Minimum Number Of Samples (Biological Replicates) Required For Co -Expression Network Analysis

Question

[Sudeep]

Dear All,

I have a dual channel microarray dataset with just three biological replicates. I plan to analyze the gene co-expression network from this data. But my biggest concern here is just 3 biological replicates adequate to do this analysis ? Does anybody know any paper mentioning the effect of sample sizes on co-expression networks ?

Thank you in advance.

[Khader Shameer]

What do you mean by gene co-expression network here ?

[Sudeep]

generating a network of genes that are co-expressed in the samples where nodes will be the genes and weighing the edges with correlation values of expression between them

Answer 1

[Sean Davis]

More is better. I do not have a good rule-of-thumb, but I would suggest that three is WAY too small. If you think about co-expression in its simplest form as the calculation of correlation coefficients between genes across samples, you can probably see how using three samples (which will equate to looking for a linear relationship among three points) will not be very fruitful.

Answer 2

[Istvan Albert]

I would say that three is the minimum for most techniques, and it is way better than none or two replicates (there are plenty of papers that have no replication whatsoever).

That being said it is not just the number of replication that helps produce a result, the consistency across replicates also matters. Replication is just a way to increase statistical power, if your data is consistent you can get by with fewer replicates. You just have to be careful to produce results that the data supports.