Inferring Differences in the Number of Classes Between Populations, in the Presence of Misclassification Errors

• Activity Number: 151 - Novel Methods and Tools in the Era of Big Omics Data
• Type: Contributed
• Date/Time: Monday, August 8, 2022 : 10:30 AM to 12:20 PM
• Sponsor: Section on Statistics in Genomics and Genetics
• Author(s): Senthil Kumar Muthiah* and Eric Slud and Christine Hehnly and Lijun Zhang and James Broach and Steven Schiff and Rafael Irizarry and Joseph Paulson
• Companies: Dana-Farber Cancer Institute and Unviersity of Maryland, College Park and Pennsylvania State University and Pennsylvania State University and Pennsylvania State University and Pennsylvania State University and Dana-Farber Cancer Institute and Genentech
• Keywords: richness; microbiome; spurious; misclassification; differential ; taxa

Abstract

Classical richness theory aims to estimate the unknown number of classes ( richness) in a population, given a survey sample. It assumes accurate class detection in the survey sample: select one member at a time, identify its class identity accurately, return it to the population, and repeat for a given number of trials. Inferences of differences in richness between populations (differential richness) are made by comparing sample-level estimates. Motivated by 16S microbiome surveys where sequencing reads are misclassified for their taxonomic groups, here we consider surveys with inaccurate class detection (misclassification). In this setting, we find that classical approaches confound differences in the population frequencies of members with differential richness. With extensive experimental and data analytic support, we argue that most 16S taxa identified at sub-“Genus” taxonomic resolution result from misclassification. Observed, systematic sampling effort dependent patterns of sub-genus taxa generation are used to derive a robust control for false taxa accumulation, which unlocks convenient regression approaches for differential richness inference for 16S microbiome surveys.