NEW YORK — A model based on metagenomic and other features can determine who may develop nonalcoholic fatty liver disease (NAFLD) years before it develops.
NAFLD is becoming increasingly common, affecting about a quarter of the world's population, but the standard for diagnosing NAFLD is an invasive and costly liver biopsy. Because of that, a noninvasive screening tool is needed for NAFLD, said Gianni Panagiotou from the Leibniz Institute for Natural Product Research and Infection Biology–Hans Knöll Institute in Germany.
The gut microbiome has also increasingly been linked to NAFLD development, suggesting that microbial and other signatures could portend the development of disease. To investigate this, Panagiotou and his colleagues examined metagenomic and metabolomic markers within stool and serum samples from a prospective community-based cohort. At baseline, none of the participants had NAFLD, but a portion then developed the condition during the more than four years of the study. Using machine learning approaches, the researchers developed a model that could distinguish individuals who later developed NAFLD from those who did not.
"Prospective studies like ours that examine healthy people before they develop the disease allow us to find the microbiome characteristics that are associated with the risk of developing the disease," senior author Panagiotou, who is also an associate professor at the University of Hong Kong, said in an email.
The study appeared Wednesday in the journal Science Translational Medicine.
For study, the researchers screened more than 2,000 Chinese adults using ultrasonography for NAFLD to home in on a set of 1,216 individuals with no signs of the condition. At follow-up 4.6 years later, the researchers found — after applying strict exclusion criteria including recent antibiotic use or consumption of yogurt — 90 individuals who developed NAFLD in that intervening time. They then matched these now-NAFLD-positive controls to 90 individuals of similar age, gender, and body mass index who did not develop NAFLD.
Using stool and serum samples collected at baseline in 2014, the researchers searched for differences in the individuals' microbiomes and metabolomes that could potentially account for their development of NAFLD. For instance, certain bacterial genera like Methanobrevibacter and Phascolarctobacterium were decreased among the group that developed NAFLD, while Slackia and Dorea formicigenerans were increased.
Additionally, amino acids were enriched in the metabolome of individuals who developed NAFLD. Many of the metabolites that differed between the two groups have been reported previously to be tied to NAFLD, such as 3-chlorotyrosine, arachidonic acid, oxoglutaric acid, and tryptophan.
Using a machine learning approach, the researchers developed a prospective model based on these baseline metagenomic and metabolomic features to detect early signs of NAFLD. A version of this model incorporating 14 taxonomic, functional, and metabolomic features could distinguish about 72 percent of those who would later develop NAFLD.
Further refinement of the model led to one that included 18 features — two genera, three pathways, nine metabolites, and four anthropometric and clinical parameters — that could correctly classify about 80 percent of individuals. This model, the researchers noted, outperforms other clinical prognostic models.
The researchers further tested their model in four case-control cohorts from Asia, Europe, or the US. Cases in these studies had been determined through biopsy or magnetic resonance spectroscopy. In each of these, the new model could also distinguish cases from controls with high accuracy.
But Panagiotou said that their study and other studies of NAFLD have been hampered by small sample sizes, though the number of cohorts and biobanks have been increasing. He and his colleagues plan to work with an international consortium to put together a large, federated NAFLD cohort. "This would be the starting point to develop high-precision AI models for patients' screening, staging, prognosis, and care guidance," he added.