Salivary Multi-Omics Biomarkers Using Artificial Intelligence for Early Oral Cancer Detection and Risk Classification

Abstract

Background: Oral cancer continues to be a significant health issue worldwide, and a late diagnosis is highly correlated with poorer treatment outcomes and shorter life expectancy. While these two steps (examining a patient and obtaining tissue cultures) are essential to this diagnosis (previously routine), they do not help to detect the patient in time at the population level. A minimally invasive, biological fluid such as saliva has many molecular components that may mirror pathological changes occurring in the oral cavity. Salivary omics domains could thereby yield a better assessment, through a combination of the domains, of the risk of oral cancer at an early stage. Artificial intelligence (AI) may further enhance the analysis of complex molecular patterns by integrating information from multiple salivary biomarker domains. Such approaches may support the identification of potential early indicators of oral cancer and improve the classification of individuals according to their estimated disease risk.

Objective: This study aimed to evaluate the potential of multi-omics salivary biomarkers (genomic/transcriptomic, proteomic, metabolomic, and microbiome) for the early detection and risk classification of oral cancer. The study also highlights the potential application of AI-based analytical approaches to integrate multi-omics biomarker information and clinical characteristics for improved oral cancer risk classification.

Methods: The approach applied was quantitative cross-sectional and analytical, and involved a number of respondents of 279 people. Demographic and behavioural risk factors and the perception of being assessed for these risk factors by biomarkers were investigated, along with oral symptoms, clinical findings, and salivary multi-omics classifications. Normality test and Cronbach's Alpha reliability test, Bartlett's Test of Sphericity, and KMO Test, Independent Samples t test, One-Way Analysis of Variance, Kruskal-Wallis, Chi-Square Test of Independence, Pearson correlation and regression analysis were used in statistical analysis.

Results: The results showed a good level of reliability with an average Cronbach's Alpha of 0.901 - 0.938. Supporting construct validity is the value of kmo which is excellent (0.959), and the Bartlett's test result shows that the p-value is less than 0.001. Gender difference, differences in age groups, differences in category of areca use and cooked chew (areca/users, non-users), differences in Betel chewing (areca, non-users) were the significant differences. The overall multi-omics risk classification was well correlated with the clinical examination status, χ² (9) = 429.967, p < 0.001. Overall multi-omics risk revealed significant associations between domains of major biomarkers, with the highest between clinical examination and overall multi-omics risk. The regression analysis resulted in a highly significant predictive model of overall risk classification (R = 0.943, R² = 0.889, p < 0.001), which explained a great variance of risk predictors that were included in the model.

Conclusion: Results showed that the multi-omics biomarkers of saliva were a prevalent candidate for use as a minimally invasive tool to aid early detection/classification of OCD. Combination of use of molecular biomarker domains with clinical examination may lead to improved identification of patients likely to show symptoms of a clinical disorder that would warrant early diagnosis referral. The disease diagnosis performance needs to be further tested in other populations before the wide-scale application of the system. Future studies should investigate the integration of salivary multi-omics biomarkers using appropriately validated AI models, with particular attention to early cancer detection, risk classification, model interpretability, and external validation across independent populations.