DynoChem 2008 was released last week. For more details or to request a trial, see http://www.scale-up.com/. You can view related webinars at www.scale-up.com/usersarea/view.htm (login required).
Among the features of this release are an expanded model library, full compatibility with Windows Vista and Office 2007 and an array of tools to support Quality by Design (QbD).
You can now explore a factor space and assess process robustness / sensitivity by driving a DynoChem model from Excel. The impact of parameter uncertainty on predicted responses can be quantified to describe the 'response volume' (rather than response surface) within which actual process results are expected to lie. Experimental error, whether pure or systematic, is captured as part of determining uncertainty.
The graphics below show how uncertainty, expressed as a fractional relative error in the endpoint predictions for an impurity (CQA), varies across a factor space in which temperature and equivalents are varied.
Impurity detection is a challenge for measurement techniques, with a low signal to noise ratio; in this case, the measured impurity levels have typical noise levels and these are factored into the uncertainty levels indicated by the model.
Uncertainty is minimized near the points at which experiments were carried out and parameters were fitted. A feature typical of a good first principles / mechanistic model is that a small number of experiments and some replicates enable a reduction in uncertainty over a wide area of the factor space. In the above example, two additional experiments reduced uncertainty overall and broadened the region in which uncertainty is at a minimum.