Methods to Address Compositional Data Challenges in Clinical Studies for the Safety Assessment of Human Microbiome...

Advances in sequencing technologies have enabled increasingly detailed characterisation of the human microbiome in clinical studies, but interpretation of microbiome modulation which has relevance to health and disease characterisation or safety assessments remains methodologically challenging. Taxonomic profiles generated by amplicon or shotgun sequencing are inherently compositional, sparse, and limited by detection, which complicates differential abundance analysis and may lead to unstable or misleading conclusions, especially in low-biomass settings where contamination and under-detection are concerns. Here, we review strategies used to analyse and complement sequencing-derived taxonomic count data, with the aim of obtaining more quantitative information on microbial differential abundance and viability. These include transformations for relative-abundance-based analyses and bias corrections between samples based on mathematical assumptions or additional experimental measurements such as spike-ins, broad-range qPCR and flow cytometry. We find that there is no consensus on which method best addresses compositionality and that detection level and significance of low-level microbes in health and diseases are overlooked. We discuss the limitations of sequence-based methods, such as the biases induced by the experimental and analytical process, as well as viability measurements for meaningful differential abundance assessment. We illustrate the need to integrate prevalence as well as abundance and the importance of covariates in models in the case of bacterial vaginosis. Overall, meaningful assessment of microbiome perturbations requires not only statistical correctness of differential abundance analysis, but also careful study design, appropriate measurement choices, quantitative context, and explicit recognition of the biological and analytical limits of sequencing-derived data.