Abstract

Plant tissue culture media composition prediction is valuable and can diminish the spending time and costs of releasing protocols. The present study assessed the possibility of using some imperative supervised Machine Learning (ML11. Machine Learning) algorithms, including Support Vector Machine (SVM22. Support Vector Machine), Gene Expression Programming (GEP33. Gene Expression Programming), and Gradient Boosting Decision Tree (GBDT44. Gradient Boosting Decision Tree), in predicting optimized in vitro media composition for proliferation and rooting of Salvia macrosiphon Boiss. and comparing them with a linear regression method, i.e., Bayesian Ridge Regression (BRR55. Bayesian Ridge Regression). Input parameters included different concentrations of macro- and micro-nutrients, vitamins, and Plant Growth Regulators (PGRs66. Plant Growth Regulators). The accuracy of constructed models’ performance was investigated according to Root Mean Square Error (RMSE77. Root Mean Square Error), Mean Absolute Percentage Error (MAPE88. Mean Absolute Percentage Error), and Coefficient of Determination (R299. Coefficient of Determination). Particle Swarm Optimization (PSO1010. Particle Swarm Optimization) system was employed to optimize the developed formulations using superior prediction models. Results revealed that ML methods had higher prediction accuracy than BRR. The GEP models were subsequently selected for optimization by PSO. According to hybrid GEP-PSO models, modified MS medium including 0.54 × NH4NO3, 1.94 × KNO3, 0.93 × CaCl2, KH2PO4, MgSO4, 2.65 × minors, 1.28 × vitamins, and myoinositol and supplemented with 0.93 mg/L 6-benzylaminopurine (BAP1111. 6-benzylaminopurine) and 0.05 mg/L indole-3-acetic acid (IBA1212. Indole-3-acetic acid) could bring about optimal proliferation. Based on the same models, MS medium containing 0.72 × macros, 1.49 × minors, 1.23 × vitamins and myoinositol, 0.37 × sucrose and 1.73 × FeEDDHA and supplemented with 1.97 mg/L 1-naphthaleneacetic acid (NAA1313. 1-naphthaleneacetic acid) and 0.53 mg/L IBA could result in the optimized rooting. This study shows the effectiveness of GBDT as an advanced ML algorithm for predicting the formulation of plant tissue culture media. GEP-constructed models were selected for optimization due to the simplicity and clearness of their results as an entire formula. At the same time, two more ML models used in this study also have adequate accuracy to be selected by the researcher.

Full Text
Published version (Free)

Talk to us

Join us for a 30 min session where you can share your feedback and ask us any queries you have

Schedule a call