Repeatability and reproducibility assessment in a large-scale population-based microbiota study: case study on human milk microbiota

Shirin Moossavi,Kelsey Fehr,Meghan B Azad,Ehsan Khafipour

doi:10.1186/s40168-020-00998-4

Abstract

BackgroundQuality control including assessment of batch variabilities and confirmation of repeatability and reproducibility are integral component of high throughput omics studies including microbiome research. Batch effects can mask true biological results and/or result in irreproducible conclusions and interpretations. Low biomass samples in microbiome research are prone to reagent contamination; yet, quality control procedures for low biomass samples in large-scale microbiome studies are not well established.ResultsIn this study, we have proposed a framework for an in-depth step-by-step approach to address this gap. The framework consists of three independent stages: (1) verification of sequencing accuracy by assessing technical repeatability and reproducibility of the results using mock communities and biological controls; (2) contaminant removal and batch variability correction by applying a two-tier strategy using statistical algorithms (e.g. decontam) followed by comparison of the data structure between batches; and (3) corroborating the repeatability and reproducibility of microbiome composition and downstream statistical analysis. Using this approach on the milk microbiota data from the CHILD Cohort generated in two batches (extracted and sequenced in 2016 and 2019), we were able to identify potential reagent contaminants that were missed with standard algorithms and substantially reduce contaminant-induced batch variability. Additionally, we confirmed the repeatability and reproducibility of our results in each batch before merging them for downstream analysis.ConclusionThis study provides important insight to advance quality control efforts in low biomass microbiome research. Within-study quality control that takes advantage of the data structure (i.e. differential prevalence of contaminants between batches) would enhance the overall reliability and reproducibility of research in this field.Dksa_EC4VEZpmNY5pbL1BCVideo abstract

Highlights

Quality control including assessment of batch variabilities and confirmation of repeatability and reproducibility are integral component of high throughput omics studies including microbiome research
We studied a subset of 1194 mother-infant dyads in the Canadian Healthy Infant Longitudinal Development (CHILD) Cohort Study [9]
Some participant characteristics varied significantly between the batches and some degree of true biological variability between batches was anticipated in the milk microbiota composition

Summary

Introduction

Quality control including assessment of batch variabilities and confirmation of repeatability and reproducibility are integral component of high throughput omics studies including microbiome research. Low biomass samples in microbiome research are prone to reagent contamination; yet, quality control procedures for low biomass samples in large-scale microbiome studies are not well established. Potential sources of batch effects in microbiome research include heterogeneity in all aspects from sample collection to library preparation and bioinformatics processing [1] leading to technical variability. Reagent contaminants pose a major challenge in microbiome profiling of low biomass samples such as milk [5, 6] and could be an important source of non-technical batch variability even when all procedures are identical

Objectives

Methods

Results

Discussion

Conclusion