RMeDPower for Biology: guiding design, experimental structure and analyses of repeated measures data for biological studies

Abstract

AbstractReproducibility is science has plagued efforts to understand biology at both basic and biomedical and preclinical research levels. Poor experimental design and execution can result in datasets that are improperly powered to produce rigorous and reproducible results. In order to help biologists better model their data, here we present a statistical package called RMeDPower in R, which is a complete package of statistical tools that allow a scientist to understand the effect size and variance contribution of a set of variables one has within a dataset to a given response. RMeDPower can estimate the effect size of variables within an experiment based on an initial pilot dataset. In this way, RMeDPower can inform the user how to predict the scope, dimension and size of biological data needed for a particular experimental design. RMeDPower employs a generalized linear mixed model (LMM) -based power analysis, specifically targeting cell culture-based biological experimental designs. This package simulates experiments based on user-provided experimental design related variables, such as experiments, plates, and cell lines as random effects variables. This package not only allows us to use pilot data to estimate variance components for power simulation, it also accepts a set of variance components, which is an estimation of variance of the random effects linked to experimental variables and transformed into Intra-class Correlation Coefficients (ICC), as input which is precalculated from different data sets. The latter case is suitable when pilot data has an insufficient number of replications of experimental variables to directly estimate associated variance components. RMeDPower is a powerful package that any scientist or cell biologist can use to determine if a dataset is adequately powered for each experiment and then model accordingly.