Automated Extraction and Meta-Analysis of a Century of Motor-Unit Research with NeuromechaniX

Abstract The scientific literature on human motor units and electromyography (EMG) spans over a century (1925-2025), comprising research impossible to synthesize manually. We introduce NeuromechaniX, a domain-specific platform for automated extraction and meta-analysis of this literature. The core component, MUscraper, is a large language model pipeline that extracts approximately 200 structured metadata fields, organized into 17 major sections spanning participant demographics, EMG acquisition parameters, muscle identification, task protocols, decomposition methods, and motor-unit outcomes, from ∼2,000 publications on human limb muscles. This automated extraction transforms heterogeneous narrative reports into a standardized, queryable database at a scale not achievable through manual review. From this dataset, we analyzed motor-unit discharge rate across 208 studies examining seven muscles. Our analyses reveal that discharge rates differ significantly among muscles (p