(89e) Process Knowledge for Drug Substance Production Via Kinetic Modeling, Parameter Estimability Analysis and Reaction Optimization | AIChE

(89e) Process Knowledge for Drug Substance Production Via Kinetic Modeling, Parameter Estimability Analysis and Reaction Optimization

Authors 

McMullen, J., Merck & Co.
Wyvratt, B. M., Merck & Co., Inc.
McAuley, K. B., Queen's University
A fundamental model is developed to study the formation of 2,6-difluoropurine-9-THP from starting material 2,6-dichloropurine-9-THP. The 2,6-difluoropurine-9-THP product is an intermediate used in the synthesis of Islatravir (MK-8591), a therapy for treatment of HIV. The corresponding reaction scheme is shown in Scheme 1 where the starting material is labeled (SM).1,2 Trimethylamine (TMA), a gaseous material, is bubbled into the liquid solution in the reactor to start the first reaction. Kinetic parameters for the first reaction and subsequent reactions are estimated from 26 batch reactor experiments. An Error-in-Variables-Model (EVM) approach is used for parameter estimation to address uncertainty in initial concentrations of TMA. Uncertainty in the TMA charge stems from the combined variability in the feed flowrate of TMA and in the TMA charge time. The EVM approach in the current study ranks the combined set of parameters and uncertain inputs from most estimable to least estimable. Then it estimates the true values of the uncertain inputs along with the model parameters. Finally, a mean-squared-error criterion is used to determine the appropriate parameters and inputs that should be estimated, based on the ranked list to avoid overfitting.2 This parameter subset selection method is used to determine that 33 out of 39 model parameters should be estimated along with 26 uncertain initial concentrations. The remaining six parameters are kept at their initial values to prevent overfitting of available data. EVM parameter estimates are compared with estimates obtained using a traditional weighted-least-squares (WLS) approach that neglects input uncertainties. The EVM estimates provide a better fit to the data and, as shown using cross-validation, improved accuracy for model predictions. The resulting model and EVM parameter values are used to find reactor conditions that maximize product yield while obeying constraints on temperature, the initial ratio of TMA to starting material, batch time, and the volume of solvent. An optimal yield of 92.04 % is predicted, which is higher than the yield of 90.26 % at the best experimental conditions in the data set. Contour plots are used to highlight the insensitivity of the optimal yield to batch time and solvent volume, indicating that a yield of 91.83 % could be obtained using a 50 % lower batch time and 33 % less solvent.

Scheme 1. Reaction scheme used to produce 9-THP-2,6-difluoropurine, which is labeled (P) in the black box

References

1. Hong CM, Xu Y, Chung JY, et al. Development of a commercial manufacturing route to 2-fluoroadenine, the key unnatural nucleobase of islatravir. Organic Process Research & Development. 2020;25(3):395-404.
2. Moshiritabrizi I, Abdi K, McMullen JP, Wyvratt BM, McAuley KB. Parameter estimation and estimability analysis in pharmaceutical models with uncertain inputs. AIChE Journal. 2023:e18168.