ZOZNAM ÚSPEŠNÝCH PROJEKTOV VO VÝZVE 3/24:
Preparing the Slovak General Language Understanding Benchmark
AUTOR: Marek Suppa
This project aims to develop the Slovak General Language Understanding Benchmark (Slovak GLUE), modeled after bgGLUE and SuperGLEBer. It will create a suite of NLP tasks for Slovak, addressing challenges in data scarcity and linguistic complexity. Using transfer learning with multilingual models, the benchmark will provide a standardized framework for evaluating Slovak NLU systems, with the ultimate goal of enhancing the language’s NLP capabilities and advancing research for less-resourced languages in Central Europe.
Colouring properties of 4-regular graphs
AUTOR: Dušan Bernát
Some of the problems interesting for 3-regular graphs are also interesting for 4-regular graphs. Almost all properties we are interested in are NP-complete. We have extensive experience with using SAT solvers that makes it manageable for small input graphs. The aim of current work is to examine certain colouring properties of 4-regular graphs beyond the limit of current database of known such graphs (18 vertices and above). Even partial results might be valuable too.
Electronic Structure and Low energy Scattering
AUTOR: Matej Veis
This project addresses two distinct research direction from physical chemistry / AMO physics.
The first part is focused on development of coupled cluster based methods that would include excited states, enumeration of properties, methods formulated in reciprocal space. Some portion will be used for running jobs of established codes that will not be under development, e.g. ReSpect, VASP.
The second part addresses collisional studies of processes in dilute gases that are based on low energy atomic/molecular scattering theory. The scattering calculations require electronic structure/ potential energy surfaces from the first part. For ultracold bosonic gases, the system will be modeled by extended Gross-Pitaevskii equation, with which the time-dependent picture (dynamics) will be explored.
Multi-Scale Modeling and Machine Learning Approaches to Molecular Interactions: From Olfactory Receptors to Disease Pathways
AUTOR: Marek Štekláč
This project aims to investigate molecular interactions involving olfactory receptors (ORs) and small molecules associated with musky odors, as well as the interaction of NSAIDs and light-activated transition metal complexes with DNA, focusing on ROS formation. Additionally, it will explore SARS-CoV-2 inhibition and tau protein aggregation linked to Alzheimer’s disease. By integrating molecular docking, machine learning, molecular dynamics, and QM/MM methods, the project seeks to develop efficient, AI-driven models for predicting molecular interactions, accelerating drug discovery, and enhancing our understanding of key biological processes related to disease and therapeutic interventions.
Evaluation and benchmarking of LLM capabilities in text generation and classification tasks
AUTOR: Dominik Macko
Multilingual machine text generation has significantly progressed in recent years thanks to a new generation of large language models (LLMs). They had a profound impact on the area of NLP, not limited to English only anymore, although low-resource languages (such as Slovak) are still quite challenging to work with. Our objective is to research LLMs’ capabilities (incl. in-context learning and synthetic data augmentation) to tackle multilingual text generation and classification. Namely, we will focus on the following tasks: 1) personalized disinformation generation, 2) machine generated text detection, 3) textual/multimodal content assessment, 4) LLM benchmark for the Slovak language, and 5) effective fine-tuning of LLMs for the Slovak language. To perform the planned tasks, we will need to infer and fine-tune different kinds of language models, covering various sizes and architectures, and explore quantization and efficient fine-tuning techniques for different tasks. The expected results will improve our understanding of advantages and limitations of LLMs and provide several improved methods.
DFT investigation of the activity and influence of selected materials on the HER reaction in water electrolysers and redox flow batteries
AUTOR: Natália Podrojková
The proposed project is primarily aimed at computational studying of the process occurring in both water electrolyses (WE) and redox flow battery systems (RFBs) – the hydrogen evolution reaction (HER). 2 main objectives are defined: 1) Development and characterization of computational models for the HER reaction in WE and RFB – this will include construction of surface slabs of phosphides and carbon doped with metal atoms; and 2) Identification of active sites of modelled materials in HER reaction – different adsorption sites on the material surface will be optimized with calculations of adsorption energy and Gibbs free energy. Calculations will be based on density functional method (DFT) which will be carried out under Quantum ESPRESSO distribution using Plane-Wave basis set and pseudopotential (PP) method with projector augmented wave method (PAW) PPs under Plane-Wave Self-Consistent Field (PWscf) code. The proposed project is the part of national project Recovery plan – Early-stage grants: Scholarships for excellent researchers R2 (09I03-03-V04-00086 HERAQUAS) and will highly contribute to solving the 1st milestone of the project.
