Machine Learning-Driven Cellular Automata for Simulating Plant Growth Regulator-Optimized Callus Induction in Caladium bicolor

Caladium bicolor (Aiton) Vent. is an ornamental foliage plant; efficient in vitro regeneration remains a challenge. This study established an integrated computational workflow combining Latin hypercube sampling (LHS), Gaussian process regression (GPR), and Cellular Automata (CA). Using cultivar ‘Liuguang Diecai’, 18 PGR combinations were designed by LHS and evaluated for callus induction and adventitious shoot regeneration, quantified by shoot induction rate and shoot number per callus after 45 days on MS + 2.0 mg/L 6-BA + 0.2 mg/L NAA. Four machine learning models were benchmarked against RSM; multi-objective optimization was performed via NSGA-II and TOPSIS, and the CA model was calibrated with dynamic morphological indicators. Results revealed that 6-BA exerted the strongest nonlinear control over callus induction rate and formation time; 2,4-D governed biomass accumulation; and NAA modulated formation time. The model-recommended combination (2.0 mg/L 2,4-D + 0.6 mg/L 6-BA + 0.8 mg/L NAA) was validated, yielding an 80.1% callus induction rate and 5.41 g fresh weight—less than 1% deviation from GPR predictions. The calibrated CA model reproduced spatiotemporal growth, with mean IoU 0.83. Optimized calli showed enhanced regenerative capacity, achieving 88.9% adventitious shoot induction. By integrating predictive modeling with dynamic simulation, this workflow offers a promising strategy for optimizing in vitro regeneration systems.