Two Color Array Getting Started

From Array Suite Wiki

Jump to: navigation, search



Experiments using 2-color design have 2 channels (labeled with 2 colors) for each array. For all the arrays, raw values are usually calculated using the same pre-defined method, for example Channel1 intensity/Channel2 intensity, no matter which channel actually contains the sample of interest. Besides Lowess normalization and ratio calculation, another special step in 2-color array preprocessing is to reverse the ratio values to make sure that every value means the fold change of sample/reference or treatment/control.

Two-Color Array with Sample in One Color and Reference in Another

The most common design for two color (competitively hybridized spotted) arrays is the 'reference design': each experimental sample (labeled with Color1) is hybridized against a common reference sample (labeled with Color2).

One example of this type of two-color array data is shown below.

2ColorArrayBestPractice image001.jpg

Experiment GSE21649 aims to investigate the different gene patterns of bone marrow samples between normal and SLE (Lupus Erythematosus, Systemic) individuals. Samples were obtained from 4 SLE patients and 4 normal controls. Each sample (labeled with cy3) is hybridized against a universal human reference RNA sample (labeled with cy5).

For this type of data, user may want to add expression data, run Normalize.pdf, Log2Ratio transformation and QC modules. The statistical analysis for this type of data would simply be a One-way ANOVA by comparing Lupus Erythematosus, Systemic with Normal.

Dye-Swap Experiment with Control and Treatment

Dye-specific bias effects, commonly observed in the two-color microarray platform, are normally corrected using the Dye-Swap design. Dye-swap design is to make duplicate hybridizations with the same samples using the opposite labeling scheme. For example, to compare two samples: Treatment & Control, make two (or an even number) arrays and hybridize them as follows:

Array 1: Treatment vs. Control ; Array 2: Control vs. Treatment

Example Dye-Swap data: Click to enlarge

Click to see enlarged picture

We say that labels for Array1 are swapped for Array2. Analysis of Dye-Swap data usually calculate the ratio of Treatment/Control for each array first and then merge the ratio values of 2 arrays.

One example of dye-swap experiment data is shown here.

Experiment GSE6820 involves the identification of the genes altered in response to cell scratching. Cell samples from 4 individuals were either treated by cell scratching (Wounds and Injuries) or remain untreated (normal). Gene expression levels of 23040 probes in these cell samples were measured at 1 hour, 3 hours, 6 hours, 15 hours and 24 hours after treatment (Not all samples from 4 individuals were measured at 24 hours). There are 18 combinations of individual and time. Each of the combination includes two Dye-Swap arrays, which generates 36 arrays in total.

Each combination of "Individual" and "Time" gives a unique identification for each Dye-Swap group. "ValueRatio" specifies how the raw ratio is calculated and "SampleChannel" specifies which Channel should be considered as the treatment (Wounds and Injures). If "SampleChannel" is denominator in ValueRatio, raw ratio will be reversed in Dye Swapping Process; otherwise raw ratio will remain the same. This makes every value to be ratio of treatment versus control.

Add Intensity Data for Both Color Channels

Modules exist for adding values from two channels, such as Import Agilent Text File module and Import GenePix Result File module, which can be found in Add Data | Add Omic Data | Add Expression Data.

For this experiment, intensity values of both channels should be imported first and the actual ratio will be calculated after normalizing the intensity values.

Note: When prompted, DO NOT attach design table when importing 2-channel data. A design table (72 rows) describing information of 2 channels for each array should have been attached automatically.

Caption Add data from 2 channels: Click to enlarge

Click to see enlarged picture

Transformation and Normalization

Before normalization, make sure the intensities are log2 transformed (See MicroArray_Transform for more detail).

Lowess normalization method will be applied to this dataset. Lowess is used for two-color experiments to help account for imbalances between red and green intensities.

Because ratio calculation is based on intensities, an Exp2 transformation (See MicroArray_Transform for more detail) needs to be performed to transform the normalized log2 values back to intensity values.

Generate Ratio

This step will, for each array, merge its 2 intensity values (from 2 channels) to one ratio value.

See Generate Ratios (Two Color Array) module for more details.

