Biological replication - uic-ric/uic-ric.github.io GitHub Wiki

Biological replicate samples are essential for any statistically valid study. Any measurement inherently involves some variation, or error, in the measured values, which can come from both technical sources, such as the way the sample is prepared and biochemical reactions run during sample processing, and from biological sources, from inherent variation between cells and organisms. In most cases, the magnitude of biological variation is much bigger than that of technical variation.

In a typical experiment, you will be collecting data from samples under different experimental groups, whether different treatments, genotypes, time points, etc. The goal is to determine what changes between those groups. To know this, you need to know not only how much the measurements changed between groups (i.e., a fold-change, or the effect size), but also how much variation is typical for samples within each group (i.e., a standard deviation). This is conceptually the idea behind a t-test or an ANOVA: compare between-group variation to within-group variation; if the between-group variation is much bigger than the within-group variation, you can say with confidence that the measurement is different (small p-value). Because the samples you will be comparing in different conditions are necessarily independent of each other, you also need independent biological replicates within each group to obtain a fair estimate of the within-group variability.

Biological replicates involve the complete repeat of the experiment with an independent sample. If one sample is derived from a cell culture, then a separate cell culture is needed for a biological replicate. If one sample is derived from one animal, then a second animal is needed for a biological replicate. If one sample is derived from a pool of five animals, then a second pool of five different animals is needed for a biological replicate.

Technical replicates involve any repetition of part of the sample preparation steps, but starting from the same biological material. For example, redoing an RNA extraction, rerunning PCR amplification, re-preparing an RNA library, or re-sequencing a library. Technical replicates may be valuable to understand how much variation is introduced into an experiment just because of technical sample preparation steps, for example in testing the reproducibility of a novel methodology, but are not substitutes for biological replicates in a statistical analysis.

It is often a good idea to aim for more biological replicates than are needed, with the anticipation that some samples may fail and the analysis will still be valid even after excluding those samples. If only the minimal number of replicates is collected and one or two samples drop out, you may end up with a failed experiment.

In general, for functional sequencing experiments from model systems (e.g., RNA-seq, ChIP-seq), we recommend 3-6 biological replicates. The specific number depends on how reproducible your model system is, the likelihood of sample failure (and how consistently you can get good quality samples), and the magnitude of your biological effect.

For studies of clinically derived human samples it is much more difficult to make estimate, as people vary widely, and clinical collection of samples often introduces addition significant sources of variation and sample quality. Additionally, there may be a number of clinical factors to include as covariates in any analysis, such as age, race, sex, or various clinical or biochemical measurements; more covariates will require more samples for appropriate estimates. We typically recommend planning for no fewer than 10 samples per group, with the caveat that that number is very likely to be underpowered. Typical well-powered studies will involve at least 80-100 samples across 3-6 covariates, but may include hundreds of samples.

Please contact the Research Informatics Core (RIC) at [email protected] with any questions.