.

Time-to-event data includes a hybrid of an observation (event vs. no event) and a series of observation times.

Time-to-event data analysis violates several assumptions made by standard tests, including normality and independence of observations.

Censoring complicates analysis and is not handled by standard statistical tests. Instead, we use the Log-Rank test for basic survival comparisons.

Unlike the survival function, age-specific mortality at any given time does not depend on previous observations.

The survfit() function creates a complex object in R containing life table information. This data can be extracted using the strata variable.

Once extracted, the life table data can be used to calculate and plot age-specific mortality (don’t forget to use log scale on your y-axis!).

The goal of experimental design is to minimize hypothesis testing errors.

The primary tool for improving statistical power is sample size.

R has basic tools for power analysis in the pwr package.

Experiment simulations can be used to understand the behavior of an idealized system (i.e. a system lacking noise from sources like subjective observation and measurement error).

Power analyses can be conducted using pilot data and (potentially) removing assumptions about distribution.

In essence, an experiment simulation is used to build a ‘real’ sampling distribution from which to draw statistical parameters like significance and power.

MCB 585 individual final projects.