Detecting correlations, generating hypotheses, and avoiding pitfalls in the analysis of timeseries in developmental...

Currently available experimental approaches enable monitoring trait dynamics in singled individuals. One important advantage of individually resolved data is the ability to reveal correlations that can provide insights into regulatory processes underlying trait dynamics. Correlations within timeseries data likely result from continuous processes that govern trait dynamics, while lack of correlation may indicate changes in the underlying mechanisms. Examples of both are found in timeseries for traits ranging from growth to division timing during development to duration of life history stages. We offer practical recommendations, including the use of a previously proposed simple statistical test, for detecting correlations and, no less importantly, the absence of correlations in the kinds of timeseries that are often generated by developmental biologists. We pay particular attention to discriminating between real, biologically meaningful correlations and the artifactual ones that often arise when data are collected in multiple batches. Data sets that can be analyzed in this way likely already exist for various model systems and can be gathered while conducting other experiments. We advocate for analysis of individually resolved data as a powerful tool for generating empirically testable hypotheses in developmental biology.