Will The Expression Level Value Of A Specific Gene Of Same Sample Be Same Even If They Are Obtained From Different Microarrays?

Question

[Ashis]

Hi,

I am very new in Bioinformatics. I am trying to create a model from a dataset consisting gene expression levels obtained from dna microarray. This model is supposed to be used to classify control group and case group. I am thinking of using MEAN of those expression level values to generate the model. Here, I should be sure about the standard of those kind of data.

Can the 'Expression Level' value of a specific gene of same sample be different if we measure it using different dna microarrays? If yes, how much do they differ? Are they close enough? If they are not close enough, how should they be normalized so that data from different microarrays do not deviate that much?

Thanks.

Ashis

Answer 1

[Philippe]

Hi,

you can use the mean to group of your replicates into one final "sample". However, it is generally more common to use the median since it is less affected by outliers.

The measured expression level of technical/biological replicate will most certainly differ but still be comparable. The differences can generally be due to a difference in hybridization, a difference in detecting the fluorophore signal intensity, the noise inherent to the technology or be actually biological. If you have biological or (even better) technical replicates you can compute the correlation of your expression levels between samples to assess the reproducibility of the methods. A good option is to perform rank-based analysis where you do not consider the expression value of one probe but rather its rank within you own sample. In principal, this rank are well conserved when similar samples/conditions are considered. One last point you should be aware of is the noise specific to microarrays. All genes have a signal detected even though there is no actual hybridization. Also, genes with lower expression have very noisy estimates (you can see this phenomenon plotting expression values of two identical samples against each other). then, be aware you can rely less on expression values of lowly expressed genes compared to highly expressed genes. Also, very highly expressed genes have generally a saturated signal, resulting in a lack of dynamic range for these values.

Concerning the normalization, this has been often debated here, like in this thread.

[Larry_Parnell]

Great response! In our experience, technical replicates show much less variation in signal intensity per gene than biological replicates.

[Philippe]

Yes, they usually do. But in absence of technical replicates (that happens), biological replicates can be of great help. Just have to be sure the sampling was done in similar environmental conditions...