Electronic and molecular structure, and molecular dynamics for material and drug design
AUTOR: Lukáš Bučinský
Materials related to hydrogen storage will be studied, electronic and molecular structure of transition metal complexes will be explored and druglikeness of organic compounds will be evaluated using molecular docking. In addition, molecular dynamics simulation will be utilized to confirm the binding affinity between hydrogen and the studied materials and between a ligand (drug) and the protein active centre.
The work related to transition metal complexes will be based on DFT and CASSCF/NEVPT2 calculations. The electronic structure will be compared with experimentally derived charge density from single crystal X-ray diffraction measurements. Hydrogen storage capacity of graphene like materials, borophenes, metal organic frameworks and transition metal complexes will be calculated on smaller sized clusters and under periodic boundary conditions. Molecular dynamics of hydrogen storage will be based on ab inition molecular dynamics (AIMD), while that of ligand-protein on usual molecular mechanics (MM).
Structure of saccharides and peptides by applying quantum chemical methods.
AUTOR: Michal Hricovíni
This project deals with theoretical analysis of saccharides and peptides by applying quantum-chemical methods (QM). We will focus on biologically important carbohydrates and their derivatives, as well as peptides and peptide-carbohydrate complexes. The emphasis will be placed on the biologically important carbohydrate group – glycosaminoglycans, which are involved in a number of essential biological processes. Thus, the aim of the project is to reveal the details of the influence of structure on the interaction between carbohydrates and proteins, the influence of substituents, ions and solvents.
Relativistic imprints in ultrafast time-resolved X-ray circular dichroism (TR-XCD)
AUTOR: Torsha Moitra
Chirality is inherently present in nature, broadly associated with the lack of mirror-image superimposability of the nuclear framework. Recently, ultrafast chirality has been characterized in the atto- to few-femtosecond timescale, before the onset of nuclear motions and controlled by pure electron dynamics. In this, a pump laser generates a chiral electronic wavepacket and the response of this non-stationary state is captured using a probe pulse. A conventional technique for measuring ground state chirality is electronic circular dichroism, which we extend to its pump-probe counterpart by computing time-resolved electronic circular dichroism in the X-ray (TR-XCD) energy region. However, processes involving element-specific core orbitals accessible using X-ray pulses have added complexity in its theoretical description due to the need to incorporate scalar and spin-orbit relativistic effects. To address these challenges, we seek the CPU computing time on Devana supercomputer to simulate ultrafast chiral properties using relativistic quantum chemical approach, namely real-time time-dependent density functional theory (RT-TDDFT) in conjunction with the atomic mean field exact two- component (amfX2C) Hamiltonian which produces reference four-component results at a two-component cost. The proposed simulations will provide insights into time-dependent variation of chirality in chemically relevant systems and support the design of future pump-probe experimental setups.
Influence of defects and disorder on power conversion efficiency of hybrid perovskite structures – DFT investigation II. (continuation)
AUTOR: Kamil Tokár
Hybrid perovskites (PV) as optical absorbers in photovoltaic cells could possess a comparable theoretical photoelectric efficiency with advanced silicon technologies, while the technology for preparing such solar cells based on PVs should be much more advantageous. However, the efficiency of light conversion to photocurrent can be affected by the occurrence of various defects states in the structure of the PV film and also by the thermal dependence of optical absorption.
The aim of the project will be continuation on research of the influence of basic types of defects-vacancies on the electronic band structure of MAPbI3,(Br3,Cl3), dielectric functions and optical absorption transitions using the quantum Density Functional Theory (DFT and hybrid-DFT) and post-DFT methods.
In the next more consumption part, the project will focus on the description of the fine structure of the levels near the valence and conduction bands and the impact of thermal lattice vibrations induced by phonon modes on the PVs optical absorption edge. In the study of lattice dynamics, it is supposed to simulate lattice dynamics by using the DFT approach and electronic levels lying close to the PV’s conduction band using more precise post-DFT methods (TD-DFT, G0W0, BSE).
