PhD position in Developing Machine Learned Force Fields to Predict the Stability of Molecular Crystals

The Faculty of Science, Technology and Medicine (FSTM) contributes multidisciplinary expertise in the fields of Mathematics, Physics, Engineering, Computer Science, Life Sciences and Medicine. Through its dual mission of teaching and research, the FSTM seeks to generate and disseminate knowledge and train new generations of responsible citizens, in order to better understand, explain and advance society and environment we live in.

Introduction

A method that can accurately and efficiently predict the stable phase of molecular crystals will greatly aid the development of new pharmaceutical drugs 1. If a drug is manufactured in a metastable phase that later converts to the stable phase, it can render the drug insoluble and ineffective. For instance, ritovanir, an HIV treatment, had to be recalled from the market after it began to convert to a more stable phase. The mistake cost approximately $250 million.

There are several reasons why predicting the stable phase of a molecular crystal is challenging. The thermodynamic stability of a structure depends on the minute balance between intramolecular and intermolecular forces, in particular van der Waals interactions, Pauli repulsion, and hydrogen bonding, all of which necessitate a quantum mechanical treatment. Additionally, the free energy of the molecular crystal includes nonnegligible anharmonic vibrational effects. A state-of-the-art density functional theory (DFT) method developed in our group is the leading method for accurately predicting the stability of molecular crystals 2. However, it is quite computationally demanding. We want to create next-generation machine learned force fields that can produce equivalently accurate results at a fraction of the cost, a game changer for drug development.

The candidate will learn how crystal structure prediction is currently done in the industry and target where machine-learning will best speed up the crystal structure prediction process. They will then develop training data sets using advanced DFT methods, and design and test machine learned force field methodologies, including kernel ridge regression and graph neural networks, that can be integrated into current crystal structure prediction workflows.

The PhD position is in the Theoretical Chemical Physics (TCP) group, led by Prof. Alexandre Tkatchenko in the Physics and Materials Science Department (DPhyMS) at the University of Luxembourg. This PhD position belongs to the PHYMOL: A Marie Skłodowska–Curie Actions Doctoral Network (MSCA DN) on Intermolecular Interactions. As such, the PhD candidate will enjoy a broad collaboration with top-notch research groups. The candidate will also be co- supervised by Dr. Marcus Neumann, founder of Avant-garde Materials Simulation Deutschland GmbH, in Freiburg, Germany, the developer of the most accurate crystal structure prediction software in the pharmaceutical industry.

The yearly gross salary for every PhD at the UL is EUR 39953 (full time)

Applications should include:

Early application is highly encouraged, as the applications will be processed upon reception. Please apply ONLINE formally through the HR system. Applications by email will not be considered.

The University of Luxembourg embraces inclusion and diversity as key values. We are fully committed to removing any discriminatory barrier related to gender, and not only, in recruitment and career progression of our staff.

Interested candidates should contact: Prof. Dr. Alexandre Tkatchenko ([email protected])

References:

1What is Crystal Structure Prediction? And why is it so difficult?​ ​- The Cambridge Crystallographic Data Centre (CCDC)

2 Johannes Hoja and Hsin-Yu Ko and Marcus A. Neumann and Roberto Car and Robert A. DiStasio and Alexandre Tkatchenko "Reliable and practical computational description of molecular crystal polymorphs" Sci. Adv., 5 , eaau3338 (2019) provides an excellent overview of the general problem and the method, DFT+MBD, that the doctoral candidate will use.