Empirical Simulation and In Silico Evaluation of an Ethically Governed, AI-Enhanced CRISPR Prime Editing Framework for TOR1A Mutation Correction in Dystonia

Hereditary dystonia, particularly the isolated form of early onset, is often caused by a deletion of the GAG in the TOR1A gene, leading to a dysfunctional TorsinA protein and severe motor impairments. In this study, we investigate an AI-enhanced CRISPR prime editing framework designed for the precise correction of the TOR1A mutation. Our framework integrates the generation of candidate pegRNA with an empirical simulation of training data, enabling the tuning of a random forest model that predicts editing efficiency. This approach generates robust quantitative outputs (validation R2 score of 0.701). SHapley Additive exPlanations (SHAP) revealed that GC content (mean SHAP = 0.22) and off-target risk (mean SHAP = 0.18) were the strongest drivers of predicted efficiency. The GC content showed strong positive correlations with melting temperature (r = 0.75) and guide efficiency (r = 0.82). By linking the simulated modeling work to the practical challenges of genetic neurosurgery in dystonia, a proof-of-concept framework is presented that is both ethically governed and clinically pertinent. This work looks toward a future of scalpel-less surgery by laying a foundation for future integration with empirical datasets and more advanced deep learning techniques.