HMW prediction model verification and application to VI process
scale up
To verify the HMW prediction model, different pilot VI runs (50-200L)
were used as the validation data set. The data set covered 7 conditions
and 12 runs. For some conditions, multiple runs were performed using the
same operating parameters in the same mixer. The PAE protein
concentration, working volume, agitation speed, and the acid addition
rate were taken into consideration as the operating parameters, which
were examined by the model simulations to mitigate the risk of excessive
product aggregation (HMW%,VIN of <
2.5%).
The geometry and operating conditions of the validation runs are
presented in Table IV. The EOVIA MF0.1N HClvalues were first examined to validate the CFD model for VIA simulation.
As presented in Table IV, variation between the model predictions and
the experimental EOVIA MF0.1N HCl values were
< 9%. Like trends of the training runs, a linear growth
profile of MF0.1N HCl was observed in a
validation run as shown in Figure 5D.
The low-pH zone profiles were then examined to validate Eq. 3. The
growth of the low-pH zones was visualized by CFD simulations shown in
Figures 5A and 5B using the conditions of S100_2 and X200_1 as
examples. Quantitatively, the ApH3.3 and ILPZ
profiles are shown in Figures 5C, 5E. Across the validation runs, the
EOVIA ApH3.3 values were of comparable magnitude
in the order of 10-4 m2 and much
less than the value of failed pilot lot S100_1 (0.0399
m2). Nevertheless, like the trends of training runs,
exponential growth profiles of both ApH3.3 and
ILPZ were observed in all validation runs.
The HMW%, VIN values were eventually examined to
validate the HMW prediction model. TheHMWtotal,VIN values of the validation runs are
shown in Figure 5F. The difference between the model predictions and the
experimental HMW%, VIN values ranged from 0.03%
in X200_3 to 0.98 % in X200_1 as presented in Table IV. Furthermore,
according to the model simulations, all 12 validation runs would haveHMW%, VIN of ≤ 1.41%, which met the in-process
criterion of < 2.5% HMW%, VIN . Agreed
with the model predictions, the 12 validation runs did achieve ≤ 2.0 %HMW%, VIN with a 100% success rate.
These results suggested that the model developed from training runs were
applicable to the validation runs for mitigating the risk of excessive
HMW formation. This case also demonstrated the application of the HMW
prediction model to facilitate the VI process scale up directly from
bench to pilot/production scale.