US FlagAn official website of the United States government
National Cancer Institute
MicroArray Analysis Pipeline (MAAPster)For Affymetrix human and mouse data

Microarrays

MicroArray Analysis Pipeline (MAAPster) is a comprehensive web tool designed by the CCR Collaborative Bioinformatics Resource (CCBR) that performs transcriptome analysis of human or mouse Affymetrix gene expression data. Samples may be uploaded locally or pulled from publicly available data by entering a GEO Series identification number. MAAPster is able to analyze data from a single experiment, or GEO Series, and allows for the investigation of multiple contrasts between groups of samples. Raw CEL files are analyzed using several Bioconductor packages in 'R', including limma, oligo, and gsva (Carvalho & Irizarry, 2010; Hanzelmann, Castelo, & Guinney, 2013; Ritchie et al., 2015). Output includes quality control plots, normalized gene expression data, differential gene expression analysis, gene expression heatmaps, pathway analysis and single sample GSEA analysis.

Input

There are 2 input options:

  1. Local: Analyze expression data by uploading CEL files from a local computer. Data must be expression data in CEL file format, and all data must be from the same type of Affymetrix chip.

  2. GEO: Analyze expression data from the Gene Expression Omnibus (GEO) https://www.ncbi.nlm.nih.gov/geo/ by providing the relevant series ID from the experiment, as noted below. Enter the entire ID, starting with GSE. Samples from different chips cannot be analyzed together. If the series ID contains samples from multiple chips, choose the type of chip from the drop-down menu. Samples analyzed on each chip will populate as the chip is chosen:

    The following chips are supported, and additional Affymetrix mouse or human chips may be added upon request. Currently, microRNA and single-cell analyses are not supported.

Human

Mouse

Analysis

Once the files are uploaded, choose groups for each sample. At least 2 groups must be created. The control group should be entered last. For example:

Disease
VS
Control

OPTIONAL: If batch effects are present in the dataset, assign batch names as shown. Batch effects will be corrected for differential gene expression analysis by incorporating batches into the linear model. For plots based on normalized gene expression and before ssGSEA analysis, normalized gene expression is corrected for batch effects with the ComBat function from the sva package (Leek, Johnson, Parker, Jaffe, & Storey, 2012). For an example showing how to identify batch effects in an experiment, find the section below titled Sample Quality Control.

Manually Add Groups and Batches:

Upload Groups and Batches:

Array Probe Quality Control:

Two plots, NUSE and RLE, display quality control metrics for the microarray probes. Plots are created with the oligo package:

Details regarding NUSE and RLE can be found in the oligo manual (Carvalho & Irizarry, 2010): https://www.bioconductor.org/packages/release/bioc/vignettes/oligo/inst/doc/oug.pdf

Normalization

Each sample is normalized to correct for technical artifacts using the Robust Multichip Averaging (RMA) method described in (Irizarry et al., 2003). Optional cyclic loess is also available and can be performed if RMA is insufficient (Ballman, Grill, Oberg, & Therneau, 2004; Bolstad, Irizarry, Astrand, & Speed, 2003). Only selected samples are used for normalization. Pre-normalization and post-normalization histograms, MA plots and box plots display relative expression (represented by probe intensity) before and after normalization. Data for the plots are generated with the oligo package (Carvalho & Irizarry, 2010). Normalized data can be downloaded in the DEG-Enrichments Results tab.

Histogram: The histogram displays the density of probes at each log-intensity. Each sample is represented by a curved line. Before normalization, samples should be distributed randomly. An outlier sample may be due to technical problems, and separate groupings of samples may indicate batch effects. Either of these potential issues should be investigated further before downstream analyses. After normalization, all samples lines should follow the same curve. Again, if samples are clustered into groups, batch effects may be present.

MA plot: Log-ratios (M) are plotted against the average log-intensities (A). Plots are generated that compare each sample to the median of all samples. After RMA, the red sample line should be relatively equal to the blue median line.

Boxplot: Displays the distribution of log-intensities for each sample. After RMA, distributions across all samples should be approximately the same.

Details regarding plots described above can be found in the oligo manual (Carvalho & Irizarry, 2010): https://www.bioconductor.org/packages/release/bioc/vignettes/oligo/inst/doc/oug.pdf

Sample Quality Control

Two plots, the sample similarity heatmap and 3D PCA, provide information about the quality of replicates in the groups.

3D PCA:

In general, samples in the same group should cluster together, and groups of samples should cluster separately from other groups. If batch effects are present, re-run the analysis and assign samples to respective batches for correction. For example, the dataset below has batch effects. The first principal component, PC1, separates replicates 1 and 2 (rep1 and rep2) from replicates 3 and 4 (rep3 and rep4):

Assign samples to batches and re-run MAAPster to correct for the batch effects, resulting in the PCA plot below. PC1 now separates group t3660 from group b62360:

Find the section Analysis/Manually Add Groups and Batches to see how to assign samples to batches, and for details regarding how batch correction is implemented.

Sample Similarity Heatmap:

Differential Gene Expression

The DEG-Enrichments Results tab displays differentially expressed genes between groups. MAAPster runs analysis in the background with the limma package. Linear modeling is performed using limma's lmFit function, and differential gene expression is determined using the contrasts.fit and eBayes functions (Ritchie et al, 2015). Using the toptable function, false discovery rates are calculated to adjust p-values for multiple testing (Ritchie et al., 2015). Detailed information regarding differential expression analysis can be found in the limma manual: https://www.bioconductor.org/packages/release/bioc/vignettes/limma/inst/doc/usersguide.pdf

Pathway Analysis

The top 500 significantly upregulated and top 500 significantly downregulated genes are extracted for pathway analysis (significance is determined as unadjusted p-value < 0.05). Pathway analysis is performed with CCBR's l2p R package (https://github.com/CCBR/l2p).

Gene Heatmaps:

Click on a pathway to generate a heatmap for the genes in the pathway. The heatmap will include groups selected in the contrast and will display scaled normalized gene expression for samples in the groups.

Single-sample GSEA (ssGSEA)

ssGSEA is performed using the gsva package as previously described (Hanzelmann et al., 2013). Pathway enrichment scores are then analyzed to determine fold changes and p-values between groups of samples, similar to the differential gene expression analysis described above. Differential pathway enrichment is ranked by p-value, and the top 50 pathways are displayed in a heatmap. Human gene set modules were downloaded from the BROAD Institute's MSigDB (http://software.broadinstitute.org/gsea/msigdb/collections.jsp), and mouse gene set modules were downloaded from the Gene Set Knowledgebase (https://www.bioconductor.org/packages/release/data/experiment/vignettes/gskb/inst/doc/gskb.pdf).

Download Results

All tables and plots generated may be downloaded.

References