This is a test version of Biostars. For the public version, visit https://www.biostars.org.
Technical replicates in Microarray data

I am trying to analyze a microarray dataset from GEO in which one of the patients has 5 technical replicates. Should I calculate the average gene expression of 5 replicates? What is the best solution?

geo microarray
• 4,360 views • link
updated by cwwong13 • cwwong13 • written by Fatemeh Nabizadeh • Fatemeh Nabizadeh

Can you elaborate what a technical replicate in microarray experiments is? In RNA-seq it means resequencing the same library > 1, but what is it in the array world?

• 0 views • link
written by ATpoint ATpoint

I think that the definitions are almost the same. Here means taking one sample from the same patient and analyzing it 5 times across multiple arrays.

• 0 views • link
written by Fatemeh Nabizadeh • Fatemeh Nabizadeh

1 answer

The best way to handle technical replicates often depends on the context and the overall experimental design, which you haven't told us much about. Why did the experimenters make the technical replicates? What were they hoping to learn? Generally, however, if you have multiple arrays from one same patient sample while all other patient samples have only one array, then averaging the 5 arrays would be the logical strategy. Alternatively you might choose the best array out of the 5 in terms of quality criteria and discard the other four.

• 0 views • link
written by Gordon Smyth Gordon Smyth

Thank U Gordon! I checked the quality of 5 arrays for this patient.

• 0 views • link
written by Fatemeh Nabizadeh • Fatemeh Nabizadeh

Based on the plots, I think that averaging the 5 arrays would be an appropriate strategy. Am I right?

• 0 views • link
written by Fatemeh Nabizadeh • Fatemeh Nabizadeh

The plots actually tell you nothing about the relative quality of the arrays. The microarrays have been quantile normalized, which forces every array to have the same expression density and the same boxplot, so plotting those things becomes uninformative. An MDS plot of all 152 arrays would be more relevant. You want to see the technical replicates clumping together relative to other arrays.

• 0 views • link
written by Gordon Smyth Gordon Smyth

Thank you so much for your invaluable comment. I'll check it out!

• 0 views • link
written by Fatemeh Nabizadeh • Fatemeh Nabizadeh

In the case of averaging the 5 arrays, would you suggest averaging on the quantile normalized values?

• 0 views • link
written by cwwong13 • cwwong13

Log in to answer this question.

More posts like this