Evaluating and developing optimal statistical practices for human microbiome data analysis

The National Institute of Allergy and Infectious Diseases awarded Mayo Clinic $100,228 on November 23, 2026, to develop and evaluate novel biostatistical methods for human microbiome metagenomic data analysis under the Allergy and Infectious Diseases Research program (CFDA 93.855).

The award funds two primary research aims. In Aim 1, the recipient will systematically evaluate widely used methods for analyzing longitudinal microbiome data—specifically mixed-effects models—and develop a simpler fixed-effects approach that is more robust in small, sparse datasets. Work will include pseudo-simulating repeated-measures data while varying sample size and effect size to measure false positive control and statistical power, then applying these methods to real-world datasets from endometrial and vaginal microbiome studies as well as publicly available longitudinal microbiome data. In Aim 2, the recipient will optimize treatment of uneven sequencing effort and create a novel extrapolation-based technique to retain samples with smaller library sizes, using simulation studies to determine optimal library-size cutoffs that balance power and false positive control, and will train a deep learning model on large, diverse datasets.

The project runs through June 30, 2029, with place of performance in Rochester, Minnesota. The funding is structured as a project grant, a standard assistance type for investigator-initiated research in the NIH portfolio.

Abstract

This fellowship application proposes to develop and evaluate novel biostatistical methods tailored to the unique challenges of human microbiome metagenomic data, with a focus on improving robustness and reproducibility. Microbiome studies are increasingly performed with longitudinal designs and in less diverse body sites, producing small, sparse, and highly variable datasets. Such complexities hinder the identification of reliable microbial biomarkers and potential therapeutic targets. By creating (1) a simpler fixed-effects model for longitudinal differential abundance analysis and (2) optimized or alternative approaches to uneven sampling effort (rarefaction), this project aims to address key limitations of existing methods in terms of false positive control and statistical power. In aim 1, i will systematically evaluate widely used methods for analyzing longitudinal microbiome data-specifically mixed-effects models-and develop a simpler fixed-effects approach that is more robust in small, sparse datasets. First, i will pseudo-simulate repeated-measures data while varying sample size and effect size, measuring false positive control and statistical power. I will then apply these methods to real-world datasets from our lab's endometrial and vaginal microbiome studies, as well as publicly available longitudinal microbiome data, comparing how well mixed- and fixed-effects models detect true signals without inflating false positives. In aim 2, i will optimize the treatment of uneven sequencing effort (commonly addressed via rarefaction) and create a novel extrapolation-based technique to retain samples with smaller library sizes. Using simulation studies across varied sample types, i will determine the optimal library- size cutoffs that balance power and false positive control. I will then train a deep learning model on large, diverse datasets to extrapolate diversity metrics for samples below the cutoff-minimizing data loss in studies with limited library size. These advancements hold significant relevance to public health. Host-microbiome interactions contribute to a range of disease phenotypes, including reproductive cancers-an active area of study in the walther-antonio lab at mayo clinic. More robust analytical tools will enable researchers to uncover meaningful associations with greater confidence, ultimately accelerating the development of microbiome-based diagnostics and interventions. The research training plan combines formal biostatistics coursework (including a newly launched concentration in biostatistics) with active mentorship from DRS. Walther-antonio and chen. Complementary activities include regular conference presentations, thesis committee presentations, and lab meetings, facilitating technical skill development and professional growth. The mayo clinic environment provides access to clinically relevant microbiome datasets and a supportive, interdisciplinary community. This integrated training approach will prepare me to become an independent investigator leading quantitative, translationally focused research on microbiome-driven human health.