Disentangling Blood-Based Markers of Multiple Sclerosis Through Machine Learning: An Evaluation Study.

Area of Science:

• Biomedical data analysis
• Computational neuroscience
• Machine learning in medicine

Background:

• Multiple sclerosis (MS) research increasingly utilizes blood-based biomarkers.
• Machine learning (ML) algorithms are applied for classifying MS using these markers.
• Significant variability exists in the ML methodologies employed in these studies.

Purpose of the Study:

• To compare different machine learning configurations for blood-based marker classification in multiple sclerosis.
• To evaluate the impact of feature selection methods and cross-validation strategies.
• To identify optimal ML approaches for MS biomarker discovery.

Main Methods:

• Comparison of various machine learning algorithms (e.g., Logistic Regression).
• Assessment of different feature selection techniques, including Random Forests.
• Evaluation using 10-fold cross-validation with heterogeneous data splits.

Main Results:

• Logistic Regression combined with Random Forests for feature selection and 10-fold cross-validation achieved the best classification performance.
• The specific blood-based markers identified were dependent on the feature selection method used.
• Heterogeneity in cross-validation data splits indicated variability in experimental setups.

Conclusions:

• The choice of machine learning algorithms, feature selection, and evaluation methods significantly impacts classification results for MS blood-based markers.
• Experimental design and data splitting strategies influence the selection of relevant biomarkers.
• Standardization of ML methodologies is crucial for reproducible MS biomarker research.