An integrated computational pipeline for machine learning-driven diagnosis based on Raman spectra of saliva samples

Area of Science:

• Biomedical Spectroscopy
• Computational Biology
• Medical Diagnostics

Background:

• Raman spectroscopy offers disease-specific spectral fingerprints in biological samples.
• Interpreting complex spectral data remains a significant challenge for experts.
• Current diagnostic methods can be invasive, costly, or time-consuming.

Purpose of the Study:

• To develop and validate an end-to-end deep-learning pipeline for classifying Raman spectral data.
• To assess the pipeline's efficacy in diagnosing SARS-CoV-2 infection.
• To evaluate the pipeline for screening neurodegenerative diseases, including Parkinson's and Alzheimer's.

Main Methods:

• Utilized an end-to-end deep-learning architecture to process Raman spectral data from saliva samples.
• Trained and validated the pipeline on spectral datasets from confirmed SARS-CoV-2 positive and negative individuals.
• Tested the pipeline's performance on spectral data from patients with Parkinson's and Alzheimer's diseases.

Main Results:

• The deep-learning pipeline successfully classified spectral data for disease detection.
• Demonstrated high accuracy in differentiating between healthy controls and patients with SARS-CoV-2 infection.
• Showed promising results in screening for neurodegenerative conditions like Parkinson's and Alzheimer's disease.

Conclusions:

• An end-to-end deep-learning approach can effectively decode Raman spectral fingerprints for disease diagnostics.
• The developed pipeline offers a non-invasive, cost-effective, and time-efficient solution for disease screening.
• This technology facilitates rapid prototyping of advanced diagnostic tests for various medical conditions.