Using chemometric models to predict the biosorption of low levels of dysprosium by Euglena gracilis

The critical rare earth element dysprosium (Dy) is integral for sustainable technologies. What is concerning is that Dy is in imminent short supply and no current replacements yet exist, coupled with increasing environmental Dy levels influenced by anthropogenic activities. This study applies chemometric methods such as response surface methodology and artificial neural networks to predict low Dy removal levels using the biosorbent Euglena gracilis. A three-factor Box-Behnken experimental design was conducted with initial concentration (1 to 100 µg L^-1), contact time (30 to 180 min), and pH (3 to 8) as the three independent variables, and percentage removal and sorption capacity (q) as dependent variables. Using Dy percentage removal as response, for the worst and best conditions ranged from 0 to 92% respectively, with an average removal of 66 ± 4%. Using sorption capacity (q) as a different response variable, q varied from 0 to 93 µg/g with 27 ± 4 µg/g capacity as average. Maximum removal was 92% (q = 93 µg/g) was at pH 3, a contact time of 105 min and at a concentration of 100 µg/L. Using sorption capacity as the response variable for ANOVA, pH and metal concentrations were statistically significant factors, with lower pH and higher metal concentration having improved Dy removal, with a desirability near 1. Statistical tests such as analysis of variance, lack-of-fit, and coefficient of determination (R²) confirmed model validity. A 3-10-1 ANN network array was used to model experimental responses (q). RSM and ANN effectively modeled Dy biosorption. E. gracilis proved to be a cheap and effective biosorbent for Dy biosorption and has the potential to remediate acid mine drainage areas exhibiting low Dy concentrations.