Quantifying Molecular Developability to Guide Decision Making in Small-Molecule Drug Discovery

Lead supervisor: Dr Fergus Imrie

Co-supervisor: Professor Charlotte Deane

Commercial partner: Dalton Tx

Designing new small-molecule therapeutics that are bioactive and safe is incredibly challenging, and significant unmet need for effective treatments remains across many disease areas. During the drug design process, numerous decisions are made regarding the design and selection of compounds. Two critical decisions are (1) which hit molecules have the greatest potential to be developed into a lead compound, and (2) which components of the molecule should be retained or altered to improve interactions with the target protein or improve other physicochemical properties, such as solubility, metabolic stability, or safety.

This project aims to capture these considerations in a concept that we will call “developability”. Broadly speaking, developability is the propensity of a molecule (or molecular series) to yield viable drug candidates through systematic modification. This project will define the notion of “developability” and establish computational methods for quantifying the developability of both individual molecules and molecular series.

There have been several previous attempts to quantify similar notions, such as “drug-likeness”. However, a fundamental limitation of such measures is that they only seek to assess the properties of a specific molecule. Our proposed notion of developability is distinct from all existing metrics in that it will not only assess the current molecule, but instead the set of possible molecules that result from reasonable modifications to the molecule under consideration. This represents a fundamental shift in what is being assessed and has the potential to be highly impactful for decision-making in drug discovery.

One key challenge is determining the set of possible molecules that could be designed from an initial hit compound. We will explore both existing approaches and develop new methods for chemical space navigation, including machine learning techniques to propose tractable modifications. These methods will be used in combination with existing tools for assessing key properties, such as solubility, metabolic stability, hERG liability, and synthetic accessibility. We will then develop approaches to combine the characterisation of the resultant chemical space into an interpretable, rational notion of developability that will be helpful for medicinal chemists and others working in drug discovery.

Evaluation will initially be carried out retrospectively using published datasets and drug discovery projects. However, validation in collaboration with Dalton Tx, either prospectively or using industrial datasets, is also possible.

By the end of the project, we will have established a computational framework that not only provides a practical definition of developability but also delivers tools to help medicinal chemists make better-informed decisions during the early stages of drug discovery. We believe such a framework will have the potential to reduce attrition rates, accelerate hit-to-lead optimisation, and improve the efficiency of drug design workflows.

Apply using course: DPhil in Statistics