Abstract
Longitudinal, prospective studies often rely on multi-omics approaches, wherein various specimens are analyzed for genomic, metabolomic, and/or transcriptomic profiles. In practice, longitudinal studies in humans and other animals routinely suffer from subject dropout, irregular sampling, and biological variation that may not be normally distributed. As a result, testing hypotheses about observations over time can be statistically challenging without performing transformations and dramatic simplifications to the dataset, causing a loss of longitudinal power in the process. Here, we introduce splinectomeR, an R package that uses smoothing splines to summarize data for straightforward hypothesis testing in longitudinal studies. The package is open-source, and can be used interactively within R or run from the command line as a standalone tool. We present a novel in-depth analysis of a published large-scale microbiome study as an example of its utility in straightforward testing of key hypotheses. We expect that splinectomeR will be a useful tool for hypothesis testing in longitudinal microbiome studies.
Highlights
Biological studies in humans are subject to significant variability and noise, often great enough to obscure all but the most dramatic differences
We provide the user with interpretable results in the form of pre-formatted plots that can be saved at publication quality and re-generated from the results object stored by the function
We have shown how a new open-source R package, splinectomeR, can quickly assess statistical significance in large longitudinal microbiome studies by summarizing longitudinal group data with splines and using randomly permuted distributions to evaluate the probability that the observed magnitude of differences between groups, or of trends over time, is due to chance
Summary
Biological studies in humans are subject to significant variability and noise, often great enough to obscure all but the most dramatic differences. A number of practical concerns often complicate analysis of longitudinal microbiome data: time points are usually not in sync or differ in number between subjects, longitudinal variation may not follow a normal distribution, and timeseries data may follow arbitrary curves, for example during the maturation of the infant microbiome. To overcome these challenges, researchers in many studies have collapsed samples across time points to average individuals’ signals or they have
Sign-in/Register to view highlights & summary Sign-in/Register to access full text options 
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Schedule a call
