Calibration of a chemometric model by using a mathematical process model instead of offline measurements in case of a H. polymorpha cultivation

Abstract

Data driven regression models such as Principle Component Regression (PCR) or Partial Least Square Regression (PLS) in combination with spectroscopic methods are increasingly applied in bioprocess monitoring. However, as the name “data driven” implies, the calibration of these regression models requires a large amount of predictor (X) and response (Y) data. The predictor data in this case mostly consists of the spectroscopic data, which are easy to generate in large quantities, but the response data typically involves offline measurements in the laboratory that require much more effort to perform in large numbers.It will be shown that in case of a H. polymorpha cultivation performed in microtiter plates, those tedious offline measurements for response data can be replaced by a mathematical process model. Here, an exponential growth model in an ideal stirred tank reactor with lag-time is applied, which has three parameters (lag time, specific growth rate, and yield coefficient). Furthermore, it will be demonstrated that knowledge about the parameter values of this process model is not required, as these values can be determined from 2D fluorescence spectra alone. The only required information about the cultivation is the predictor data, 2D-fluorescence spectra in this case, and the initial state of at least three different cultivation runs, that is the initial values of biomass and substrate (glycerol) concentration.The smallest prediction error for biomass and glycerol obtained by the new calibration procedure are 0.19 g/L and 0.79 g/L respectively, and 0.19 g/L and 1.12 g/L, if a classical procedure using off-line measurements is applied. The inherently calculated process parameters of lag time, specific growth rate and yield coefficient are 4.77 h, 0.154 h−1, and 0.457 g/g, which are similar to values which are determined with offline measurements and least square fit 4.48 h, 0.139 h−1, 0.466 g/g respectively.