Integrating machine learning and multi-omics analysis to identify and validate key genes associated with periodontitis.

AIMS: Early detection and intervention are crucial for effective management of periodontitis. Our study aims to identify diagnostic biomarkers for periodontitis by integrating single-cell RNA sequencing analysis, Mendelian randomization, and experimental validation.

METHODS

Gene Expression Omnibus (GEO) dataset GSE164241 was downloaded to analyze the cellular compositions and intercellular communications in periodontitis. GEO datasets GSE10334, GSE16134, GSE23586, and GSE106090 were downloaded, and differential expression genes (DEGs) and functional enrichment were analyzed, followed by investigation of endothelial-fibroblast signaling pathways. Molecular subtypes were defined based on myeloid and fibroblast markers, and their immune characteristics were analyzed. A diagnostic model was built using 106 algorithm combinations from 11 machine learning methods (Lasso, Ridge, Enet, Stepglm, SVM, glmBoost, LDA, RandomForest, GBM, XGBoost, and NaiveBayes) to screen reliable biomarkers. Key genes were identified using Mendelian randomization and validated via reverse transcription and quantitative polymerase chain reaction (RT-qPCR).

RESULTS

This study identified four novel periodontitis subtypes using consensus clustering based on myeloid cell and fibroblast markers: fibroblast-dominant, myeloid-dominant, fibroblast/myeloid quiescent, and fibroblast/myeloid mixed. These four subtypes exhibited unique biological patterns, potentially leading to different disease progression. Patients in the fibroblast/myeloid mixed and myeloid-dominant groups, in particular, may have a stronger inflammatory microenvironment. In addition, we found that IGFBP4, IL1B, LAPTM5, PSAP, and SRGN were closely related to the occurrence and development of periodontitis through random forest (RF) analysis and Mendelian randomization. RT-qPCR validation in gingival tissues from 13 stage III/IV periodontitis patients and healthy controls confirmed significant differences in the expression of IGFBP4, IL1B, and LAPTM5 (p < 0.001).

CONCLUSION

Our findings suggest that IGFBP4, IL1B, and LAPTM5 could be potential biomarkers for periodontitis, paving the way for future research on its pathogenesis, diagnosis, and treatment.