Review article: Statistical challenges in longitudinal microbiome data analysis
With Saritha Kodikara and Susan Ellul, we have published our latest review and benchmarked existing methods for longitudinal microbiome analysis on simulated data and a case study. The methods we have identified can be categorised into differential abundance analysis, clustering and network modelling.
The key points from our review are:
Longitudinal microbiome studies are conducted to understand the temporal variations of the microbiome, which is inherently complex, with dynamic interactions between microorganisms, host and environment factors.
Longitudinal microbiome data have inherent data characteristics that are common to both microbiome data (such as compositionality, sparsity and over-dispersion) and longitudinal data (such as within-subject correlation, high-variability between time points).
We identified three analysis objectives, ranging from differential abundance analysis, to clustering and network modelling.
Most methods developed to address one of the three objectives do not take the data characteristics into account, which may lead to biased or spurious results, and lack flexibility in their applications.
Methods for longitudinal microbiome data are still at their infancy, and require substantial methodological developed to understand biological and temporal relationships between microorganisms.