The project brief included the following objectives:
- Identification of appropriate test powders and characterization of their relevant physical and chemical properties.
- Establishment of a test method to quantify material adhesion on compaction tooling over an industrially relevant range of process and environmental conditions.
- Identification of key factors affecting the amount and/or rate of powder adhesion on
- compaction tooling such as: molecular, crystal, surface, and mechanical properties of the powder, surface finish and chemistry (including coated surfaces), process conditions (e.g., pressure/stress) and environmental conditions (temperature, relative humidity)
- Establish predictive criteria for the propensity of adhesion given a set of molecular/crystal properties and process/environmental conditions.
The sticking behaviour was characterized for the following materials: Ibuprofen, Acetylsalicylic Acid (Aspirin), Acetaminophen (Paracetamol), Mannitol (Pearlitol), Sorbitol (Neosorb), Maize Starch B and Microcrystalline cellulose (Microcel) which was used as a reference non-sticking material.
In addition to standard compaction studies, two new test methods were developed: heated die and high-rate compaction. The experimental conditions for a diagnostic test were established.
The key factors controlling sticking were determined as: compaction pressure, compaction rate, temperature and relative humidity.
The team at Leicester conceived a predictive method that uses machine learning to determine the sticking probability of sticking of any existing or new chemical entity from molecular information as follows. The chemical formula together with the structure of the molecule is encoded in SMILES (Simplified Molecular Input Line Entry System), for which molecular descriptors are calculated (Mordred). Machine learning tools including linear discriminant analysis (to rank the descriptors), feature engineering (to balance the data set), principal component analysis (to determine weighting for the descriptors), and support vector machine (to classify sticking and assign probability).
The experimental data generated in the project was used to train the algorithm, together with known sticking and non-sticking materials from the literature. The sticking probability was determined for the materials published in the Handbook of Pharmaceutical Excipients and molecules proposed by the industrial partners.