How to Replicate your Product Across Multiple Scales or Sites
When is the day done for a product formulator? Although it can be tempting to walk away once the product has been perfected in the lab, some believe that product formulation is not truly complete until the product is successfully produced at the final manufacturing scale or site.1 Replicating lab-scale results at a larger manufacturing scale can be a challenging task. What is more, completing this task in a cost- and time-effective manner is critical to the successful release of a new product to market.
A related product formulation challenge is that of product transfer from one existing manufacturing site to another. Here the goal is to reproduce, at a target plant, the same product quality characteristics achieved at the source plant.2 Among other factors, a lack of detailed mechanistic models, and differences in process design and process measurements at the two sites typically complicates this transfer problem. Nonetheless, it may be necessary to produce the product at two different sites in order to fulfill production quotas.
The ProSensus Approach2,3,4,5
Whether the challenge is process scale-up or product transfer, multivariate statistical analysis (MVA) is a powerful tool that can be used by product formulators and process experts to efficiently guide the selection of operating conditions that will replicate the desired product.
- Both challenges assume that historical manufacturing data is available for similar products from both the source (or smaller scale) and target (or larger scale) sites. This enables assembly of an amalgamated dataset, composed of historical data from both manufacturing sites. Once assembled, this dataset is used to identify a statistical relationship between the source and target sites.
- The MVA approach is well-equipped to handle differences between operating condition measurements at each site. One condition however, is that the same quality variables must be measured at each site.
PLS Inversion and Joint-Y PLS
Two approaches are available to solve product transfer and product scale-up problems in the MVA framework; these are PLS Inversion and Joint-Y PLS (JYPLS).
After data pre-processing, the next step of either approach is to confirm that the various products presently manufactured at the two different sites have similar latent structure; in other words, the first assumption listed above must be mathematically proven. To accomplish this goal, a PCA model is built on the quality variables of the target site. The quality variables of the source site are subsequently projected onto this model. The resulting location of these projections, relative to the original score space, and the SPE reveal the validity of the first assumption. Next, a specialized PLS model is built incorporating the following historical data:
The specific data structure used by each of the two methods differs, as does the approach to inverting the PLS model. Regardless, by completing inversion by either method, one arrives at the solution in the reduced (latent) space. Due to the relative dimensionality of X and Y common to this type of dataset, the inversion is often underdetermined and results in a “window” of solutions that should all result in the same desired product, rather than a single solution. The window is a result of a null space component; within this null space, changes to the operating conditions should have no impact on the product quality.
Finally, the window of physical operating conditions can be computed from the reduced space solution and the null space component. The preferred solution is often the one achieved by keeping operating conditions at the target (or larger scale) site as close as possible to past operating conditions.
Replicate Your Product Today
- Courtney, P. (2017). Predictions for the future of formulation. European Pharmaceutical Review. Volume 22 Issue 1.
- Garcia Munoz, S., MacGregor, J., Kourti, T. (2005). Product transfer between sites using Joint-Y PLS. Chemometrics and Intelligent Laboratory Systems. 79(2005) 101-114.
- Jaeckle, C., MacGregor, J. (2000). Product transfer between plants using historical process data. AIChE Journal. October 200. Vol. 46, No. 10.
- Jaeckle, C., MacGregor, J. (2000). Industrial applications of product design through the inversion of latent variable models. Chemometrics and Intelligent Laboratory Systems. 50(2000) 199-210.
- Liu, Z., Bruwer, M.J., MacGregor, J., Rathore, S., Reed, D., Champagne, M. (2011). Scale-up of a pharmaceutical roller compaction process using a joint-y partial least squares model. Industrial & Engineering Chemistry Research. 2011, 50, 10696-10706.