An alternative approach to biokinetic modelling for phenol pollutant degradation and microbial growth using Genetic Programming.

Biokinetic models can optimise pollutant degradation and enhance microbial growth processes, aiding to protect ecosystem protection. Traditional biokinetic approaches (such as Monod, Haldane, etc.) can be challenging, as they require detailed knowledge of the organism's metabolism and the ability to solve numerous kinetic differential equations based on the principles of micro, molecular biology and biochemistry (first engineering principles) which can lead to discrepancies between predicted and actual degradation rates. More recently, data-driven machine-learning techniques have emerged as a promising alternative for modelling microbial systems. A few machine learning models (such as ANN, SVM, RF, DT, XG BOOST, etc.) have been used recently for modelling phenol degradation, but they lack the robustness of generating mathematical models. This gap is addressed in this study using Genetic Programming (GP) as the modelling approach for modelling the phenol degradation. This study utilises the microalgae Acutodesmus Obliquus, finding that phenol degradation of 98% required 216 hours. Both the traditional kinetic approach and the Genetic Programming (GP) approach were used to determine the specific growth rate (µ_max) and saturation constant (K_s). It is noted that without any a priori information on the form of the mathematical mode, GP can evolve a model which closely fits the Monod kinetics, thus demonstrating that data-driven models can bring out the first engineering principles on which biokinetic models are dependent or framed in a most swift and effective way. Performance was assessed using root mean square error (RMSE) and correlation coefficient (R), with the GP model showing superior predictive accuracy.