Implementation of multilayer perceptron (MLP) and radial basis function (RBF) neural networks for predicting Shatavarin IV content in Asparagus racemosus accessions.

The principal bioactive ingredient of Asparagus racemosus is Shatavarin IV, a steroidal saponin found in roots with a wide range of medicinal properties. A. racemosus roots are used in Ayurveda system of medicine for curing a spectrum of diseases like diarrhea, nervous disorders, dysentery, dyspepsia, tumors, inflammations, neuropathy, hyperdipsia, hepatopathy, cough, hyperacidity, bronchitis and certain infectious diseases. Many of the factors influence the production and accumulation of secondary metabolites including environmental, developmental, and genetic factors. An artificial neural network (ANN) based model was developed in this study to evaluate the influence of abiotic factors (climate and soil) and to forecast a suitable site for collecting and cultivating A. racemosus with high Shatavarin IV concentration. The experimental dataset contains 103 A. racemosus accessions collected from various phytogeographical zones of Odisha. Fourteen input parameters such as soil factors (nitrogen, phosphorus, potassium, sulphur, organic carbon, pH, electrical conductivity) and climatic factors (altitude, relative humidity, UV index, average temperature, minimum temperature, maximum temperature, annual precipitation) were considered for the study. For training, testing and validation, the datasets were randomly divided into 75%, 15%, and 15%, respectively. A validated high performance thin layer chromatography (HPTLC) method was used to determine the Shatavarin IV content. The quantity of Shatavarin IV in A. racemosus root extracts varied from 0.010% to 0.400% on dry weight basis among 103 accessions collected from various regions of Odisha. Two ANN model algorithms namely MLP (multilayer perceptron) and RBF (radial basis function) were used for prediction and optimization of Shatavarin IV content using environmental and soil factors. The ANN model with MLP algorithm was found to be better model as compared to the RBF algorithm. The results demonstrated that an ANN-MLP architecture with a single hidden layer of 8 neurons having 14–8–1 topology could reliably predict Shatavarin IV content with a coefficient of determination (R2) of 0.970. The prediction efficiency of the developed model was 90.740%. Sensitivity analysis revealed that nitrogen and phosphorus have a greater impact on Shatavarin IV biosynthesis than the other parameters. The content of Shatavarin IV could be increased from 0.245% to 0.300% by managing these sensitive factors in the developed model. The developed ANN model would be very useful in predicting the best regions/sites for A. racemosus with optimum Shatavarin IV yield.