Electromyography Agreement and Musculotendon Parameter Sensitivity Analysis of Predictive Cycling Simulations

Predictive musculoskeletal simulations are increasingly used to study human movement, as they allow estimation of internal biomechanical variables that are costly or physically demanding to measure experimentally. In cycling, these approaches are particularly attractive for rehabilitation and assistive applications, where systematic exploration of cadence, task constraints, and neuromuscular strategies could inform clinical decision making. However, the extent to which predictive cycling simulations reproduce experimentally measured muscle activation patterns, and how sensitive this agreement is to common musculotendon parameter assumptions, remains insufficiently quantified. In this study, we show that predictive cycling simulations capture cadence-dependent activation timing and phase relationships across eight lower-limb muscles per leg, but exhibit muscle-specific limitations in activation magnitude and robustness to parameter variations. Across 10 healthy participants and 2 participants with incomplete spinal cord injury, simulated muscle activations showed moderate agreement with electromyography (EMG) primarily in temporal features rather than amplitude, with the tibialis anterior and knee extensors showing the most consistent correspondence. Agreement was systematically lower for posterior chain muscles, including the gastrocnemius and soleus, across all cadences. Sensitivity analysis revealed that perturbations in tendon slack length produced substantially larger changes in EMG simulation agreement than equivalent perturbations in optimal fibre length, with asymmetric effects across muscles. In contrast, variations in fibre length, maximal isometric force, and pennation angle had comparatively small and spatially diffuse effects on correlation outcomes. These results indicate that EMG simulation agreement in cycling is highly dependent on tendon-related parameters and muscle-specific coordination patterns. These findings provide a quantitative benchmark for the current capabilities and limitations of predictive cycling simulations in a rehabilitation context. More immediately, they highlight tendon slack length as a critical parameter for model tuning and interpretation, supporting more cautious and informed use of predictive simulations as clinical support tools rather than direct surrogates for measured neuromuscular activity.