ALS molecular subtypes are a combination of cellular and pathological features learned by deep multiomics classifiers
Kathryn O’Neill, Regina Shaw, Isobel Bolger, Oliver H. Tam, Hemali Phatnani, Molly Hammell
Abstract
Amyotrophic lateral sclerosis (ALS) is a complex syndrome with multiple genetic causes and wide variation in disease presentation. Despite this heterogeneity, large-scale genomics studies revealed that ALS postmortem samples can be grouped into a small number of subtypes, defined by transcriptomic signatures of mitochondrial dysfunction and oxidative stress (ALS-Ox), microglial activation and neuroinflammation (ALS-Glia), or TDP-43 pathology and associated transposable elements (ALS-TE). In this study, we present a deep ALS neural net classifier (DANCer) for ALS molecular subtypes. Applying DANCer to an expanded cohort from the NYGC ALS Consortium highlights two subtypes that strongly correlate with disease duration: ALS-TE in cortex and ALS-Glia in spinal cord. Finally, single-nucleus transcriptomes demonstrate that ALS subtypes are recapitulated in neurons and glia, with both ALS-wide and subtype-specific alterations in all cell types. In summary, ALS molecular subtypes represent a combination of cellular and pathological features that correlate with clinical features of ALS.