Drug Discovery via Accelerated Quantum Mechanics Simulations on Spinnaker Cloud Computing

Mentored by Jens Meiler, Christian Mayr
at University Leipzig, Medical School, Institute for Drug Discovery

Virtual high throughput screening (vHTS) is an essential tool in drug discovery to select molecules for experimental testing as the space of drug-like molecules (up to 1060) is far too large for experimental screening. vHTS techniques in academia and start-up companies experience a revolution as three developments coincide: a) new companies provide large (1010) make-on-demand chemical libraries where compounds that emerge from vHTS can be ordered economically; b) innovative artificial intelligence (AI) methods developed by us and others improve vHTS results by providing better quantitative structure-activity relationship (QSAR) models; and c) deep learning protein structure prediction methods such as AlphaFold and RoseTTAFold are adapted by us and others to create protein structures suitable for vHTS.

Specifically we will develop Quantum Chemical 3D Descriptors for QSAR Modeling. Molecular descriptors used to encode the ligand such as charges and polarizabilities are typically computed as rough but rapid approximations in order to work with hundreds of thousands of ligands. Precise Quantum Mechanical (QM) methods are neglected as they are considered as too time-consuming. We propose to develop AI-accelerated QM descriptors for QSAR modeling. a) Leveraging predictive equivariant neural networks to determine electron density. b) Utilizing existing electron-density based descriptors from CDFT. c) Employing size-agnostic topological autocorrelation vectors as a descriptor framework. d) Conducting QSAR benchmark studies to assess the predictivity of existing and developed descriptors.

To complete these calculations ins a finite time, here we propose to combine these innovative technologies pioneered in the Meiler laboratory at Leipzig University within SECAI with an innovative hardware setup also built within SECAI, the newly developed Spinnaker2 at TU Dresden. It provides up to ten million computational cores. We hypothesize that Spinnaker2 can be programmed to meet the computational demands of vHTS and thus allow the screening of complete make-on-demand compound libraries for the first time.

Work environment

You will be working with both PIs to implement drug discovery algorithms on Spinnaker2. Your algorithms will be used for drug discovery. Molecules will be synthesized and tested for biological activity on a number of application projects. Current application projects include drug development for neurological diseases including Schizophrenia, Alzheimer's, and Parkinson’s, probing the fundamentals of G-protein-coupled receptor signaling, understanding cardiac arrhythmia as caused by the complex interplay of potassium channel regulation and drug interactions, multidrug resistance in cancer and bacterial cells related to multidrug transporter proteins, and the structural basis of viral infections and antibody activity, among others.

Prerequisites

To conduct this research, you should hold a very good university degree (MSc or an equivalent) in computer science, chemistry, bioinformatics or related disciplines. The candidate will need a solid background in bioinformatics or structural biology as well as proficiency in scientific programming and interest in data analysis. Prior expertise in machine learning, programmkinlanguages, and/or AI techniques will be an advantage.

Further details on the requirements and application process can be found in SECAI's announcement for open PhD positions in 2024.