Skip to main content
 

Genetic model of MS severity (GeM-MSS) predicts future accumulation of disability

Wednesday, September 12, 2018 — Poster Session I

12:00 p.m. – 1:30 p.m.
FAES Terrace
NIAID
GEN-10

Authors

  • KC Jackson
  • K Sun
  • C Barbour
  • JL Milstein
  • JT Phillips
  • P Kosa
  • B Bielekova

Abstract

Genome-wide association studies (GWAS) have successfully identified and validated more than 200 susceptibility genes for Multiple Sclerosis (MS). However, similar studies have failed to validate genetic associations with MS severity, measured by the MS severity score (MSSS), likely due to the small effects of individual variants and insensitivity of the outcome. To mitigate these shortcomings, we employed following advances: 1. We modelled 113 unvalidated variants associated with MS severity in previous GWAS; 2. Rather than evaluating effects of individual variants we used an unbiased learning method Random Forests (RF) that capture cumulative, non-linear variant effects and interactions between variants. 3. As an outcome, we employed recently-developed, sensitive MS disease severity scale (MS-DSS) that integrates effect of treatments and central nervous system (CNS) tissue destruction measured by MRI and predicts future rates of disability progression. In the training cohort of 205 MS patients, variants were removed recursively from each RF model until the out-of-bag (OOB) error stabilized or increased by >1%. The final Genetic Model of MS Severity (GeM-MSS) included 20 unique variants at 13 loci. In an independent validation cohort (n=94), the GeM-MSS predicted outcome explained 4.4% of the variance in MS-DSS (R2=0.0441; p=0.043) and correlated with MSSS (Pearson r=0.202; p=0.055). GeM-MSS represents the first machine-learned model of MS severity and highlights the need to investigate the complex interactions between variants, ideally within large, multicenter collaborations.

Category: Genetics and Genomics