tagged: biological replicates, biological samples, experimental design, statistics, technical replicates

Technical Replicates and Biological Samples

Posted by ludesi in 2DE Knowledge Base

A good experimental design can be one of the deciding factors for successful scientific discoveries and in the field of 2D gel based proteomics one of the topics that continues causing headaches is replicates.

So to follow up on our previous post on technical replicates this post provides some more basic background on the difference between technical replicates and biological samples.

The principles behind technical replicates and biological samples are logical and intuitive and easy to follow once the topic is presented in a correct manner.

For illustrative purposes, assume we want to investigate if gray mice are heavier than white mice.

Necessity of biological samples

If we randomly select one white and one gray mouse and measure their weights, we will not be able to draw any conclusions about weather gray mice are heavier in general. This is because we only have two biological samples.

If we repeat the measurements, let’s say we weigh each mouse five times then we will have ten different measurements. But this cannot be used to prove that white mice are heavier than gray mice in general, we still have only looked at one white and one gray mouse.

Using the terminology above the five measurements of each mouse are technical replicates.

What we need to do is to select five different white mice and five different gray mice. Then we would have more than two biological samples and be able to say if there is a statistical difference between white and gray mice in general.

Purpose of technical replicates

This does not mean that technical replicates are useless. If we don’t trust the weighing scale it is good to repeat the measurements to make sure we have measured the mice’s weight correctly.

Multiple biological samples and technical replicates

Below is an example where two white and two gray mice have been weighed five times each.

The measurements are grouped so that the first five corresponds to mouse 1 and so on.

Falsely assuming that the measurements come from twenty different mice will yield a pvalue of 0.003 – i.e. highly significant.

This is in correspondence with ones intuition, if the values actually came from twenty different mice, I at least would believe that the average gray mouse is heavier than the average white mouse.

But in reality we only have measurements from four different mice and the p-value will be 0.36, i.e. there is no statistical evidence for any differences.

This is also as one would suspect. Four observations: one white is below average, one white is on average, one gray on average, and one gray above would not convince me that there is a statistical difference.

Analyzing data correctly is not difficult, one just has to tell the algorithm which technical replicates corresponds to which biological sample. Unfortunately this is not possible in some of the 2D gel image analysis software.

2DE

Running multiple gels of the same biological sample is like weighing the same mouse multiple times. It is useful for eliminating differences that arise from running conditions, but we still need multiple biological samples to be able to draw any conclusion about the differences between the groups.

Depending on what the goal of your experiment is, you should therefore always plan at least 3 biological replicates. The more biological variation you have to compensate for, the more biological replicates you will need. As a general rule of thumb, the more biological replicates you have, the better your statistical confidence will be.

If you want to read more about calculating your ideal samples size, try and google for statistical power and required sample size. These are topics for a whole new post, so we’ll save that for another time. : )

If you have any questions or experiences you would like to share, we’d love to hear your comments.