Simulations of key electronic and photoelectric parameters may provide modelling support to the experimental part focusing on the study of passivation of defects during hybrid-PV films growth and tuning the opto-electronic characteristics.
Design and Optimization of New Photochromic Molecules for Biomedicinal Applications
AUTOR: Martin Ošťadnický
Applicability of molecular photoswitches in vivo is a fascinating and intriguing topic, which has drawn considerable attention over past years. To meet the requirements for their use in living organisms, various strategies have been proposed, mostly revolving around the photochromic unit extension of the molecular structure, effectively shifting the absorption maxima into the VIS/IR region. Although successful, this approach has its limitations, e.g., the loss of photochromic properties, the low penetrability through skin, etc. An alternative approach is based on the effect of sonoluminescence, i.e., generating light using ultrasound, which is however of low intensity. To be effectively used inside the human body, this strategy calls for photochromic molecules with high molar absorption coefficients and adequate thermal half-lives. With the use of static quantum chemistry approaches and non-adiabatic molecular dynamics simulations, our aim is to elucidate the structure-photochemical properties, relationships and photoreaction mechanisms for selected classes of photochromic compounds, thereby enabling design and optimization of photoswitches activable by low-intensity visible light.
Study of critical airflow velocity in tunnel using Fire Dynamics Simulator
AUTOR: Lukáš Valášek
The proposed project will focus on the realization of series of large parametric studies consisting of a large number of simulations of fire in various tunnels in order to contribute to the improvement of the estimation of the critical airflow velocity in tunnels. Simulations will be realized by the parallel MPI model of FDS. The research builds on the previous results and preliminary computer simulations results carried out. Based on the literature study, we found that existing estimates of the critical airflow velocity, even those included in international standards overestimate the value of the critical velocity with respect to the data measured during experiments in the tunnels. The main goal of the research is to realise a series of computer simulations of fires for different tunnels and fire intensities to estimate critical airflow velocity.
Theoretical prediction of mechanism and kinetics of isobutanol to butenes transformations
AUTOR: Tomáš Bučko
Within this project, we shall use static and molecular dynamics based approaches to investigate mechanisms and kinetics of isobutanol transformations into butenes, which is a catalytic reaction highly debated in the context of sustainable chemistry. The research problems that will be addressed include the investigation of the effect of water on reaction mechanism and kinetics of the most important elementary steps from the reaction network identified in our previous work, and the identification of active sites on the external surfaces of zeolite ferrierite. Furthermore, adaptive training strategies for obtaining a reliable machine learning potentials to accelerate the ab initio molecular dynamics simulations of chemical reactions will be developed and used to predict rate constants with accuracy beyond the transition state theory.
Training of the Slovak T5 large model
AUTOR: Daniel Hládek
The development of large monolingual language models for the Slovak language remains largely unexplored. Current support for Slovak in multilingual models is limited or absent. Slovakia’s relatively small market size, the insufficient amount of high-quality training data, and the lack of adequate computational resources compound this issue.
This project aims to address these challenges by training a monolingual T5 model with 1 billion parameters, which, to our knowledge, would be the first of its kind for the Slovak language. We believe this model will strike a balance between the hardware requirements and the availability of training data.
The resulting model will be a foundational resource for future work, including instruction fine-tuning and data augmentation. Moreover, the model will be freely available to the public, fostering further development and research in Slovak natural language processing (NLP).
Disintegration, Dissolution, and Population Pharmacokinetic Modeling of a Pharmaceutical Finished Product
AUTOR: Stefan Horkovics-Kovats
Aim:
This project evaluates the pharmacokinetics (PK) of a parent drug and its metabolite(s) using concentration-time profiles from bioequivalence studies, integrating a disintegration-dissolution model within the PK/PD framework. The aim is to assess how disintegration and dissolution affect drug absorption and examine the impact of covariates (e.g., age, weight, sex) on PK variability in healthy volunteers.
Goals:
Implement a disintegration-dissolution model to assess dosage form effects on bioavailability.
Model absorption, distribution, non-linear metabolism, and elimination (ADME) of the parent drug.
Explore covariates influencing PK variability.
Methods:
Integrate a mechanistic disintegration-dissolution model into the PK/PD framework.
Use S-ADAPT software for PK modeling of the parent drug and metabolite(s).
Perform stepwise covariate analysis to identify key factors.
Validate the model using diagnostic and predictive tools.