45P - Gut microbiome signatures for exploring the correlation between gut microbiome and immune therapy response using machine learning approach

Background

Despite the remarkable efficacy of immune checkpoint inhibitor(ICI)-based immunotherapy in various cancers, it still faces prominent issues such as a limited population benefiting from the treatment. Numerous studies have shown that the gut microbiome can participate in immune modulation. Therefore, utilizing the gut microbiome to accurately predict the effectiveness of immunotherapy, thereby achieving precision diagnosis and treatment, has become a research hotspot.

Methods

Metagenomic sequencing was conducted on stool samples from patients receiving ICIs across three centers. Machine learning techniques were employed for model development, while SHapley Additive explanations (SHAP) were utilized to interpret the models.

Results

Distinct characteristics were observed in the gut microbiome between patients with different responses to ICIs. The areas under the receiver operating characteristic curve (0.81 in the validation cohort) and the precision-recall curve (0.80 in the validation cohort) demonstrated that the gut microbiome signature (GMS) exhibited excellent predictive performance. SHAP analysis identified the top five taxa driving GMS predictions: Anaerobutyricum hallii, Bacteroides eggerthii, Faecalibaculum rodentium, Clostridiales bacterium, and Anaerostipes hadrus. Survival analysis indicated that GMS also has prognostic value, showing significant associations with overall survival (P value < 0.001 for both log-rank test and multivariate Cox analysis) and progression-free survival (P value < 0.001 for log-rank test and P value = 0.008 for multivariate Cox analysis). In the validation cohort, GMS outperformed traditional single gut microbiome biomarkers (P value > 0.05) in terms of predictive performance (P value < 0.05).

Conclusions

The GMS demonstrated excellent performance in predicting responses to ICIs and prognosis, positioning it as a promising new generation of non-invasive biomarkers for evaluating ICIs efficacy. The taxa identified through SHAP analysis may serve as potential targets for enhancing the effectiveness of ICIs.

Legal entity responsible for the study

D-X. Wang.

Funding

Has not received any funding.

Disclosure

All authors have declared no conflicts of interest.