Lightweight Bayesian-Neural-Network Framework for Quantifying Cementum Regeneration Potential of Periodontal-Ligament Stem Cells

Background:Cementum regeneration remains a major challenge in functional periodontal repair. Traditional computational models often rely on large deep-learning architectures that provide point estimates but limited information on prediction confidence. Objectives:To develop a compact Bayesian neural network (BNN) capable of predicting cementum thickness with high accuracy and well-calibrated uncertainty, while remaining computationally efficient for laptop-class hardware. Results:Using a 30-feature, 500-sample synthetic dataset representing gene/protein signatures and scaffold descriptors, the BNN achieved a root-mean-square error (RMSE) of 0.54 mm and explained approximately 77% of outcome variance. The model also provided interpretable posterior intervals, offering a measure of prediction confidence. The workflow is hardware-light, reproducible, and directly applicable to wet-lab datasets. Conclusion:The proposed BNN framework enables accurate, uncertainty-aware cementum thickness predictions on standard hardware, facilitating reproducible and translational computational modeling for periodontal regeneration research