Models can be a way to swiftly screen multiple formulation candidates, assessing a single combination of components or various possible combinations, while keeping relations between input material properties, their interactions and process settings' influences in mind using robust quantitative projections. These means are designed to be a support to the experience and process knowledge of the formulator, to quantitatively de-risk the proposition of formulation candidates by fast assessment through the knowledge and relations captured in the models.

Below, a simple example case is given for direct compression, using Elegent in-house models for prediction of formulation flow behaviour, which aides in selecting the first formulation candidates, and subsequently prediction of tablet tensile strength and tabletting ejection stress in function of main compaction pressure. 

Three points are sampled at the maximum active pharmaceutical ingredient (API) concentration that is projected to result in good flowability for the blends. The microcrystalline cellulose (MCC) and di-calcium phosphate (DCP) mass fraction ratios can still be varied to assess the effect on the other two CQAs taken into account here. 

The tensile strength target of 2MPa is reached for all of these filler ratios examined, yet it is already clear that it takes a different main compaction pressure value to actually achieve this strength. A logical observation, yet the model's projections make a quantitative estimation possible. 

That this necessary compaction pressure, among others, has an effect on the ejection stress, also comes natural. It can be seen that the DCP-heavier formulation exhibits worse properties in this respect. Keeping in mind the upper bound set for allowable ejection stress to prevent tooling degradation, the predictions show it is effectively an inappropriate formulation candidate: target tensile strength is only reached at pressures where ejection stress is projected to be too high.

The only experimental need, moreover, to perform this analysis and use these predictions, is material characterisation of the powders that have not been measured yet before. In this case, this is only the API's properties relevant to the model, around 20g of material was needed for this.