Machine learning-driven diagnosis of multiple sclerosis from whole blood transcriptomics.

Area of Science:

• Neuroimmunology
• Bioinformatics
• Genomics

Background:

• Multiple sclerosis (MS) is a neurological disorder involving immune system dysregulation.
• Clinically isolated syndrome (CIS) is the first manifestation, potentially progressing to definite MS.
• Early diagnosis and prediction of MS progression are crucial for patient management.

Purpose of the Study:

• To evaluate the diagnostic and predictive value of blood transcriptional alterations in MS and CIS using machine learning (ML).
• To develop non-invasive biomarkers for early MS detection and risk stratification.

Main Methods:

• Deep sequencing of over 200 blood RNA samples from CIS, MS, and healthy individuals.
• Application of ML binary classification to distinguish MS from healthy subjects.
• Utilizing ML Time-To-Event models to predict CIS conversion to MS.
• Algorithm benchmarking via nested cross-validation and independent dataset validation.

Main Results:

• A blood transcriptional signature accurately classified definite MS from healthy subjects with 97% accuracy.
• ML models predicted CIS conversion to MS with 72% accuracy using paraclinical data and 74.3% using blood transcriptomes.
• Blood-based classifiers demonstrated significant predictive power for MS development.

Conclusions:

• Blood transcriptomics combined with ML can identify predictive signatures for MS diagnosis and CIS conversion.
• These findings support the development of early, non-invasive diagnostic approaches for multiple sclerosis.
• Transcriptional profiling in blood offers a promising avenue for predicting MS risk in individuals with CIS.