Application of artificial neural network and machine learning algorithms for modeling the in vitro regeneration of chickpea (Cicer arietinum L.)

In vitro whole plant regeneration protocol of desi chickpea was established followed by generating the model prediction using different machine learning algorithms. Surface sterilized seeds with 5% bleach (NaOCl) were cultured on Murashige and Skoog medium with six different concentrations (0.25, 0.50, 1.00, 1.50, 2.00 and 3.00 mg/L) of Benzylaminopurine (BAP), Thidiazuron or Kinetin (KIN). BAP and KIN enriched medium produced normal shoots and relatively high shoot induction frequency (%) was recorded 98.14–100%. Application of TDZ induced medium generated five and sevenfold more shoot counts than BAP and KIN respectively. Maximum shoot length was recorded as 10.67 cm and 9.90 cm on medium containing 0.25 mg/L BAP or 0.50 mg/L KIN respectively. Regenerated shoots were rooted on medium containing IBA. The establishment of plantlets were done in growth chamber adjusted to 24 ± 1 ºC, 60% relative humidity and 16 h light photoperiod where plant established flowering and set seeds. Machine learning algorithms of support vector regression, gaussian process regression, XGBoost, random forest (RF) models and multilayer perceptron neural network were used to predict the shoot count and length. It was found that the RF model indicated the highest performance to predict the outputs. To confirm the validity of the models, Leave-One-Out cross validation was used. The evaluation was performed using the parameters of R2 (coefficient of determination and MSE (mean squared error) scores. In our study, The R2 and MSE scores of RF model were 0.99 and 2.86 for shoot count, 0.98 and 0.29 for shoot length respectively.