KTH & AstraZeneca Partner to Revolutionise Drug Development with HPC & AI

The KTH Royal Institute of Technology and the pharmaceutical giant AstraZeneca have launched a cutting-edge collaboration to accelerate drug development using high-performance computing (HPC), quantum chemistry, and machine learning (ML). At the heart of this initiative is VeloxChem, a next-generation quantum chemistry software developed at the PDC Center for High Performance Computing at KTH.

As drug development increasingly relies on computational methods, VeloxChem represents a major leap forward. Written from scratch to exploit modern supercomputers, including GPU-enabled systems like the National Academic Infrastructure for Supercomputing in Sweden (NAISS) Dardel supercomputer system hosted at KTH, VeloxChem can handle molecular simulations on a vastly larger scale than traditional tools. Its hybrid design – using Python for user interaction and high-performance languages like C++, HIP and CUDA for intensive computations – makes it both powerful and user-friendly.

The collaboration aims to build a cloud-based environment where AstraZeneca researchers can access VeloxChem remotely to perform complex simulations and train Artificial Intelligence (AI) models. This setup will support tasks such as predicting drug reactivity, optimising synthesis pathways, and modelling molecular interactions – all critical stages in the pharmaceutical pipeline. AstraZeneca, with a strong presence in Sweden and a growing focus on AI-driven research, plans to use VeloxChem to generate large, high-quality datasets which are needed to train graph-based neural networks. These models could significantly improve predictions of molecular properties like stability, selectivity, and toxicity, cutting both time and cost in drug discovery.

With support from Vinnova’s “Advanced Digitalisation” programme for a pilot study, the project’s first milestone is a cloud-accessible prototype of VeloxChem running on the Dardel supercomputer, complete with a user-friendly interface and secure access. A workflow for calculating reactivity parameters in organic molecules will also be developed, laying the groundwork for broader applications such as electrosynthesis and crystallisation prediction.

Long-term goals include expanding the solution to other industrial sectors and enabling more secure handling of confidential data. KTH researchers believe this collaboration will also push academic software development to better align with real-world needs. This strategic partnership showcases how academia and industry can unite to harness Sweden’s national HPC infrastructure, driving innovation in drug development and beyond.