Title: A data-driven framework for reactive musculoskeletal modeling: Integrating proteomics, biomechanics, and machine learning
Abstract:
Reactive, predictive musculoskeletal (MSK) models that integrate high-throughput proteomic, biomechanical, and physiological data offer a powerful tool for forecasting tissue adaptation under altered mechanical loading. Spaceflight—and its ground-based analogs such as rodent hindlimb unloading, human 6° head-down tilt, and limb offloading protocols—accelerates MSK deconditioning, providing an efficient testbed for model calibration. To address the lack of integrated, data-driven platforms in current MSK simulations, we have developed a modular framework that leverages longitudinal serum proteomics, analog biomechanical metrics, and vascular–bone intervention outcomes to train machine-learning algorithms capable of predicting multi-protein responses, ocular pressure shifts, and bone density changes. These initial modules establish the foundation for a reactive, in silico MSK platform.
Methods: In partnership with the University of Central Florida (UCF), we assembled a multi-source dataset—longitudinal serum proteomics from rodent hindlimb unloading studies, public E-PROT-10 data, and a simulated primer dataset—to train a Random Forest–based, multi-output regression pipeline in scikit-learn. Model interpretability was achieved via SHAP, and performance was evaluated using R² and MAE. Concurrently, at Embry-Riddle’s Biomechanics and Aerospace Laboratory (ERBal), we conducted a 6° head-down tilt study in healthy volunteers to quantify sex-based differences in intraocular pressure (IOP). AdventHealth Innovation Labs is contributing data from one-limb rest protocols and vascular–bone interventions, expanding the mechanical stimuli represented in the model. We are also integrating participant genetic profiles to account for inter-individual variability.
Results: Across 15 muscle-related proteins, the pipeline achieved a mean R² of 0.70 and MAE of 0.18. Top-performing proteins (FABP3, Lipocalin-2, Annexin A1) exceeded R² = 0.85. In the IOP study, female participants exhibited ~20% greater IOP elevation than males (p < 0.01), suggesting hormonal or anatomical influences. Genetic integration is ongoing and expected to enhance model specificity.
Conclusion: These initial modules offer a reactive, data-driven foundation for predictive MSK modeling. By layering multi-domain datasets—proteomic, mechanical, ocular, vascular, and genetic—this framework supports individualized, in silico evaluation of countermeasures for spaceflight and rehabilitation applications.