Refer to the 72-row design table for how to set "Array ID" and "Channel" parameters.

At this moment, the 36-row design table can be attached to this RatioData. To do this, right-click on the "Design" and select "Import".

Note: Remember to change the column property if categorical values are uploaded as numerical values.

Process Dye Swapping

This step will, for each sample, merge its 2 ratio values (from two paired arrays) to one ratio value.

See Process Dye Swapping module for more details.

This will automatically merge the design table (e.g. from 36 rows to 18 rows) based on Sample ID. The previous example will generate this design table:

2ColorArrayBestPractice image004.gif

"Sample ID" is the ID specifying which 2 arrays come from the same sample resource. In the example above, "Sample ID" should be Individual_Time.

Strategy to decide flipping - The example above uses two columns (SampleChannel and ValueRatio) to determine whether the ratio is calculated correctly. User can also use one single column to determine, for example a column telling "TRUE", "YES", "Flip" or "1".

Statistical Tests

All the preprocessing steps convert two-color array data to simple one-color array data. All analysis strategies applied to one-color array data can be applied to processed two-color array data. After running Quality Control, appropriate statistical analysis can be applied to different testing interest:

  • One Group Test
Take the previous experiment for example, to identify the genes altered in response to cell scratching, we need to compare Log2Ratio values of all biological groups (indicated by Time) to 0. The H0 hypothesis is Log2(Wounds/normal)=0. Use module One Group Test to run this type of analysis.
  • Linear Model
General Linear Model can also be used to examine gene expression changes at different time points (Expression = Time + Individual).

Dye-Flip Experiment with Control and Treatment

In Dye-Flip experiment, treatment and control for some arrays are labeled with cy5 and cy3, for others are labeled with cy3 and cy5. The difference between Dye-Flip and Dye-Swap is that in Dye-Flip one cannot find pairs of arrays which originate from the same sample resource, which means that there is no need to merge arrays. Comparing sample and reference in Dye-Flip experiment uses same method, except replacing treatment groups with samples and control groups with reference.

One example of dye-flip experiment data is shown below.

2ColorArrayBestPractice image005.gif

Experiment GSE4804 aims to identify differentially expressed genes in response 3 different treatments (Anti EGFR Ab and IL-13, Anti EGFR Ab, IL-13). Each one of the treatment groups is compared to control and 3 replicates exist for each treatment. In some arrays, treatment group is labeled in cy3 and control is labeled with cy5, while the other arrays have labels flipped (treatment with cy5 and control with cy3). None of arrays originates from the same sample recourse, so there is no need to merge arrays, and calculating inverse ratio of some arrays is necessary.


The steps to add 2-channel data, transformation, normalization and ratio generation for dye-flip experiment data

Flip Ratio

Ratio values should be flipped for arrays whose original ratio are calculated as (Control Channel/ Sample Channel).

Take GSE4804 for example, first click the design table under data object "RatioData".

Next, in the "View Controller", select the "Observation" tab and then expand "SampelChannel". Choose "Channel2 (4)". Now only these 4 arrays should be flipped.

2ColorArrayBestPractice image006.jpg

Right click on "List" under Solution Explorer and choose "Add List From Visible Columns":

2ColorArrayBestPractice image007.jpg

Choose "GSM" as list source and rename the list as "Flip".

2ColorArrayBestPractice image008.jpg

Transform the arrays on the "Flip" list by multiplying their values with -1.

Select MicroArray | Preprocess | Transform and choose "RatioData".

2ColorArrayBestPractice image009.jpg

When selecting Observations, choose "Customized observations" and click "Select" to choose the "Flip" list just generated.

2ColorArrayBestPractice image010.jpg

Note: REMEMBER to leave the Output field blank to overwrite the original ratio data.

Under Options, check "Multiply a constant" and change the number to "-1". Under Transformation method, choose "No Transformation".

Statistical Tests

Now the two-color data has been converted to simple one-color array data. All analysis strategies applied to one-color array data can be applied to this data. In the GSE4804 example, One-Way ANOVA can be run to identify differentially expressed genes in three biological group.