RMA normalization of microarrays (October 24, 2005).

This page is moving to a new website.

If you ask most statisticians if they want raw data or processed data, they will, for the most part, prefer to look at the raw data. There are two reasons for this. First, the statisticians want to understand the processing of the data and how that might influence the precision of any further calculations based on the raw data. Second, statisticians may want to try alternative approaches for processing the data and see if that produces better results.

An example of this involves the normalization of Affymetrix microarray chips. The Affy chips as they are often called, use photolithography to build the chip, a process similar to the approach used to create semiconductor chips. This approach allows a very high number of genes to be tested on a slide that can fit in the palm of your hand. Unlike spotted arrays, that places spots of the full sequence of the gene (or Expressed Sequence Tag) on a slide, the Affy chip selects 20 probes for each gene, each of which has a length of 25 base pairs. Next to each of these probes is another probe that represents the same 25 base pair sequence except that the middle base is changed. The 20 probes are called PM (Perfect Match) probes and the probes with the changed middle base are called MM (MisMatch) probes. The MM probes are an attempt to measure and control for cross-hbridization, the tendency for genese that are similar, but not identical to a particular gene sequence to bind weakly to these sites.

Affymetrix has developed an algorithm, MAS5, that adjusts the PM probes by effectively subtracting the signal found on the MM probes. This subtraction is done carefully to avoid producing negative signals and to minimize the effect of outliers. Most of the data I have received from Affy chips has used MAS5 to summarize the 20 PM and MM probes to produce a single intensity value for each gene.

You would think that we statisticians would be grateful for getting a data set that is effectively 40 times smaller, since the microarrays are already quite huge. But some statisticians have been experimenting with the data from the individual PM and MM probes to see if they could produce an alternate summary of these probes that behaves better.

The raw data from an Affy chip comes in a CEL format, and you also need information about the layout of the Affy chip, which comes in a CDF file. Two alternatives to MAS5, dChip and RMA, were developed to provide better summaries of the PM and MM probes. A comparison of MAS5, dChip, and RMA appears in

The RMA approach appears to be very promising because it does not involve an implicit subtraction of the MM probe values. The subtraction can lead to a lot of noise at low signal values. Instead, RMA looks at the distribution of the PM probe values and fits a combination of two distributions, a "noise" distribution that is normally distributed, and a "signal" distribution that is distributed like an exponential distribution. The normalized values are estimated through the expected value of the signal distribution.