Harnessing artificial neural networks to model caffeine degradation by High-Yield biodiesel algae Desmodesmus pannonicus.

In this study, Desmodesmus pannonicus IITISM-DIX2, outperforming Chlorella sorokiniana IITISM-DIX3 in caffeine degradation, was used to develop an artificial neural network (ANN) model for predicting caffeine removal efficiency under varying pH, photoperiods, caffeine, and indole acetic acid (IAA) concentrations. The ANN model, designed with a 4-15-1 multilayer perceptron and trained on 120 data points, achieved high predictive accuracy (R > 0.96) and bias/accuracy factors between 0.95-1.11. Sensitivity analysis identified pH as the most critical factor. IAA enhanced lipid content in Desmodesmus by 91 % in caffeine-containing simulated wastewater. FAME analysis was performed under optimal lipid-producing conditions (10 ppm caffeine, 5 ppm IAA). IAA upregulated key metabolic pathways, increasing secondary metabolites in Desmodesmus and Chlorella. In summary, the modeling results are key for improving system performance, guiding parameter selection to enhance caffeine removal by Desmodesmus. IAA also enhanced resilience and lipid yield, increasing the economic feasibility of caffeine removal and biofuel production.