Machine learning-assisted evaluation of antioxidant and metal chelating capacities in in vitro propagated Ceratophyllum demersum L. under different LED light conditions

Abstract The aim of this study is to explore the effects of different LED light spectra on the antioxidant capacity of Ceratophyllum demersum L. under in vitro culture conditions, using machine learning techniques to predict and analyze the plant’s metabolic responses. Both White and Red LEDs achieved 100% shoot regeneration, with Red LED producing the highest shoot count (85.27) and longest shoots (3.16 cm). Additionally, antioxidant analysis showed significant variations in phenolic and flavonoid content based on light and extraction methods. Red LED acetone extracts had the highest phenolic content (63.99 µg GAE/mg), while Blue LED acetone extracts yielded the highest flavonoid content (167.58 µg QE/mg). White LED acetone extracts showed the strongest DPPH scavenging activity (90.14% at 400 µg/mL), indicating broad-spectrum light enhances antioxidants. Metal chelation was highest in White LED water extracts. Numerous machine learning techniques were employed to predict DPPH radical scavenging and metal chelation activities. XGBoost emerged as the top-performing algorithm for DPPH activity prediction, achieving the lowest MAE (3.754) and the highest R² (0.887), along with one of the lowest RMSE values (5.027). MLP (Multilayer Perceptron) also showed strong performance with relatively low RMSE (5.528) and MAE (4.200) on the test set. For metal chelation activity, Cubist demonstrated the best performance, with the lowest test RMSE (5.129) and MAE (4.141) values, along with one of the highest R² values (0.899). This study highlights the potential of machine learning algorithms in predicting antioxidant activities and the significant impact of light conditions on these activities.