Light-emitting diodes induced in vitro regeneration of Alternanthera reineckii mini and validation via machine learning algorithms

Abstract

Optimization of in vitro regeneration protocol using multiple input variables is highly significant, and can be achieved by validating the data using machine learning algorithms. Shoot tip and nodal segment explants of Alternanthera reineckii mini were inoculated on Murashige and Skoog (MS) medium enriched with different concentrations of benzylaminopurine (BAP), and cultured under five different monochromic light-emitting diodes (LEDs). The attained results were validated through the application of four different supervised machine learning models (RF, XGBoost, KNN, and GP). The prediction of the data were validated by using regression coefficient (R2), mean squared error (MSE), and mean absolute percentage error (MAPE) performance metrics. Results revealed R2 values of 0.61 and 0.59 for shoot counts and shoot length, respectively. The results of MSE were registered between 3.48–5.42 for shoot count and 0.40–0.74 for shoot length, whereas, 28.9–35.1% and 13.2–18.4% MAPE values were recorded for both shoot count and shoot length. Among the utilized models, the RF model validated and predicted the results more accurately, followed by the XGBoost model for both output variables. The results confirm that ML models can be used for data validation, and opens a new era of employing ML modeling in plant tissue culture of other economically important plants. Schematic structure presenting input features and outputs together with ML models, used validation and performance metrics