Fig. 2: MS-GRS distribution and ROC-AUC analysis across three cohorts.

a–c MS-GRS distribution violin plots: comparative distribution of MS-GRS (multiple sclerosis genetic risk score) among different participant groups in three datasets: UK Biobank (a), Geisinger (b), and FinnGen (c). Groups are defined on the x-axis: healthy controls (Controls), individuals with optic neuritis without MS (ON only), MS-associated optic neuritis (MS-ON), and individuals with MS without optic neuritis (MS only). The mean is represented as a white circle, interquartile range as a black box, and the outside line shows the kernel density estimate of the underlying distribution. Each colour corresponds to a specific group: healthy controls (yellow), ON without MS (red), MS without ON (blue), and MS-ON (purple). The statistical analysis utilized two-sided Welch’s t-test with Bonferroni correction term to account for multiple comparisons. d–f ROC-AUC analysis: receiver operating characteristic area under the curve (ROC-AUC) analysis for differentiation between any form of MS (including MS only and MS-ON) versus healthy controls in three distinct datasets: UK Biobank (d), Geisinger (e), and FinnGen (f). The null model (grey line) encompassed the same covariates as the MS-GRS+covariates model (red line) but excluded the MS-GRS. MS-GRS without covariates is shown as a blue line. Covariates included in the models were: sex, TDI (Townsend Deprivation Index), age at cohort entry, and the first four principal components for UK Biobank; reported sex, index age, and the first four principal components for Geisinger; and sex, age at DNA sample collection, and the first four principal components for FinnGen. The ROC-AUC analysis provides insight into the discriminatory power of the models in distinguishing between MS cases and healthy controls.