A classification-based machine learning strategy for nonlinear optimization of the antiparkinson drug L-DOPA and relevant catecholamine intermediates from Faba bean (Vicia faba L.) leavesMohammad Danish Jawed Ansari, Sandip Garai, K. K. Kanaka, Chinta Sravani, Tushar Kashyap, Tanmaya Kumar Sahu, Arnab Roy Choudhury, Vijai Pal Bhadana, Sujay Rakshit, Rakesh Kumar Sinha, and Sujit Kumar Bishi ICAR-Indian Institute of Agricultural Biotechnology, Ranchi, India
E-mail: sujitbishi@gmail.com Received: 7 May 2026 Accepted: 8 June 2026 Abstract: Faba bean (Vicia faba L.) is an underutilized legume that accumulates the antiparkinson drug L-3,4-dihydroxyphenylalanine (L-DOPA) and co-occurring catecholamine (CA) intermediates, i.e., L-tyrosine and tyramine, in its leaf tissues. This study investigated the optimization of simultaneous extraction conditions for these bioactive compounds using a full factorial design (33 = 27 combinations, 81 observations with triplication) incorporating three factors: temperature (30 °C, 45 °C, 60 °C), sample-to-liquid ratio (SL ratio: 1:10, 1:20, 1:30), and hydrochloric acid concentration (HCl: 0.01 N, 0.105 N, 0.20 N). Quantification was performed using Ultra-Performance Liquid Chromatography (UPLC) with validated recovery (87–96%) and intra-day precision (RSD \(\le \) 3.5%). The yield datasets exhibited significant non-normality (Shapiro–Wilk test, p < 0.001 for L-DOPA and tyramine), high skewness (tyramine: 1.27), and non-linear dependency structures (BDS test, p < 0.01), characteristics that limit the reliability of conventional linear regression for optimisation. Accordingly, a Classification and Regression Trees (CART) model was developed, framing yield optimisation as a classification task by categorising responses into Low, Medium, and High-performance tiers based on interquartile thresholds. The CART model achieved classification accuracies of 84%, 84%, and 68% for L-DOPA, tyramine, and L-tyrosine, respectively, on a held-out test set. A naive ensemble of CART, Random Forest, and Gradient Boosting further improved accuracy for L-DOPA to 92%. Variable importance analysis identified HCl concentration as the dominant extraction driver for L-DOPA and tyramine, while SL ratio governed L-tyrosine yield. The identified optimal conditions were: L-DOPA at 30–45 °C, SL 1:20, 0.105 N HCl; tyramine at 45–60 °C, SL 1:10, 0.01 N HCl; L-tyrosine at 45 °C, SL 1:30, 0.01 N HCl. These results support the potential utility of CART-based classification as a complementary, interpretable tool for multi-compound phytochemical extraction optimisation, with full assessment of reproducibility across genotypes and industrial-scale applicability remaining as important future work. Graphical abstract Keywords: CART; Decision support system; Industrial biotechnology; L-DOPA; Machine learning; Process optimization Full paper is available at www.springerlink.com. DOI: 10.1007/s11696-026-05185-7
Chemical Papers 80 (11) 13357–13372 (2026) |
Saturday, November 14, 2026 
|
|||
© 2026 Chemical Papers |
|
|||
