A new study led by co-corresponding authors, Min Zhang, MD, PhD, and Dabao Zhang, PhD, professors of epidemiology and biostatistics at the University of California, Irvine’s Program in Public Health, used artificial intelligence (AI) to analyze ‘big data’ sets of biological samples to reduce unwanted variations by more than 95%.
The results, which are published in the journal PNAS, hold promise for the field of biostatistics and metabolomics – the study of small molecules – as it has shown that using AI can dramatically reduce variations in -omics data collected from different cohorts, such as metabolomics data or small biological molecules.
Analyzing metabolite levels provides an immediate snapshot of an organism’s response to environmental changes, lifestyle factors, or disease states. Metabolite profiling of blood is a powerful tool to better detect, diagnose, and treat diseases. Blood is the one fluid most used in metabolomics as it provides a detailed picture of phenotypes (observable genetic traits) to be used in the field of precision medicine.
Using the power of artificial intelligence and applying biostatistics to this study, we were able to essentially wipe out the variations that had previously challenged researchers like me.”
– Min Zhang, MD, PhD
Yet, using metabolite profiles of blood samples for research has proved difficult due to its sensitivity to external factors. These factors can often be challenging to collect, leading to inconsistencies in research findings and complicating the use of metabolomics in precision medicine.
The research team, also led by co-corresponding author Daniel Raftery, PhD, professor at the University of Washington School of Medicine, collected more than 400 healthy human plasma samples from three geographically dispersed sites along with demographic information. This method mirrors common challenges in preserving blood metabolite levels such as collection methods, geographical sites, and demographic, clinical, and genetic factors.
The team then analyzed the samples using machine-generated modeling or AI and saw a dramatic reduction in the variation of metabolite levels (90%), especially their site-to-site variation (95%). This novel approach allows a more accurate prediction of certain chemicals in the body, and all while utilizing a smaller sample.
“Using the power of artificial intelligence and applying biostatistics to this study, we were able to essentially wipe out the variations that had previously challenged researchers like me,” said Min Zhang. “Our findings are an important step forward in the field of precision medicine and AI’s potential in the research world.”
Future studies will include an expansion of the number of samples processed and identifying abnormal biomarkers of disease.
Additional authors include doctoral students Danni Liu and Zhongli Jiang, and postdoctoral researcher Kangni Alemdjrodo, all from Purdue University and G.A. Nagana Gowda from the University of Washington.
This work was supported by U.S. National Institutes of Health grants R01GM131491, R01GM131491-02S2 and P30DK035816.
