R Tutorial: The limma package

Key Takeaways

now you'll learn about the limo package for differential expression using the limo package for differential expression has multiple clear advantages over writing your own bespoke code first testing thousands of genes individually would require writing a loop with boilerplate code like counting variables thus obscuring the intent of the analysis every limo function performs its action on every gene in the dataset making it easier for you to write the code and easier for others to understand your analysis but it's not only more convenient for you it's also better statistics the loop in the code above treats every gene as a completely independent analysis but because all the genes were measured in the same experiment limo uses a statistical technique known as empirical base to share information across the genes which is especially helpful for studies with small sample sizes now empirical Bayes methods would deserve its own separate course so I won't elaborate more but know that not only are you getting convenience but also sophisticated statistics lastly lemma has many functions for processing the data before and after performing the differential expression test you'll learn about some of these in chapter 3 to get started using the limo package install it using BIOS Eli recall that the goal of the breast cancer study is to identify differentially expressed genes between er- tumors in which the estrogen receptor protein is absent versus ER positive tumors in which the estrogen receptor protein is present thus for each gene you need to fit the following linear model the response variable Y is the expression level of the given gene beta naught is the mean expression level of the gene in the ER negative tumors beta 1 is the mean difference in expression level of the gene in the ER positive tumors compared to the ER negative tumors explanatory variable x1 indicates er status an epsilon models the random noise you translate the statistical model to our code using the function model dot matrix in the formula syntax you use to create box plots however this time you only need to specify the explanatory variable the result is called a design matrix and limo will then apply this to every gene here I create a design matrix for the breast cancer data by including the column er as the explanatory variable I don't need to surround it in quotes because I specify the source of the data is the phenotype data frame each column of the design matrix corresponds to a coefficient in the linear model if the sample in a given row is modeled by this coefficient then it has the value 1 and 0 otherwise looking at the first two samples the intercept is 1 because it is 1 for all samples the first sample is er negative it has zero for beta1 conversely the second sample is they are positive and has one summing the columns of the design matrix is a good sanity check the number of samples modeled by the er coefficient should equal the number of er positive samples 209 now that you have specified a design matrix you can proceed with the standard limit pipeline first load the package second fit the coefficients of the model with LM fit by passing it the expression set object and design matrix third calculate the T statistics by passing it the fitted model object to the eBay's function lastly count the number of genes with higher or lower expression an ER positive tumors compared to ER negative there are 5000 for upregulated genes 6000 276 downregulated genes and 11,000 3g news expression level is not different between the tumors note that in a call to decide test I specified the coefficient er since it wouldn't be meaningful to test the intercept coefficient now it's your turn to build and test a model for the Lukey