Predição das velocidades específicas da produção de xilitol a partir do bagaço de caju por kluyveromyces marxianus CCA 510, utilizando redes neurais artificiais

Abstract

The northeastern region of Brazil has the highest national production of cashews. From this process, a substantial amount of cashew bagasse remains, which serves as a material for generating various sugars, notably glucose and xylose. Both, processed through biotechnological methods with microorganisms, generate by-products such as ethanol and xylitol, which have significant commercial value. The main objective of this study is to propose a mathematical model using Artificial Neural Networks (ANN) applied to the biotechnological process of obtaining xylitol from cashew bagasse hydrolysate using the yeast Kluyveromyces marxianus CCA 510. This allows for examining the behavior of concentration profiles of biomass, xylose, glucose, ethanol, and xylitol, as well as the specific rates of each. Experimental data from the thesis "Biotecnological Production of Xylitol from Cashew Bagasse Hydrolysate” (ALBUQUERQUE, 2014), were used for this purpose. The data underwent treatment through regressions and interpolations to smooth experimental measurement noise and increase the dataset for training Artificial Neural Networks (ANNs). In this study, specific rates are predicted by ANNs as a function of the instantaneous concentration of cells and substrate for each type of cashew bagasse hemicellulosic hydrolysate (HBC). Simulation results were statistically evaluated using the Mean Absolute Error (MAE), in addition to validation against a new experiment. According to the results, the models used proved effective in predicting specific rates according to the behavior of the evaluated concentration profiles for each tested condition, with R² for training and validation from 0.99 to 0.94 for biomass, 0.99 to 0.90 for xylose, 0.99 to 0.93 for glucose, 0.96 to 0.90 for xylitol and 0.98 to 0.94 for ethanol, showing no significant variation between the analyzed conditions. This indicates good quality in the maden adjustments and the suitability of the models to the observed data. High accuracy values were achieved where MAE was less than 1%.