Averaging Summarization is a simple summarization method that when selected, adjusts the probe intensity data in two steps:
1) Data across all of the chips is normalized using the selected normalization method:
•Global Mean: Scales the probe intensity values so that the geometric mean per array is the same across all arrays. Each experiment in your project will be scaled so that they all have the same mean signal value. When Global Mean is specified, the geometric mean will be calculated, which utilizes the logarithm of values rather than the raw values.
Note: Log2 values may differ for paired experiments, making it appear as if they have not been correctly normalized. This is due to dissimilarities in how different read technologies deal with missing or zero values: With Affymetrix and NimbleGen, linear data are always positive numbers, so these data are always normalized as expected. With RNA-Seq data, missing linear values are not reported, while zeroes are reported for areas with no reads mapping. These can both cause issues with normalization. For ChIP-Seq, peaks and fragments always have positive linear values. However, genes may have missing values if no regions with mapping reads are found nearby, which can also affect normalization.
•Global Median: Scales the probe intensity values so that the median per array is the same across all arrays. Each experiment in your project will be scaled so that they all have the same median signal value.
•None: No adjustments will be made to the probe intensity values.
Note: When Global Mean or Global Median is chosen as the normalization method for Averaging Summarization, a scaling factor is calculated for each chip and displayed in the Experiment Attributes section of the Experiment List View under Preprocessing. This scaling factor can be a useful measure of quality when analyzing and comparing chips.
2) Probe data are summarized into gene level signal values by using averaging summarization, which calculates the geometric mean of the probe intensity values.