What's the logic/motivation behind pre-filtering data based on variance?

Question

I am currently working with gene expression data where I have a number of genes (variables) measured over a number of samples. I could not make sense of the statistics and visualization suite we have, so I started checking out the tutorial and came up with the following bit:

You specify a lower cutoff for the variance. You can either enter a value, or use the slider. The value is relative to the variance of the variable with the largest variance. The variance is calculated over the active samples.

Example: If you select the value 0.1, then all variables that have a variance σ greater than or equal to 0.1 ·σ_max are kept, where σ_max is the variance of the variable with the largest variance.

I realize that it might be very fundamental question, but why would I want to filter my variables based on the variation alone? What I would want to know is which variables "explain" the difference between the samples (which are hypothesized to belong to a couple of different biological conditions)

Answer 1

I don't really understand what variance you are talking about, is it a variance of variables over one measure ? a variance of a variable over multiple measures ?

The more a variable has variance the more it contains information. Think about it in variation: if a variable has little variance it does not evolve much around its mean and then it won't be usefull to learn from the variable. On the contrary, variables with lots of variance will be usefull.