Application of a Multilayer Perceptron Artificial Neural Network for the Prediction and Optimization of the Andrographolide Content in Andrographis paniculata

Andrographolide, the principal secondary metabolite of Andrographis paniculata, displays a wide spectrum of medicinal activities. The content of andrographolide varies significantly in the species collected from different geographical regions. Therefore, this study aims at investigating the role of different abiotic factors and selecting suitable sites for the cultivation of A. paniculata with high andrographolide content using a multilayer perceptron artificial neural network (MLP-ANN) approach. A total of 150 accessions of A. paniculata collected from different regions of Odisha and West Bengal in eastern India showed a variation in andrographolide content in the range of 0.28–5.45% on a dry weight basis. The MLP-ANN was trained using climatic factors and soil nutrients as the input layer and the andrographolide content as the output layer. The best topological ANN architecture, consisting of 14 input neurons, 12 hidden neurons, and 1 output neuron, could predict the andrographolide content with 90% accuracy. The developed ANN model showed good predictive performance with a correlation coefficient (R²) of 0.9716 and a root-mean-square error (RMSE) of 0.18. The global sensitivity analysis revealed nitrogen followed by phosphorus and potassium as the predominant input variables influencing the andrographolide content. The andrographolide content could be increased from 3.38% to 4.90% by optimizing these sensitive factors. The result showed that the ANN approach is reliable for the prediction of suitable sites for the optimum andrographolide yield in A. paniculata.