A Multimodal Multiomics Machine Learning (MMM) approach for biomarker discovery and acceleration of clinical trial readiness for childhood-onset neurological disorders
Journal:
medRxiv
Published Date:
Jul 22, 2026
Abstract
Background Childhood neurodegenerative disorders are usually rare, genetic, and life-limiting. Whilst targeted approaches present huge potential, significant hurdles include disease rarity, geographical dispersion of patients, funding, clinical trial design, and execution. Crucially, the paucity of robust biomarkers and objective measures of disease progression hampers evaluation of efficacy, drug development and regulatory approval. To address this paradigm, we developed a Multimodal Multiomics Machine Learning (MMM) framework, integrating large-scale, multi-source patient datasets to generate quantitative metrics for disease stratification and longitudinal tracking. We applied MMM to PLA2G6-associated neurodegeneration (PLAN), an ultra-rare condition currently lacking validated biomarkers, where precision gene therapy approaches are at an advanced preclinical stage. Methods A large, single time-point international natural history study (n = 310) was conducted alongside development of a disease-specific rating scale (CoPLAN-DRS), prospective longitudinal neuroimaging, and multiomic biomarker discovery. Machine learning methods were applied to the integrated dataset. Results Kaplan-Meier analyses enabled estimates for survival and time to loss of ambulation. Multiple clinical, radiological, and biofluid biomarkers were identified, clearly correlating with disease progression. The CoPLAN-DRS and brain MRI Quantitative Susceptibility Mapping showed strong positive correlation with age (rho = 0.69, 0.96 respectively). Nicastrin, a critical structural component of the gamma-secretase complex in Amyloid Precursor Protein (APP) processing, was identified as a novel biomarker. Neurofilament light levels showed strong negative correlation with disease progression (rho = -0.74). The complex multi-dimensional dataset was distilled into a simplified, clinically intuitive Digital Disease Dashboard (DDD), enabling real-time visualisation of disease severity. Conclusions Our study highlights the clinical utility of MMM in integrating multi-dimensional data from rare disease cohorts, delivering an unbiased, data-driven, optimised biomarker set. Condensing this into the DDD provides a pragmatically useful tool for clinicians, facilitating longitudinal tracking of disease. The MMM and DDD have accelerated clinical-trial readiness for PLAN, and potentially applicable to a broad range of neurogenetic disorders.
