Salivary Molecular Spectroscopy with Machine Learning Algorithms for a Diagnostic Triage for Amelogenesis Imperfecta

Area of Science:

• Biochemistry
• Genetics
• Medical Diagnostics

Background:

• Amelogenesis imperfecta (AI) is a genetic disorder affecting tooth enamel formation, often caused by mutations in specific genes.
• The complex phenotypes and genetic basis of AI make diagnosis challenging, necessitating advanced diagnostic tools.
• Current diagnostic methods for AI can be invasive and costly, highlighting the need for accessible screening platforms.

Purpose of the Study:

• To evaluate the efficacy of attenuated total reflection Fourier-transformed infrared (ATR-FTIR) spectroscopy combined with machine learning algorithms for discriminating AI patients from healthy controls.
• To identify specific salivary vibrational modes that can serve as potential biomarkers for AI detection.

Main Methods:

• A case-control pilot study was conducted using saliva samples from AI patients and matched controls.
• ATR-FTIR spectroscopy was employed to analyze salivary vibrational modes.
• Machine learning algorithms, including linear discriminant analysis (LDA), random forest, and support vector machine (SVM), were utilized for data analysis and classification.

Main Results:

• The support vector machine (SVM) algorithm demonstrated the highest performance in discriminating AI subjects, achieving 100% sensitivity, 79% specificity, and 88% accuracy.
• Shapley Additive Explanations (SHAP) identified five key vibrational modes (1010, 1013, 1002, 1004, and 1011 cm⁻¹) as significant features for AI detection.
• These findings suggest specific spectral regions as potential salivary biomarkers for AI screening.

Conclusions:

• ATR-FTIR spectroscopy coupled with machine learning offers a noninvasive and accurate approach for discriminating Amelogenesis imperfecta from control subjects.
• The identified salivary vibrational modes represent a promising, pre-validated spectral area for AI screening.
• This methodology holds potential for developing low-cost, accessible diagnostic platforms for AI.