AI and Neurological Rare & Neuroimmune Diseases ── Multiple sclerosis ── 2026-11-09
Multiple sclerosis: treatment, trials and AI
- Abstract
Primary Progressive Multiple Sclerosis (PPMS) and HTLV-1-Associated Myelopathy/Tropical Spastic Paraparesis (HAM/TSP) are clinically overlapping progressive myelopathies characterized by spastic paraparesis and autonomic dysfunction. However, they originate from fundamentally distinct pathogenic processes: autoimmune demyelination in PPMS versus chronic retroviral-driven neuroinflammation in HAM/TSP. Current diagnostic approaches are frequently inadequate, as HTLV-1 screening is often omitted in non-endemic regions and positive serology in endemic areas does not rule out concomitant PPMS, highlighting the urgent need for complementary molecular diagnostic tools. In this study, we developed an integrative bioinformatics framework combining network analysis, differential expression profiling, and machine learning to identify a transcriptional signature capable of distinguishing PPMS from HAM/TSP. Disease-specific gene modules were identified through cross-dataset preservation analysis and module-trait correlations. Random Forest classification with recursive feature elimination was employed to derive a parsimonious Diagnostic Risk Score (DRS), which was validated on an independent external cohort under a frozen-threshold, leakage-free design. Weighted Gene Co-expression Network Analysis (WGCNA) revealed striking topological divergence between the two diseases: HAM/TSP exhibited a consolidated network architecture with a single dominant module, whereas PPMS displayed a highly fragmented topology with numerous smaller modules. Cross-dataset preservation analysis identified 11 PPMS-specific and 63 HAM/TSP-specific candidate biomarker genes. Machine learning refinement reduced this to a 4-gene signature comprising ZFAND1, MAPRE2, TBK1, and LPAR6, which achieved robust classification performance in cross-validation (macro-averaged accuracy: 0.982; macro-averaged F1-score: 0.987; AUC-ROC: 0.995) and robust performance on external validation (macro-averaged accuracy: 0.975; macro-averaged F1-score: 0.937; AUC-ROC: 0.800). The 4-gene signature captures distinct biological dimensions: ZFAND1 marks proteostatic stress from viral inflammation, TBK1 excludes the HTLV-1 interferon signature, MAPRE2 reflects a HAM/TSP-associated cytoskeletal program, and LPAR6 links inflammatory signaling to oligodendrocyte maturation. This multidimensional architecture explains the panel's high accuracy and offers mechanistic insights. The proposed approach offers a practical, transcriptome-based adjunct to current diagnostic methods, with potential to guide therapeutic stratification and reduce misdiagnosis of PPMS and HAM/TSP.
Journal IF-equivalent: 2.6 (OpenAlex 2-year mean citedness, value as of 2026-11-08, retrieved 2026-11-09; not the official Clarivate IF)Reference: Zarei Ghobadi M, Afsaneh E, Mozhgani SH. A four-gene diagnostic panel for distinguishing primary progressive multiple sclerosis from HTLV-1-associated myelopathy/tropical spastic paraparesis. Journal of Neuroimmunology. 2027 Feb;423:579120. doi:10.1016/j.jneuroim.2026.579120.Checked: Abstract only