Advancing Statistical Methods for Next-Generation Microbiome Data Analytics

TITLE: Advancing Statistical Methods for Next-Generation Microbiome Data Analytics Abstract: The focus of our research lab is to support human microbiome research with rigorous, robust, and practical statistical and quantitative methodology. In the next five years, our goals are to address critical data analysis gaps raised by recent advancements in microbiome epidemiology and bioinformatics, and to develop novel statistical methods that will facilitate the next generation of microbiome analytics. First, expanding microbiome consortia and public databases increase precision in testing the microbiome’s association with host conditions, but introduce more sources for unmeasured confounding such as population heterogeneity or uncollected covariates (e.g., medication). Recent statistical research has focused on confounding from the data’s compositional nature but largely ignored these unmeasured and potentially stronger factors. We propose to benchmark unmeasured confounding effects in microbiome studies with diverse real-world data, and to develop specialized latent factor modeling techniques for adjustment. This will improve false discovery control and facilitate robust findings in large-scale microbiome association studies. Second, modern bioinformatics can generate rich whole-microbiome genetic profiles with millions of microbial genes. But these profiles tend to have extreme levels of sparsity and lack functional annotations, limiting interpretability and statistical power in downstream findings. We propose to leverage recent breakthroughs in artificial intelligence and genomic large language models (LLMs) towards this problem. We will utilize the inherent functional dependency structure encoded by genomic LLMs to create putatively functional orthologs of microbial genes, which can meaningfully aggregate sparse and under-annotated “dark-matter” genes. This will empower downstream analyses and unlock the potential of modern microbiome bioinformatics. Third, metatranscriptomic (MTX) protocols are increasingly available, which characterize the microbiome beyond its functional capacities (“what can microbes do”) and reveal functional bioactivities in situ (“what are microbes doing”). MTX expressions depend on the underlying gene abundances, which are dynamic and error prone. Existing methods do not account for this and can generate biased findings. We will develop a rigorous error-in-variable model that will properly adjust for dynamic and noisy gene abundances, thus enabling robust differential expression analysis in new MTX studies. Our developed methodologies will be validated in different populations, disease settings, and microbiome ecologies, and implemented as open-source software. Derived data resources will be provided as publicly available databases. Through this proposal, our overall vision is to address critically unmet analytical needs from recent advancements in microbiome epidemiology and bioinformatics, and provide a publicly available toolkit of novel statistical methods that will support the next phase of human microbiome research and catalyze translational discoveries. 1 Project Number: 1R35GM162668-01 | Fiscal Year: 2026 | NIH Institute/Center: National Institute of General Medical Sciences (NIGMS) | Principal Investigator: SIYUAN MA | Institution: VANDERBILT UNIVERSITY MEDICAL CENTER, NASHVILLE, TN | Award Amount: $433,985 | Activity Code: R35 | Study Section: Special Emphasis Panel[ZRG1 MGG-F (55)] View on NIH RePORTER: https://reporter.nih.gov/project-details/11272359