Machine learning-guided optimization and Shiny-based deployment for in vitro shoot propagation of sumac (Rhus coriaria)
摘要
The in vitro micropropagation of perennial woody and shrubby plants often relies on basal media like Murashige and Skoog medium (MS) and woody plant medium (WPM). However, wild forms of semi-woody and shrubby plants, known for their recalcitrance and high phenolic content, present challenges for in vitro propagation. Advances in plant tissue culture focus on developing new media for species resistant to conventional protocols. This study evaluated shrub plants medium (SPM) alongside MS, WPM, and Driver and Kuniyuki walnut medium (DKW) for the in vitro propagation of sumac. The media were supplemented with varying 6-Benzylaminopurine (BAP) hormone concentrations (0, 0.5, 1, and 1.5 mg L⁻¹). Responses were shoot number (SNE), shoot length (SL), fresh weight (FW), and dry weight (DW). The highest SNE (22.23 number) and SL (46.08 mm) were observed in 0.5 mg L⁻¹ BAP + SPM, followed by SPM (control) with 18.74 shoots per explant and 45.65 mm shoot length. The highest FW (4.42 g) and DW (0.207 g) were observed in 1.5 mg L⁻¹ BAP + SPM. The highest SNE value (15.56 number) and SL value were at 0.5 mg L⁻¹ BAP treatment. The highest FW value (3.67 g) was at 1.5 mg L⁻¹ BAP treatment, while the highest DW value (0.17 g) was at 1 mg L⁻¹ BAP treatment. Support vector regression (SVR), XGboost, elastic net (EN), and Gaussian processes (GP) algorithms were used to predict in vitro features. The deployment model for each attribute was chosen based on the highest R2 along with the lowest minimum root mean square error (RMSE), mean absolute percentage error (MAPE), and mean absolute deviation (MAD) ratings. XGboost had the highest level of accuracy for SNE, FW, and DW when compared to other models, while GP demonstrated the highest level of generalization for SL. The models demonstrated strong predictive ability, achieving R² values above 0.90 for parameters (SNE, SL, FW, DW), confirming their reliability for in vitro optimization. During the pre-experimental exploration phase, the best models (Xgboost and GP) were presented in an interactive Shiny interface. Optimization of sumac in in vitro growth conditions is possible using an SPM-based procedure and machine learning. Shiny can allow users to explore medium × BAP combinations and reduce experimental overhead.