Advertisement

Machine Learning Identifies Role of Impaired Purine Metabolism in Gout Pathogenesis

Published on: 

Machine learning uncovers gut microbiome biomarkers linked to purine metabolism, enhancing diagnostic strategies for gout and hyperuricemia.

A new machine learning analysis of the gut microbiome has identified the purine metabolism pathway as the main contributor in distinguishing gout from other groups, revealing key gut microbiome biomarkers and potentially aiding new diagnostic strategies for hyperuricemia (HUA) and gout.1

“In healthy individuals, approximately two-thirds of UA is excreted through the renal system, while the remaining one-third is eliminated via the intestinal tract. As renal function declines, the proportion of UA excreted through the gut may increase to two-thirds. However, renal dysfunction and excessive UA production limit the compensatory role of intestinal elimination, predisposing patients to HUA.2 Therefore, investigating gut microbiota differences among patients provides a valuable approach to understanding the pathogenesis of gout and HUA,” lead investigator Jia-Wei Tang, The Marshall Centre for Infectious Diseases Research and Training, The University of Western Australia, Perth, Australia, and colleagues wrote.1

Tang and colleagues collected 16S rRNA amplicon sequencing data of 233 fecal samples from 5 population studies conducted in Japan, South Korea, and China, for a total dataset of 100 healthy controls (HC), 93 in the HUA group, and 40 in the gout group. These groups had average ages of 40.87, 54.88, and 50.27 years, respectively, and mean BMIs of 24.34, 26.22, and 24.08. The groups consisted of 75 (75%), 62 (67%), and 38 (93%) male participants, respectively.1

Using the Shannon diversity index and inverse Simpson index, investigators found that the gout group had the lowest diversity while the HC group had the highest diversity, with significant differences observed at the genus level (P <.05).

One-way ANOSIM further confirmed significant separations among the three groups (HC vs. HUA: R = 0.167, P = .001; HC vs. GOUT: R = 0.651, P = .001; HUA vs. GOUT: R = 0.452, P = .001) (Fig. 2c). Further evaluation of the interactions among genera revealed that the HC group had the most complex microbial community network compared to the HUA and gout group.1

Machine learning (ML) and Shapley Additive exPlanations (SHAP) interpretability algorithms found that 5 genera, g-Christensenellacese, g-Streptococcus, g-Prevotella, g-Coprococcus, and g-Erysipelotrichaceae, were shared between HC and HUA groups, and 7 genera, g-Subdoligranulum, g-Agathobacter, g-Collinsella, g-Dorea, g-Alistipes, g-Lachnospira, and g-Bacteroides, were shared between HC and gout groups. Among these genera, Christensenellaceae exhibited a high contribution in the SHAP analysis and Subdoligranulum exhibited a strong correlation and significance with the HC group in both LEfSe and SHAP analyses. Alistipes was identified by SHAP as the most contributive genus.

Between HUA and gout groups, 8 genera, Subdoligranulum, g-Agathobacter, g-Blautia, g-Akkermansia, g-Lachnospira, g-Fusobacterium, g-Phascolarctobacterium, and g-Bacteroides, were shared. Among these, LEfSe found Subdoligranulum as more enriched in the HUA group and SHAP analysis identified Lachnospira as the most important contributor for model classification. Among unique, significant genera, Halomonas was found to be the most critical genus for model classification and Rhodococcus ranked second in contribution for model classification.1

A random-forest (RF)-based SHAP approach was chosen to assess the performance of the top 20 genera identified by LEfSe and SHAP based off best diagnostic performance, with prediction accuracy ranging from 82 to 96%. Using Tax4Fun2, Tang and colleagues also found that compared to HC, the HUA group showed reduced activity in thiamine, fructose, mannose, and propanoate metabolism, suggesting possible metabolic suppression contributing to hyperuricemia. Compared to the HC group, purine metabolism and fructose and mannose metabolism showed the most significant differences in the GOUT group. This was further reinforced when comparing HUA and gout, in which significant differences in purine metabolism were also observed.1

“In summary, the ML-based SHAP approach for identifying core taxa developed in this study outperforms the conventional LEfSe method in terms of classification accuracy. Functional prediction of the identified core taxa revealed significant enrichment in metabolic pathways associated with HUA and gout, further supporting the reliability of the proposed method. As the size and diversity of available datasets continue to grow, this approach holds promise for informing novel diagnostic and therapeutic strategies for HUA and gout,” Tang and colleagues concluded.2

REFERENCES
  1. Tang, JW, Tay, ACY, Wang, L. Interpretive prediction of hyperuricemia and gout patients via machine learning analysis of human gut microbiome. BMC Microbiol 25, 429 (2025). Doi: 10.1186/s12866-025-04125-x
  2. Ichida K, Matsuo H, Takada T, Nakayama A, Murakami K, Shimizu T, et al. Decreased extra-renal urate excretion is a common cause of hyperuricemia. Nat Commun. 2012;3(1):764.

Advertisement
Advertisement