Longitudinal differential abundance analysis of microbial marker-gene surveys using smoothing splines

Abstract

BackgroundHigh-throughput targeted sequencing of the 16S ribosomal RNA marker gene is often used to profile and characterize the taxonomic composition of microbial communities. This type of big high-through sequencing data is rapidly being applied to various infectious diseases like diarrhea. While many studies are limited to single "snapshots" of these communities, there is increasing recognition that longitudinal profiling of these communities are required to understand community dynamics and the complex relationships between dynamics and phenotypes of interest. Statistical methods that determine microbial features that are differentially expressed are required as an initial step to characterizing phenotypic associations with community dynamics in big data and infectious diseases.

Airway Microbiota Dynamics Uncover a Critical Window for Interplay of Pathogenic Bacteria and Allergy in Childhood Respiratory Disease

“…We summarized the results for OTUs with an absolute fold change of >1.5 and FDR adjusted p-value <0.025 in either age strata. We picked eight representative OTUs to more precisely investigate how the associations changed over time, and modelled the longitudinal structure of the data using smoothing splines ANOVA with 100 permutations to assess significance ( Paulson et al., 2017 ). All models for the ARI vs. healthy association were adjusted for age, season, gender, and any antibiotics within the last 4 weeks.…”

Faecal microbiota shift during weaning transition in piglets and evaluation of AO blood types as shaping factor for the bacterial community profile

“…In order to test taxonomic differential abundances, the family aggregated data were normalized by cumulative sum scaling approach and analysed in metagenomeSeq package [21], the effect of genotype was tested with the procedure for longitudinal data “fitTimeseries”[22] and the “fit-zig model” implemented in the same package was used for the other pairwise contrast.The level of significance was defined by p values (P) <0.05. For multiple comparisons the Benjamini& Hochberg correction was applied (Padjust).…”

Generalized Additive Mixed Modeling of Longitudinal Tumor Growth Reduces Bias and Improves Decision Making in Translational Oncology

“…While the slope ̂ assumes underlying log-linear growth, the eGaIT statistic ̂ assumes only a smooth response over the study period. Though the eGaIT estimates ̂ are scaled so as to align well with their group-matched LMM slope estimates ̂ in the log linear case, they are area under the curve (AUC) statistics (44): an alternative interpretation of the eGaIT statistic is as the constant log-linear growth rate that would have been needed to yield the AUC that we actually observed (see appendix). This interpretation gives some intuition as to when summarizing longitudinal curves by eGaIT is likely to enhance power to discriminate among treatment effects.…”