Biological and Technical Replicates

Historical Origins of the Replicate Nomenclature

Until the 1990s, much of the reproducibility testing in biomedical research, especially in the wet laboratory experimental sciences such as molecular and cellular biology, employed what would be considered technical replicates today. The push to systematically include biological replicates in experiments originated primarily in these fields in the 2000s with the widespread realization that biomedical research findings were not sufficiently reproducible, in large part because technical replicates by themselves did not properly account for the variability of the underlying biological phenomena. Major funding agencies in these fields, such as the U.S. National Institutes of Health, spurred the widespread adoption of the current replicate nomenclature and practices of replicate design by incentivizing researchers to employ both biological and technical replicates in their research as a way of enhancing the reproducibility of the research findings.

While the practice of quantitative measurements and statistical testing was much better established in many other fields of biomedical research, such as epidemiology, psychophysics, ecology and evolutionary biology, reproducibility of results was not necessarily commensurately better in these fields, arguably also because of poor replicate design.

Reproducibility Requires Representative Replicates

Research is primarily about learning general truths about the phenomenon under study. A set of findings is useful only to the extent that the same findings are obtained when the given experiment is independently but precisely repeated. The term reproducibility typically means this type of across-experiment reproducibility of the findings or conclusions and not of the measurements on which the conclusions are based. One accepts the mathematical reality that the measurements themselves will not be exactly reproducible from one instance to the next, be it within or across experiments, even as one expects the conclusions to be reproducible.

The only way one can draw reproducible conclusions based on inherently variable measurements is to use sound practices of statistical sampling and, where necessary, statistical testing. To the extent that the empirical measurements are truly representative of the underlying phenomenon, one can have a quantifiable degree of confidence, say 95%, that the conclusions will be reproduced when the experiment is exactly repeated. Thus, the key to obtaining reproducible results is to ensure that replicates as a group adequately represent the relevant statistical properties of the phenomenon of interest. This, in a nutshell, is the goal of replicate design: to ensure that the replicates are representative and that they adequately capture the study-relevant statistical properties, including the variability, of the phenomenon.

...