Normalization of Affymetrix data

Question

[XBria]

Hi everyone !

I have normalized data,however the result seems not normal; after normalization of 11 probes (14 samples) , all points are regularly presented between 200-400.

Thanks !

Answer 1

[cpad0112]

Could you please address following doubts I have:

1) What normalization method did you use?

2) Are you expecting any thing different than existing normalized values?

[XBria]

Of course !

I normalized my data by Quantile normalization method in Matlab.

I suppose both figures of pre and post normalization are different. The normalized one should not be too regularly presented though. At least some differences should be seen.

I 'm curious to know if it is normal ,what may be the reason behind this presenting.

thanks !

[cpad0112]

In general, once normalized, the extent of data normalization is property of raw data it self and the method used for normalization. There are several ways to check the effect of normalization, visually. Plot MA plots assuming that you are working on expression chips. 1) Plot MA values and check IQR and R2 for Normalized Vs Raw data 2) Plot data using Box plots. These plots should be able to tell you visually, if your data is normalized or not, compared to raw data. However it is difficult to give the data range post normalization as it depends on the raw data range.

[XBria]

Thank you so much ! :)

I'm gonna check it.

Answer 2

[XBria]

Of course !

I normalized my data by Quantile normalization method in Matlab.

I suppose both figures of pre and post normalization are different. The normalized one should not be too regularly presented though. At least some differences should be seen.

I 'm curious to know if it is normal ,what may be the reason behind this presenting.

thanks !

Answer 3

[Lemire]

Maria, I think your problem is a semantic one. Normalization does not mean that the end result will be normal (gaussian). Normalization here refers to the process of reducing non-biological (e.g. technical) variation. If your raw data doesn't look bell-shaped, your quantile-normalized data won't look bell-shaped either.

[cpad0112]

simply put, quantile normalization makes the distribution same across the chips and per se it won't change the distribution of input data.