Lyme disease (LD), caused by Borreliella burgdorferi, is the most common tick-borne illness in the United States, yet early-stage diagnosis remains challenging due to the limitations of current serological diagnostics. Raman spectroscopy (RS), paired with partial least squares discriminant analysis (PLS-DA), showed promise as an alternative diagnostic tool. Using RS, we analyzed 107 coded human blood samples (42 LD-positive and 65 LD-negative) obtained from the Lyme Disease Biobank. PLS-DA models showed nearly perfect internal validation performance with a sensitivity and specificity of 97.1% and 100.0%, respectively, indicating robust predictive capabilities. External validation of the developed chemometrics model with 80/20 training/validation split of all spectra gave true positive rates of 92.7% and 87.3% for serological positive and negative spectra, respectively. These findings highlight the potential of RS as a rapid and noninvasive diagnostic platform for LD, particularly when integrated with machine learning.

This dataset contains the raw (background-subtracted) and preprocessed, labelled Raman spectra for the referenced study. Specifically, the preprocessed spectra represent the 9-spectra averaged final model C data.