Identifying candidate therapeutic targets in amyotrophic lateral sclerosis through a transcriptome-wide machine-learning consensus approach for drug repurposing.
TL;DR
The results provide transparent, hypothesis-generating gene, pathway, and compound priorities, but they do not establish biomarkers, therapeutic efficacy, or clinical suitability and require validation in independent cohorts and experimental ALS models.
Abstract
Amyotrophic lateral sclerosis (ALS) is a heterogeneous neurodegenerative disease for which effective disease-modifying therapies remain limited. This study aimed to derive internally recurrent ALS-associated transcriptional signatures and generate directionally interpretable drug-repositioning hypotheses using a consensus machine-learning framework. Two publicly available transcriptomic datasets from motor cortex (E-MTAB-2325) and blood (E-TABM-940) were analyzed using four feature-selection methods within 100 repetitions of 4-fold cross-validation. Probes recurrently selected in models achieving an accuracy of at least 0.90 were prioritized and examined using COGENA pathway enrichment and Connectivity Map drug-signature analysis. Fifteen qualifying models were obtained for the motor-cortex dataset and 55 for the blood dataset. No exact prioritized gene or probe identifier was shared between the two top-100 signatures, but pathway-level integration identified complementary evidence involving glial and immune regulation, proteostasis and vesicle trafficking, MAPK-related stress signaling, cytoskeletal and extracellular remodeling, and RNA-related processes. The motor-cortex dataset additionally emphasized astroglial support, glutamate handling, and inclusion-body regulation, whereas the blood dataset highlighted cytokine regulation and directionally heterogeneous immune, mitochondrial, and metabolic signals. Deferoxamine and disulfiram showed the clearest reversal-compatible profiles in motor cortex, whereas yohimbic acid and atovaquone showed reversal-compatible profiles in blood. Ciprofloxacin, prochlorperazine, and a compound group led by androsterone instead showed concordant connectivity. The results provide transparent, hypothesis-generating gene, pathway, and compound priorities, but they do not establish biomarkers, therapeutic efficacy, or clinical suitability and require validation in independent cohorts and experimental ALS models.