PreMiEr Seminar Series: Dr. Anthony Fodor

PreMiEr-sponsored seminar hosted by University of North Carolina at Charlotte.

Speaker: Anthony Fodor, Professor, College of Computing & Informatics, PreMiEr Associate Director

Adventures with the Poisson distribution: on simplicity and reproducibility in microbial analyses

Abstract

In this talk, we explore ways in which simple algorithms can sometimes outperform more complex ones in metagenomics pipelines.  We present recent results where straight-forward classical statistical approaches to differential analysis can out-perform more recently developed negative-binomial based methods.  We show that data normalizations based on simple or “naïve” proportions can have some attractive properties when compared to more sophisticated, but more complex, compositionally aware transformations.  And we explore how one-parameter models based on the Poisson distribution can provide reasonable null error models for 16S-based ASVs and WGS approaches to taxonomic classification.  As recent controversies highlight, analysis of metagenomic data can often get bogged down in complexity.    Adopting simpler methods where possible has the potential to yield more easily interpretable results that are less likely to be the result of arbitrary choices made in data analysis pipelines.

Click here to register!