An in vitro slow-growth callus conservation strategy for several medicinal plants using response surface methodology and machine learning

In vitro culture of callus is an effective method for conserving the genetic resources of economically important crops. However, continuous subculturing is a costly and labor-intensive process. Therefore, establishing an efficient in vitro long-term conservation system applicable to various plant species is required. In this study, calli derived from five medicinal plant species, Camellia japonica (Cj), Centella asiatica (Ca), Ligusticum afficinale (Lo), Panax ginseng (Pg), and Sageratia thea (St) were used to optimize storage conditions and establish a suitable in vitro conservation strategy. Calli cultures were maintained on the appropriate culture medium at 5 °C for 120 days. Cell viability and regrowth rate were assessed during the storage period, and correlations between growth and antioxidant traits were examined. Subsequently, pretreatment optimization using sucrose (3–9%), MeJA (0-200 µM), and CTR (0–20 mg/L) was performed using RSM, and the effects of pretreatment and storage temperature on callus conservation were evaluated. In addition, machine learning models such as GRNN, MLP, RF, SVR, and XGBoost were applied to the experimental data. The findings demonstrated that, in comparison to Ca and St, Lo, Pg, and Cj exhibited noticeably higher antioxidant capacity while maintaining high cell viability and regrowth rates. Interestingly, Ca and St drastically decreased viability and regrowth after 60 days, whereas Lo, Pg, and Cj maintained viability and regeneration for up to 90 days. Both TPC and AC (measured by FRAP assay) showed a high positive correlation with cell viability and growth rate, according to correlation analysis. RSM predicted that the optimal pretreatment medium for enhancing antioxidant capacity was the species-specific proliferation medium supplemented with 3% sucrose, 135 µM MeJA, and 20 mg/L CTR, while the highest TSSC was achieved using the species-specific proliferation medium supplemented with 9% sucrose and 200 µM MeJA. When the storage temperature was set to 15 °C following the antioxidant-enhancing pretreatment derived from the RSM optimization, all five species showed improved cell viability and regrowth rates, among the storage methods. Among the ML models tested, XGBoost demonstrated the most stable and accurate predictive performance for both viability and regrowth during in vitro conservation. SHAP-based analysis of the XGBoost model, focusing on regrowth rate, revealed that storage duration was the most influential factor affecting regrowth prediction, followed by storage temperature, while pretreatment conditions showed secondary but meaningful contributions. This study demonstrates that long-term callus conservation is closely associated with AC and TPC. Medium supplemented with sucrose 3%, 135 µM MeJA, and 20 mg/L CTR, followed by storage at 15 °C, significantly improved viability and regrowth, and calli could be maintained up to 120 days without subculturing. This approach provides an efficient and broadly applicable in vitro strategy for the conservation of diverse plant genetic resources.