This is a test version of Biostars. For the public version, visit https://www.biostars.org.
Can beta values fall outside 0-1 range?

Hi, I am working with a beta value CSV file from a 850K Microarray experiment. Note, I don't have the raw idat files.

From what I understand beta values should be restricted between 0 and 1. However, a small fraction of the beta values are either >1 or <0.

Are the values in this CSV wrong? If so, is it because of incorrect preprocessing/ normalisation?

Any help will be greatly appreciated.

Best, D

microarray
• 1,105 views • link
written by biod • biod

1 answer

You are spot on that, by definition, true beta values must lie between 0 and 1.

Most likely is that the beta values have been converted to M-values or there have been some normalisation efforts of the beta values.

• 0 views • link
written by samuel.a.odonnell samuel.a.odonnell

Thank you for your response.

I was wondering if I should remove rows with these out of range beta. Is that a safe thing to do or should they be clamped to 0 (or a small value epsilon) and 1 (1-epsilon) for beta<0 and beta>1 respectively?

• 0 views • link
written by biod • biod

Sort of hard to say without knowing why these values exist...

• 0 views • link
written by samuel.a.odonnell samuel.a.odonnell

Unfortunately, I have not done the preprocessing of the IDAT files. Although, I have a note about the data processing which says

"

Data preprocessing

The Illumina DNA methylation array IDAT files were converted into methylation signals (beta values).

Sample quality control and normalization

The methylation signal intensities were normalized using the intra-sample Beta-Mixture Quantile (BMIQ) normalization procedure, and the inter-array Gaussian Mixture Quantile Normalization (GMQN) method. A total of N out of the N provided samples demonstrated an adequate beta distribution and were successfully normalized, which corresponds to a high proportion of good quality samples.

"

Wonder if there is any clue to what went wrong?

• 0 views • link
written by biod • biod

That makes sense; I am pretty sure those normalisation methods (particularly GMQN) can create values outside of 0-1.

I think in most cases trimming them to 0/1 would be tolerated but don't take my word for it.

• 0 views • link
written by samuel.a.odonnell samuel.a.odonnell

Thank you for your response.I was thinking trimming them back to 0-1 would still make them dicey. Probably, I will just remove them altogether. Any thoughts?

• 0 views • link
written by biod • biod

Log in to answer this question.

More posts like this