A commentary on “Oral microbiota as a biomarker for predicting the risk of malignancy in indeterminate pulmonary nodules: a prospective multicenter study”

Dear Editor,

We read with great interest the recent multicenter study by Ma et al^[1] on the use of oral microbiota profiling to predict malignancy in indeterminate pulmonary nodules (IPNs). In a prospective cohort of 1040 IPN patients across three centers, they successfully collected saliva, throat swab, and tongue-coating samples from 460 IPN patients and 70 healthy controls, applying 16S rRNA sequencing coupled with seven machine learning (ML) models to identify predictive microbial signatures. This integrated strategy, combining high-throughput sequencing, fluorescence in situ hybridization validation, and a visualized IPN risk prediction system, is a novel strength of the study. As recommended by the Transparency In The reporting of Artificial Intelligence (TITAN) guideline 2025^[2], we ensure the article is compliant with the guideline and respectfully propose the following points for further consideration and potential enhancement of future work.

First, the study offers robust statistical power. Prospectively enrolling and following patients addresses potential selection biases, thereby enhancing the internal validity and clinical relevance of the findings. By comparing saliva, tongue, and throat samples, the authors systematically identified saliva as the optimal site for IPN prediction. This comparison of niches represents a clear advance over prior studies, which typically focused on a single sample type. Then, the use of fluorescence in situ hybridization to visualize bacteria within tumor tissue provides confirmation that microbes identified in oral samples are present in lung lesions. In addition, testing the identified biomarkers against independent external cohorts (e.g., other lung cancer and non-lung disease datasets) strengthens the specificity and generalizability of the findings. Seven ML models (support vector machine, logistic regression, naive Bayes, multi-layer perceptron, random forest, gradient-boosting decision tree, and LightGBM) were trained and compared, ensuring optimal model selection. Notably, the authors used SHAP (Shapley Additive exPlanations) to interpret the saliva-LightGBM model, highlighting which core taxa drive the prediction. This transparency is a friendly methodological asset in a clinical context.

Second, the clinical relevance of this work is important. Differentiating malignant from benign IPNs is a major challenge in thoracic surgery. Current tools (e.g., radiographic models) have limited sensitivity and often lead to invasive biopsy or resection with attendant risks. A saliva-based test could serve as a safer, noninvasive “liquid biopsy” to augment decision-making. Saliva-LightGBM model achieved high discriminative performance (area under the curve [AUC] = 0.887, 95% confidence interval [CI]: 0.865–0.918) and identified biologically plausible oral genera (Actinomyces, Rothia, Streptococcus, Prevotella, Porphyromonas, and Veillonella) as malignant predictors. By stratifying IPN malignancy risk before any invasive procedure, such a tool could help avoid unnecessary surgeries or delays.

Third, these findings build upon and substantially support current understandings of microbiota-driven oncogenic mechanisms in lung cancer. Yang et al^[3] reported dysbiosis of salivary bacteria in non-smoking women with lung cancer, revealing Sphingomonas and Blastomonas positively correlated with immunocytochemistry markers (CK7 and Napsin A) and functional enrichment in p53 signaling pathway, pathways in cancer, apoptosis, and tuberculosis. Additionally, Zhou et al^[4] found that periodontal pathogens, including Fusobacterium nucleatum, Aggregatibacter actinomycetemcomitans, and Porphyromonas gingivalis, correlate with lung cancer risk in the atherosclerosis risk in communities study. Besides, a recent study^[5] also showed that specific principal component vectors of microbial communities and increased Streptococcus abundance were associated with lung cancer risk. However, Ma et al’s study is distinct in its design: no prior study has compared multiple oral sample types in a large IPN cohort or rigorously integrated explainable ML. The present study thus represents a significant advance in the field. The systematic approach and confirmatory analyses give confidence that the identified taxa are reproducible biomarkers, rather than chance associations.

Finally, we congratulate the authors on this comprehensive study and offer a few suggestions for further work. Applying the saliva-based model in independent, pre-operative IPN cohorts (e.g., in a clinical trial or pre-surgical assessment setting) would confirm its real-world utility. Combining microbiota profiles with radiomic features from CT scans or with pathomics might yield even higher accuracy. A unified model leveraging oral microbial, imaging, and histopathologic data could be highly powerful. Tracking microbiota changes over time in IPN patients may reveal how microbial shifts relate to nodule progression. Functional studies could elucidate mechanisms behind the associations.

In summary, Ma et al provide a compelling proof-of-concept that oral microbiota profiling can noninvasively stratify IPN malignancy risk. This work paves the way for integrating microbiome diagnostics into lung cancer screening algorithms. We look forward to further validation and clinical translation of these findings.

Ethical approval

Not applicable.

Consent

Not applicable.

Sources of funding

All the authors declare to have received no financial support or sponsorship for this study.

Author contributions

Z.W.: conceptualization, writing – original draft, writing – review & editing. S.Z.: writing – review & editing. C.G.: writing –review & editing.

Conflicts of interest disclosure

The authors declare that they have no competing interests.

Research registration unique identifying number (UIN)

Not applicable.

Guarantor

Chundong Gu.

Provenance and peer review

Commentary, internally reviewed.

Data availability statement

Not applicable.