Implementation of new machine learning algorithms for the optimisation of drug formulations

Correctly developing and predicting crystalline forms with specific physico-chemical properties is essential to the pharmaceutical industry.

The main challenge this industry faces is the fact that most active pharmaceutical ingredients in most drugs can interconvert into a different (usually more stable) polymorph, potentially reducing the solubility of the drug, slowing down the release of the API and affecting the pharmacokinetics, bioavailability and efficacy of the drug.

For instance, due to the complex interplay between thermodynamics and kinetics, it often happens that unexpected polymorphs emerge either in development (best case scenario) or long after the drug has been approved for market (worst case scenario).

A previously known stable form that disappears or the sudden appearance of an even more stable form can have grave consequences.

For instance, the new form may have new properties that are not suitable for the intended purpose of the drug, leading to significant economic and public health repercussions.

This ERC Proof of Concept project aims to implement new machine learning approaches that would allow to accelerate the process of predicting crystal structures by a factor of 100, thereby making it sustainable and enabling industry to investigate other crystal structures of the same drug to find the most suitable formulation (e.g. hydrates, salts, co-crystals, etc).

Beyond pharma (which is our target application for MACHINE-DRUG), polymorphism of chemical structures has significant importance across many other different industries.

For instance, the polymorphism of a pigment can generate a different colour, or the polymorphism of a chemical structure can lead to a material with significantly different properties (thermal, plastic, etc.). As such, MACHINE-DRUG is a lean, targeted project with a clear scope, but its potential applications are limitless.