AI-based exploration of crystal spaces to accelerate drug discovery


Motivation. The commercial anti-viral drug Ritonavir has several stable crystalline structures and not all of them have desired properties. With the drug already in the market, a new and more stable crystalline structure was unexpectedly produced. The two polymorphic structures have the same chemical composition, but differ in molecule assembly in the solid state. The geometric arrangement of molecules in the new crystalline structure made it more stable and less soluble, modifying significantly bio-availability of drugs. Hence thousands of patients took much smaller than expected doses, which was a big scandal in the pharmaceutical industry.

Success rates for drug development are very low and a tenth of the 5 to 10 thousand drug candidates qualifies for test in humans. Drug developments from initial discovery to marketplace last a decade on average. The ongoing pandemic shows that drug discovery needs to be continuously and substantially accelerated. This acceleration can be based on justified methods to classify crystal structures in a new continuous way [AK]. The paper [NS] warns about other potential Ritonavir situations and how computational methods can assist minimising risk.

State-of-the-art. Crystal Structure Prediction (CSP) aims to identify putative crystal packing for a given molecule. CSP is becoming integral part of the functional materials design workflow [P], including drug development [NR]. Most CSP methods rely on a random search in the infinite space of all possible crystal packings.

Prof Sally Price FRS (UCL) has described the state-of-the-art in CSP as `embarrassment of over-prediction', because modern CSP tools output too many (thousands or millions) computationally generated putative crystals, without similarity relations, while there are almost always only less than 10 experimentally accessible crystals.

Problem statement. Simulated crystals are obtained by an iterative optimisation as approximate local minima of a potential energy that determines the stability of the predicted crystals. We need a reliable way to establish structural similarities between predicted structures and to reveal energy barriers for accessing experimental structures or phase transformations.

The aim of this PhD project is to design efficient algorithms based on new crystal invariants [AK, MK] to guide a search in a space of crystalline drugs. Establishing reliable similarities of crystal structures will inform us about potentially synthetically accessible drug structures.

Due to a recent change in UKRI policy, this is now available for Home, EU or international students to apply. However, please be aware there is a limit on the number of international students we can appoint to these studentships per year.

An ideal candidate should have at least a 2:1 degree (or equivalent) in Computer Science or Mathematics or a computational area of Physical Sciences, e.g. Computational Chemistry. The preferred programming languages are C++ and/or Python. Excellent communication skills are essential to work in a large team and to collaborate with industry partners.

The other co-supervisors are the Materials Innovation Factory: Director Prof Andy Cooper FRS:, and Dr Angeles Pulido: from the Cambridge Crystallographic Data Centre:

For any enquiries please contact: Dr Viktor Zamaraev () and Dr Vitaliy Kurlin ()

To apply for this opportunity, please visit:


Open to students worldwide

Funding information

Funded studentship

This studentship is funded by the EPSRC DTP within the AI and Future Digital Health Doctoral Network and is offered for 3.5years in total. It provides full tuition fees and a stipend of approx. £15,609 tax free per year for living costs. The stipend costs quoted are for students starting from 1st October 2021 and will rise slightly each year with inflation. The funding for this studentship comes with a budget for research and training expenses of £1000 per year, and for those that are eligible, a disabled students allowance to cover the costs of any additional support required.



[AK] O.Anosova, V.Kurlin. An isometry classification of periodic point sets.
[MK] M.Mosca, V.Kurlin. Voronoi-based similarity distances between arbitrary crystal lattices. Crystal Research and Technology, 1900197, 2020.
[NR] J.Nyman, S.M.Rutzel-Edens. Crystal Structure Prediction is changing from basic science to applied technology. Faraday Discussions 2018, 211, 459-476.
[NS] M.A.Neumann, J. van de Streek. How many ritonavir cases are still out there? Faraday Discussions 2018, 211, 441-458.
[P] A.Pulido et al. Functional materials discovery using energy-structure-function map. Nature 2017, 543, 657-664.