In fboulnois/made: Microarray Analysis of Differential Expression

Introduction

Non-small cell lung cancers (NSCLC) are a group of heterogenous diseases with diverse pathological features and multiple histologies. Among the two most common NSCLCs are adenocarcinomas and squamous-cell carcinomas. Both of these histologies are genetically very different. Squamous-cell carcinomas typically have high levels of keratinization, which means that keratin genes may be useful biomarkers for this disease.

We will be downloading a NSCLC dataset and comparing the patterns of expression between controls, adenocarcinomas, and squamous-cell carcinomas, along with identifying potential batch effects or other quality issues in the data.

Before starting the analysis we will want to load the made library:

library(made)

Downloading and extracting the Affymetrix data

The raw Affymetrix data can be downloaded directly from the Gene Expression Omnibus (GEO) under the accession identifier GSE18842. The raw data is available at the following link (400MB).

Extract the Affymetrix data to a folder: this folder will be used as the analysis directory for the rest of the pipeline. Store the full path of this folder in a variable in R.

affymetrixDir <- "full_path_to_Affymetrix_folder"

Defining the configuration and group file

The configuration file, or config, is used by almost all the functions in the made package. The config consists of a set of desired options which can be used to reproducibly analyze a set of oligonucleotide microarrays. Sharing the config with the microarray data allows other researchers to replicate an experiment exactly. The config is generated using the write.yaml.config function.

The write.yaml.config function has many optional parameters (see ?write.yaml.config), but requires at least two parameters: the analysis directory (analysisDir) and how samples should be grouped (groupBy). Since we have individual Affymetrix samples, we will group the samples by "files". If we had multiple folders of samples that we wanted to compare instead, we might want to groupBy = "dirs" instead.

Since all the optional parameters have defaults, we could create the config file and assign each sample to a group manually:

config <- write.yaml.config(analysisDir = affymetrixDir, groupBy = "files")

This can be tedious when there are a large number of samples, however, and so instead we will assign the groups programmatically using the optional parameters contrast.groups and groups.df.

contrast.groups describes how the samples should be compared: in this case, we have controls, adenocarcinomas, and squamous-cell carcinomas. We will create these three groups and compare each to the other:

comparisons <- "adenocarcinoma-control, squamous-control, adenocarcinoma-squamous"

groups.df relates samples with their associated groups in a data.frame. We load a file describing these relationships using information from the original study:

groups <- read.table(system.file("extdata", "example-affymetrix-groups.txt", package = "made"), header = TRUE, comment.char = "#")

We can now create the complete config file:

config <- write.yaml.config(analysisDir = affymetrixDir, groupBy = "files", contrast.groups = comparisons, groups.df = groups)

Running the pipeline

Once the config is generated, the entire pipeline is easily run using the command ma.pipeline. This command will read in and normalize the samples, remove batch effects, summarize the probe data, and generate a report:

ma.pipeline(config)

Note: Supported microarray chips in made depend on one or two additional packages for functionality. These packages are large and unique to each chip type and are thus not automatically installed with made. The two packages are composed of a platform design and an annotation database. The platform design describes the geometry of the chip including the location of probes and is required for Affymetrix arrays. The annotation database maps probes to accession numbers, chromosomal positions, genes, proteins, and pathways, and is required for both Affymetrix and Illumina microarrays.

If the previous step caused an error which mentions that certain packages are missing, install them as follows:

BiocInstaller::biocLite(c("pd.hg.u133.plus.2", "hgu133plus2.db"))

Reading the report

The pipeline generates a microarray report summarizing the details of the experiment. Open this report in any browser and you will see a table of contents followed by several sections. The table of contents can be clicked on to jump to that section. The experimental setup section provides a brief summary of the chosen options and group and sample information. The quality assessment section includes important indicators relating to the overall quality of the microarray experiment. The differential expression section shows the differentially expressed genes between each comparison of interest. The coexpression section identifies correlated patterns in gene expression across samples. The pathway analysis reveals biological terms or pathways that are statistically associated to the differentially expressed genes in each comparison of interest.

The quality assessment section shows that samples are of high quality but that there is a potential source of batch effects. The log-intensity values distribution plot is smooth and the distributions are very similar, indicating that there are no obvious outliers in the samples. On the other hand, two of the samples are misclassified in the hierarchical clustering of samples, indicating that Surrogate Variable Analysis (SVA) should be used to remove any potential sources of batch effects (this option is on by default).

A number of keratin genes (KRTxx) can be observed with large log-fold changes and high-statistical significance (qvalue column) in the comparison between squamous cell carcinoma and the other groups. This indicates that keratinization is an important process in the development of squamous-cell carcinomas.

The most differentially expressed genes are highly correlated across samples in the coexpression heatmap.

Finally, the pathway analysis reveals a number of statistically significant biological terms and pathways relating to the cellular junction suggesting that these pathways are dysregulated between non-small cell lung cancer subtypes.

fboulnois/made documentation built on May 16, 2019, 12:01 p.m.