Figure 2: Biomarker for predicting MS severity using in silico approach.

(A) Classification modelling was based on exploratory analysis on the variables in the dataset with the shortlisted six predictors, i.e., QA, PA, KA, FGF-basic, TRP and TNF-α and its ranked importance for the predictive analytics. (B) Training Set using a C5.0 Decision Tree comprised of pie chart proportions of healthy control or MS subtype after being split by the six predictors. To optimize the split, calculated cut-off concentrations for each predictor were determined by the analytic software. The aim is to define a set of predictors that results in a full circle for each experimental group. For example, a QA concentration ≥494 nM (#) results in isolation of the SP and PP MS subtypes, then applying a PA concentration of <313 nM (#), as the next predictor, results in 89.1% isolation of the PP MS subtype. The experimental groups are denoted: healthy control (HC; green), RRMS (RR; yellow), SPMS (SP; orange) and PPMS (PP; red). (C) The numbers of observed and correctly predicted HC and MS subtype in the Training Set are shown (blue boxes) along with proportions of true (sensitivity) and false (specificity) predictions. (D) A different Test Set was used to validate the predictive model built from the Training Set. The numbers of observed and correctly predicted HC and MS subtype in the Test Set are shown (blue boxes) along with proportions of true (sensitivity) and false (specificity) predictions.