Biomarker selection and a prospective metabolite-based machine learning diagnostic for lyme disease

Area of Science:

• Biochemistry
• Bioinformatics
• Medical Diagnostics

Background:

• Accurate and early diagnosis of Lyme disease remains a significant clinical challenge.
• Liquid chromatography-mass spectrometry (LCMS) offers a powerful platform for identifying disease-specific biomarkers in biological samples.
• Existing diagnostic methods for Lyme disease can be limited in sensitivity and specificity.

Purpose of the Study:

• To develop and validate a computational pipeline for the prospective diagnostic testing of Lyme disease using LCMS serum data.
• To leverage machine learning techniques for robust biomarker discovery and classification model development.
• To establish a generalizable methodology applicable to other metabolomic LCMS datasets.

Main Methods:

• A comprehensive pipeline was designed for data preprocessing, feature selection, and classification of LCMS serum samples.
• Machine learning algorithms, including sparse support vector machines (SSVM) and iterative feature removal (IFR), were employed for biomarker identification.
• K-fold cross-validation and feature ranking were utilized to build a discriminant model for Lyme disease detection.

Main Results:

• The developed model achieved a high balanced success rate (BSR) of 98.13% on a sequestered test set of LCMS serum samples.
• Several key biomarkers indicative of Lyme disease were identified through the ML-driven selection process.
• The pipeline demonstrated robust performance in classifying samples, distinguishing between Lyme disease and control groups.

Conclusions:

• The developed LCMS-based pipeline with ML integration provides a highly accurate method for Lyme disease diagnosis.
• The identified biomarkers and classification model hold potential for a prospective diagnostic test.
• The generalizable nature of the methodology allows for its adaptation to other metabolomic studies and diagnostic challenges.