Deep Learning and Machine Learning Modeling Identifies Thidiazuron as a Key Modulator of Somatic Embryogenesis and Shoot Organogenesis in Ferula assa-foetida L.

Khushbu Kumari, Samaksh Mittal, Kritika Sharma, Sanatsujat Singh, Jyoti Upadhyay, Vishal Acharya, Virender Kadyan, Sudesh Yadav, Rohit Joshi

The spice Ferula assa-foetida L., also known as asafoetida, is widely recognized for its medicinal and culinary applications. The non-native status of the plant and the prolonged dormancy of its seeds pose significant challenges for large-scale cultivation in India. In vitro organogenesis offers an effective solution to these obstacles. Establishing reliable in vitro regeneration protocols requires standardized statistical methods to evaluate univariate and multivariate data for optimizing specific traits. However, these methods have limitations when handling complex, nonlinear inputs, often producing large prediction errors that reduce the reliability of trait optimization. This study developed an in vitro regeneration system for F. assa-foetida L. and identified optimal PGRs for somatic embryogenesis and shoot organogenesis through image-based morphological analysis. Predictive models were created using DL and ML algorithms. Calli induced from leaf explants was cultured on the Murashige and Skoog medium supplemented with various combinations and concentrations of thidiazuron (TDZ), 6-benzylaminopurine (BAP), and α-naphthaleneacetic acid (NAA), as experimental variables. Seven ML approaches, namely random forest (RF), support vector machine (SVM), k-nearest neighbours (kNN), decision tree (DT), extreme gradient boosting (XG Boost), naïve bayes, and logistic regression, alongside five DL models—convolutional neural network (CNN), MobileNet, region-based convolutional neural network (RCNN), residual neural network (ResNet), and visual geometry group (VGG19)—were employed to predict the best PGRs for somatic embryogenesis and shoot organogenesis. Among them, the convolutional neural network (CNN) achieved the highest accuracy (87%), outperforming baseline ML models such as logistic regression and decision tree (82%). This pioneering study in F. assa-foetida L. presents an AI-driven, image-based framework for predicting optimal PGRs, offering a scalable approach to enhance micropropagation in endangered medicinal plants.