An accessible GenePattern notebook for the copy number variation analysis of Illumina Infinium DNA methylation arrays

Implementation

The entire workflow is implemented as a GenePattern notebook, which can be accessed at the GenePattern Notebook Repository (http://www.genepattern-notebook.org/) and run there by the user. Data preprocessing and CNV analysis steps are implemented as a GenePattern module (Reich et al., 2006) and utilized by the MethylationCNVAnalysis notebook.

Load and preprocess data

To begin the analysis, two sets of data are required: the query sample data for which the copy number profiles are to be analyzed and appropriate control sample data used to establish baseline copy number profiles for comparison (Figure 2). The input data for this notebook (query and control samples) are raw IDAT files generated by the microarray scanner, representing two different color channels prior to normalization. As described in the minfi documentation, IDAT files are the most complete data types, because they include measurements on control probes, which are necessary for assessing bisulfite conversion efficiency and for normalizing technical variability.

Figure 2. Copy number variation analysis GenePattern Notebook interface.

The “MethylationCNVAnalysis” module is presented as an input form using the GenePattern Notebook graphical user interface. The user links or uploads input files and selects analysis parameters before pressing “Run” to execute the workflow.

To load the Illumina Infinium methylation array data into the notebook, the IDAT files must be combined into a single archive (.zip or .gz formats). The archive can be organized either as a flat archive where all IDAT files are packed without subfolders, or as an archive in the standard folder structure as presented in the Illumina demo dataset. The IDAT archive can be selected and loaded through the graphical user interface of the GenePattern notebook. Both 450k array and EPIC array types are compatible as long as all samples in a single archive are of the same array type. If the query samples or control samples are of different array types, only the common set of probes between 450k and EPIC array types are evaluated across all samples.

For each sample, the data is normalized with respect to background and positive control probes on the arrays according to the implementation in Illumina’s proprietary GenomeStudio software. Upon loading the data, the notebook generates a quality control report containing two plots for identifying poor quality samples. The first plot shows the log₂ median intensity of the methylated versus unmethylated channels (Figure 3A). Poor-quality samples tend to have lower median intensities and separate from the good quality samples. The second plot shows the DNA methylation levels (Beta values) of all probes on the array and for all samples as a density plot in which we expect to see a bimodal distribution with peaks at zero (no methylation) and one (100% methylation) (Figure 3B).

Figure 3. Plots of query and control samples.

(A) Median intensity plot of query & control samples. Log median intensity of the methylated channel is along the x-axis and log median intensity of unmethylated channel is along the y-axis. Bad-quality samples fall under the threshold and are colored red. There is no bad quality sample in this plot. (B) DNA methylation (Beta-value) density plot of query & control samples. A density plot showing the distribution of beta values across each sample. Beta values should be bimodal and peak around 0 and 1.0.

Control samples should be free of CNVs and have a similar methylation profile as the samples of interest. The best practice is to use control samples of the corresponding normal tissue type. If control samples are included in the query sample dataset, no separate data needs to be loaded. Instead, the control samples can be specified by providing the sample names in the CNV analysis step. Otherwise, the control data will be loaded as a separate archive of IDAT files.

CNV analysis

As outlined in the conumee documentation, the copy number analysis is performed as follows: each query sample is normalized to the control samples by multiple linear regression yielding the linear combination of control samples that most closely fits the intensities of the query sample. Next, the log₂ ratio of probe intensities of the query sample versus the combination of control samples are calculated. Probes are then combined within predefined genomic bins. Intensity values are shifted to minimize the median absolute deviation of all bins to zero to determine the copy-number neutral state. The genome is segmented into regions of the same copy number state using the circular binary segmentation algorithm (Seshan & Olshen, 2018).

Genomic loci of genes to be highlighted in the CNV plots are retrieved from the hg19 Ensembl database using the BiomaRt R package (Durinck, 2005; Durinck, 2009). The notebook also offers an option to exclude regions from analysis, such as highly polymorphic regions that would yield inaccurate copy number calls. In addition, X and Y chromosomes can be excluded to avoid misleading results in case no appropriate control data is available.

Operation

To run the MethylationCNVAnalysis notebook, the user must have a GenePattern account that can be created on the GenePattern Notebook website (http://genepattern-notebook.org). After logging in, the notebook can be found in the “Community” section of the “Public Notebooks” page. The notebook can then be run from the GenePattern Notebook site, with no additional software installations needed.