Figure 3 – Pre-clinical preformulation decision tree to identify a strategy for development
simplified example of a typical preformulation workflow.
AI in pre-formulation
Like how medicinal chemists use in silico screening and AI and ML to aid in the design of new structures, the pharmaceutical chemist has access to these tools to help with the prediction of solvate formation, propensity toward polymorphism and the likelihood of forming salt or co-crystal versions. 7 In fact, most of the pre-formulation essentials listed in Table 1 can be aided by some element of model or screening tool.
These are extremely useful tools when applied at a phase-appropriate juncture during development. AI and screening tools represent a growing part of the development toolbox and supplement the more traditional screening and manufacturing activities rather than replace them. Perhaps better put, they function to streamline those experiments applied and validate the results obtained.
From a pre-formulation perspective, challenging molecules require a suitable strategy to be in place early in the drug substance’ s life cycle. Consideration of the approaches that may be taken to deliver the substance to its clinical target( route of administration) relies on an underlying pool of data that lists characteristics requiring modification. 8
Thorough characterisation and performance evaluation of a molecule with the route of administration in mind should answer the question‘ What are the CQAs that the selected solid form should provide?’
This question is continually challenged during development. For example, the acceptable and desirable attributes for a parenteral application may well differ significantly from that required for a respiratory indication, where crystal morphology and the ability to micronise to a satisfactory size distribution often require greater consideration.
Iterative processes
There is no‘ one size fits all’ when it comes to development. Compounds should be developed on a material basis, as screening and selection should never be formulaic. The entire process must be both iterative and pragmatic when required, and having the ability to integrate the various aspects of development, especially in the early phases, is the ideal.
Given the pace of early-phase development and the challenging nature of many NCEs, having an integrated development team coordinating process chemistry, solid-state data and formulation strategy enables rapid and successful progression. If this is combined with modern in silico tools, taking the time to understand what is important from the outset, allows the construction of a risk-mitigating development plan.
The foundation stone of preformulation studies is solubility and behaviour in organic solvents and water. A well-prepared polymorphism screen is a benchmark to set in place early in development. Traditional wet methods of screening can be aided by in silico solvent screening in support of crystallisation and polymorphism studies.
These rely on molecular descriptors to organise solvents by chemotype, polar surface area, molecular mass, dipole moment and hydrogen bonding potential. The application of the Random Forest algorithm is well documented in support of this type of screening approach and, for example, has been applied to a model for predicting crystal packing of olanzapine solvates. 9
Developing salts & crystals
Salts and co-crystals are also targets of development in this area. Cocrystals are of particular interest as traditional screening methods relying on the interrogation of supramolecular synthons for the molecule and co-formers that can be considered as laborious relative to salt screening.
This is of course not always true and many successful co-crystal development candidates have been identified by those skilled in the art without the aid of in silico tools. However, it remains an area of continued development and interest.
Computational techniques based on molecular modelling, molecular descriptors, hydrogen bond propensity and ML have been developed to guide such screens. 10
These often use the Cambridge Structural Database( CSD), which contains structural information on over 1.3 million compounds. I have also used this technique to validate a traditional‘ wet’ co-crystal screen and expand upon the range of co-formers tested in search of alternatives. The hits that were predicted were then verified experimentally.
32 SPECIALITY CHEMICALS MAGAZINE ESTABLISHED 1981