Users and their projects

Ribosomal tunnels

Call: 33rd Open Access Grant Competition; OPEN-33-14

Researcher: Michal Kolář

Institution: University of Chemistry and Technology Prague

Field: Biosciences

The image shows a cross-section of a bacterial ribosome, with the large subunit in shades of grey and the small subunit in shades of yellow. The nascent protein is dynamically displayed in blue and red on the right side inside the ribosomal tunnel.

Ribosomes are key molecular factories in cells – they produce essential proteins for life. The nascent protein chain passes through a tunnel inside the ribosome that is approximately 10 nm long. Although the mechanism of protein synthesis is the same in all organisms, ribosome tunnels can vary considerably. The shape and chemical composition of the tunnel walls influence, for example, the speed and accuracy with which proteins are synthesised. Michal Kolář and his team from the University of Chemistry and Technology in Prague use the Karolina and LUMI supercomputers for atomistic simulations to observe how proteins move inside the ribosomal tunnel. This provides detailed information about a complicated process to study experimentally. The results of the simulations allow them to understand better how ribosomes are regulated and their evolutionary history. This directly impacts the development of antibiotics and drugs for neurodegenerative diseases, in which the ribosome, its tunnel, and nascent proteins play a key role.

This research is also supported by the Czech Science Foundation (project Towards an atomic understanding of the first moments in the life of protein, 23-05557S).

Advanced speaker recognition using artificial intelligence

Call: 33rd Open Access Grant Competition; OPEN-33-6

Researcher: Lukáš Burget

Institution: Brno University of Technology

Field: Informatics

Lukáš Burget from Brno University of Technology will use the Karolina and LUMI supercomputers to develop an advanced speech recognition system. The aim is to create a technology that can cope with multiple speakers speaking simultaneously, even in noisy and acoustically challenging environments using multiple microphone inputs. The team will build on the successful use of large-scale pre-trained models such as OpenAI Whisper and focus on adapting and integrating them with complementary tools, such as speaker diarisation or sound source separation. The result will be a robust system that finds applications in research and practice - for example, in healthcare, smart home, and crisis communication.

The research is supported by the European Horizon Europe (ELOQUENCE project) and Marie Skłodowska-Curie (ESPERANTO project) programmes, as well as by national projects of the Ministry of Interior of the Czech Republic (“112” and NABOSO projects), which focus, among others, on security, AI trustworthiness and combating voice-based deepfakes.

Artificial intelligence and the search for new drugs to treat pain and prostate cancer

Call: 33rd Open Access Grant Competition; OPEN-33-54

Principal Investigator: Rafael Doležal

Institution: 2nd Faculty of Medicine, Charles University

Field: bioscience

The figure illustrates the basic principle of virtual screening of new TRMP8 receptor antagonists using molecular docking and deep neural networks (AI & GPU).

Rafael Doležal from Charles University uses artificial intelligence (AI) and supercomputers, specifically Karolina and LUMI, to streamline drug design. In this project, he focuses on identifying new antagonists for the TRPM8 receptor, which is associated with the perception of cold and diseases such as migraine, steatohepatitis, and prostate cancer. In his research, he intends to use AI for ultra-fast screening of up to 1 billion chemical compounds to find suitable candidates for new drugs. This process combines molecular docking with machine learning, allowing symbolic representations of chemical molecules to be evaluated extremely quickly. The goal is to design new drug candidates for certain types of neuropathic pain and cancer using advanced AI technologies, which will be verified experimentally in in vitro tests.

A universal approach for video understanding tasks with language grounding

Call: 33rd Open Access Grant Competition; OPEN-33-23

Researcher: Evangelos Kazakos

Institution: The Czech Institute of Informatics, Robotics and Cybernetics at CTU in Prague

Field: Informatics

Output of the GROunded Video caption gEneration (GROVE) model on an instructional video. The model outputs a video-level caption (bottom) with key noun phrases in the caption coloured and localised (grounded) in the video by temporally consistent bounding boxes (top).

Evangelos Kazakos from the Intelligent Machine Perception team at the Czech Institute of Informatics, Robotics and Cybernetics, Czech Technical University in Prague, is leveraging the LUMI supercomputer to develop a versatile video-language neural network model capable of addressing a wide range of spatio-temporal grounding tasks in long videos. The model integrates a large language model (LLM) to interpret task requirements and generate video captions, alongside a visual grounding module that identifies both the temporal span of described activities and the bounding boxes of relevant objects.

Existing models face significant challenges in spatio-temporal reasoning–particularly over long videos–an ability crucial for applications such as robotic manipulation and autonomous driving. This project aims to unify diverse video grounding tasks into a single, flexible framework, offering a more holistic solution. The outcomes could enable robots to follow natural language instructions for object manipulation and enhance communication between autonomous vehicles and human drivers, contributing to safer and more intuitive interactions.

This work is part of the ERC Advanced Grant FRONTIER (GA no. 101097822).

Laser acceleration modelling

Call: 33rd Open Access Grant Competition, OPEN-33-83

Researcher: Alexander Molodozhentsev

Institution: ELI-Beamlines

Research Area: Physics

Alexander Molodozhentsev from ELI-Beamlines will use the Karolina and LUMI supercomputers to develop new models for laser particle acceleration, aimed at generating high-quality electron beams with energies in the giga-electronvolt (GeV) range. The goal of this research is to improve methods for efficiently and compactly accelerating electrons using laser radiation, which could lead to smaller and more affordable particle accelerators. Such technology could be applied in scientific fields such as biology and medicine, including advanced imaging and cancer treatment methods. This research is part of the European  EuPRAXIA project, and the team closely collaborates with experimental teams, accelerating the transfer of results to real-world applications.

Machine learning and semi-empirical quantum chemistry

Call: 33rd Open Access Grant Competition; OPEN-33-68

Researcher: Jan Řezáč

Institution: Institute of Organic Chemistry and Biochemistry of the Czech Academy of Sciences

Field: Material Sciences

Jan Řezáč from the Czech Academy of Sciences of the Czech Republic will use the Karolina and Barbora supercomputers to develop a new method in computational chemistry that combines semiempirical quantum mechanical (SQM) computation with machine learning (ML). The result will be a tool that offers the accuracy of density functional theory (DFT)-based methods but with significantly lower computational requirements. The method's current version, PM6-ML, has already been published in the Journal of Chemical Theory and Computation.

With the support of supercomputers, the research team is now working on improving the machine learning model and expanding the training dataset. The goal is to create an even more accurate and versatile method that retains excellent scalability for large-scale computations, especially for applications in drug development. Based on open-source software, the tool under development is already freely available to the scientific community. The research is supported by the Grant Agency of the Czech Republic.