Question

within array vs between array normalization (methylation)

1

Entering edit mode

5.3 years ago

rorio ▴ 20

Hello all,

This may seem to be a silly question, but what is the difference between "within array" and "between array" normalization, specifically within the context of Illumina methylation microarrays? I understand the concept of normalization, but not really this distinction. It seems to be used inconsistently. What, exactly, is an "array"? Does it refer to a single sample on a chip? If so, I understand that "within array" normalization could be used to correct for the Inf I/Inf II shift, and between array normalization would then be to ensure each sample has the same distribution of intensities.

However, is an array referring to a chip with 8 / 16 samples on it? If so, why is there a distinction between the type of normalization within an array (ie, normalizing the 8 /16 samples to each other) and normalizing this array to other arrays?

Thank you for your help!

normalization methylation microarrays illumina • 3.5k views

ADD COMMENT • link updated 5.2 years ago by Kevin Blighe 87k • written 5.3 years ago by rorio ▴ 20

score 1 · Answer 1 · 2019-01-27

My answer will be brief:

In most contexts, 'chip', 'array', and 'microarray' refer to the same thing, and only a single sample will be used per chip / array / microarray.

Within-array normalisation will take each array (thus, each sample) as its own entity and perform some normalisation procedure separately on each - during this normalisation procedure, information from other arrays / samples in your dataset is not taken into account. The within-array normalisation method can differ across different platforms. For example, for 2-colour microarrays, your sample's DNA/cDNA is hybridised with a control DNA/cDNA, i.e. on the same chip, in which case, within-array normalisation occurs per sample and using information from the probes that have hybridised to the control sample. To a certain degree, within-array normalisation occurs on every chip due to the presence of replicate probes (multiple probes with same target) and / or positive and negative controls.

Between-array normalisation, then, looks at information from all of your samples and provides a metric to normalise all together. This is 'necessary' for the purposes of differential expression analysis. One of the most widely-used between-array normalisation methods is quantile normalisation, which is used by the widely known RMA (robust multiarray average) method. In a nutshell, quantile normalisation aims to match the data distribution across all of your samples. This is why, after quantile normalisation, the distributions of your samples usually look quite similar in a box-and-whisker plot.

By just performing within-array normalisation, there is a risk that your samples are not cross-comparable afterward, i.e., for the purposes of differential expression analysis. Biases / technical artifacts may exist as a result of, for example, how different samples were prepared and then loaded on the chips. We can mirror this to RPKM and FPKM expression units in RNA-seq: RPKM/FPKM are produced by a normalisation strategy that is within-sample only. This is why usage of RPKM/FPKM is not recommended in some circles for differential expression analysis.

Trust this helps.

Kevin