Viome Life Sciences Bothell, Washington, United States
Objectives: Develop a technology platform that uses high resolution molecular data obtained from stool, blood, and saliva samples, and machine learned algorithms, to provide personalized nutritional recommendations, then test the ability of the platform to improve health outcomes for several chronic diseases.
Methods: Molecular data are generated with metatranscriptomic analysis (unbiased RNA sequencing) of stool, blood, and saliva samples, then bioinformatically interpreting the data into 1. quantitative strain and species level taxonomic classification of all microorganisms, 2. quantitative expression of microbial genes, clustered into KEGG orthologs (KOs), and 3. quantitative expression of human genes.
Machine learned algorithms were developed from numerous studies. One such study developed an algorithm that predicts the personalized postprandial glucose response (PPGR) using stool metatranscriptomic data. The study included 550 participants in USA and 550 in Japan, who consumed a total of 60,000 meals during the study. the PPGR for each meal was measured using a continuous glucose monitor (CGM).
Using the same molecular data and machine learned algorithms, we carried out a randomized, placebo-controlled trial (RCT) to test the ability of data-driven personalized nutrition to reduce HbA1c in adults with pre- and diabetes.
Results: We first developed three metatranscriptomic tests, for stool, saliva, and blood. Sample collection, transportation, and metatranscriptomic analysis were all clinically validated and licensed under CLIA.
A machine learned algorithm was developed based on stool metatranscriptome's ability to predict with 80% accuracy the personalized PPGR for each food.
The preliminary RCT results will be presented. The ANCOVA analysis method was employed to control for potential confounders measured at baseline. The personalized intervention (n=12) reduced HbA1c more than the placebo (n=15) in a clinically significant manner (HbA1c difference of 0.42%, p = 0.028).
Conclusions: We demonstrate that a precision nutrition program based on personalized metatranscriptomic data and AI/ML analyses can significantly reduce HbA1c relative to USDA-recommended nutrition in a prediabetic and diabetic population.
Funding Sources: Technology development and clinical trials were funded by VC firms, as part of normal financing of a startup.
