Abstract Rational combination therapies are urgently needed for aggressive KRAS-mutant cancers such as pancreatic ductal adenocarcinoma (PDAC), which has a 5-year survival rate below 10%. VINI is a multimodal in silico platform that integrates AI, quantum-informed algorithms, pathway modeling, and structural data to accelerate the discovery of effective drug combinations. Its foundation is based on KEGG cancer pathways, with gene expression and mutation data from the Cancer Cell Line Encyclopedia (CCLE), molecular structures from PubChem and DrugBank, three-dimensional protein structures from RCSB PDB or AlphaFold predictions, and protein sequences from UniProt and DrugBank (for monoclonal antibodies), capturing disease-specific biology. VINI uses AI and semi-empirical quantum-informed virtual screening (planned for implementation in the upcoming Horizon project) to predict intracellular drug efficacy and multi-drug synergy. Classical computational chemistry tools, including Rosetta, AutoDock Vina, UCSF Chimera, and Scripps MGLTools, are used by VINI to generate high-quality predictions, with high-performance computing enabling rapid evaluation. Proof-of-concept studies demonstrated VINI’s ability to predict novel triple-drug combinations targeting ALK, BCL-2, mTOR, DNA repair, and androgen pathways in hormone-sensitive prostate cancer, achieving 79.3% agreement with clinical outcomes across 16 cancer types and 100% for DU-145 and PC3 lines. VINI has also been applied to SARS-CoV-2 drug combinations, illustrating its broad applicability. In the upcoming EU Horizon project, VINI will be instrumental in identifying effective three-drug combinations against KRAS-mutant PDAC: (1) novel KRAS inhibitors developed at the University of Toronto, (2) DNMT1 inhibitors discovered at RBI, and (3) targeted monoclonal antibodies against PDAC hallmarks, also identified at RBI using VINI. Through this multinational consortium - including the University of Toronto, Lund University, Fraunhofer, EPFL, and seven other leading institutions - VINI will integrate AI, structural modeling, and computational chemistry to provide actionable insights for treatment-resistant cancers and guide future preclinical validation. Citation Format: Drasko Tomic. VINI: A multimodal in silico platform for discovering rational drug combinations in KRAS-mutant pancreatic cancer [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2026; Part 1 (Regular Abstracts); 2026 Apr 17-22; San Diego, CA. Philadelphia (PA): AACR; Cancer Res 2026;86(7 Suppl):Abstract nr 977.

Background Traditional publishing models, open access and major publishers, cannot adequately address the key challenges of academic publishing today: Speed of peer review, recognition of work and incentive mechanisms, transparency and thrust of the system. Methods To address these challenges, the authors propose Democratisation of Academic Publishing (DAP) platform, which is based on the novel HashNET DLT platform. The DAP introduces several innovative components: tracking the activities of all participants in the peer review process using blockchain and smart contracts, the introduction of the Scholarly Wallet for holding reputation (non-fungible) and reward (fungible) tokens, the use of the Scholarly Wallet as the main interface to the DAP platform, the Virtual Editor that enables automatic discovery of the research area and invitation of reviewers, and finally the global database of evaluated reviewers, ranked by the quality of their previous work. Results The DAP platform is in the development phase, with the design and functionalities of all modules defined. An exception is the central component of DAP, the Scholarly Wallet module, whose first prototype has already been created, tested and published. The implementation of DAP is planned for the next phase of the HorizonEurope TruBlo project and other research initiatives. The DAP platform will be connected to the publishing ecosystem: 1) as a backend system (distributed blockchain database) for existing publishing platforms and 2) as a standalone publishing platform with its own API interface. Conclusions The authors believe that DAP has the potential to significantly improve academic peer review and knowledge dissemination. It is expected that the use of blockchain technology, the fast HashNET consensus platform and tokens for reward (fungible) and reputation/ranking (non-fungible) will lead to a more efficient and transparent way of rewarding all participants in the peer review process and ultimately advance scientific research.

To address the challenge of finding new combination therapies against castration-sensitive prostate cancer, we introduce Vini, a computational tool that predicts the efficacy of drug combinations at the intracellular level by integrating data from the KEGG, DrugBank, Pubchem, Protein Data Bank, Uniprot, NCI-60 and COSMIC databases. Vini is a computational tool that predicts the efficacy of drugs and their combinations at the intracellular level. It addresses the problem comprehensively by considering all known target genes, proteins and small molecules and their mutual interactions involved in the onset and development of cancer. The results obtained point to new, previously unexplored combination therapies that could theoretically be promising candidates for the treatment of castration-sensitive prostate cancer and could prevent the inevitable progression of the cancer to the incurable castration-resistant stage. Furthermore, after analyzing the obtained triple combinations of drugs and their targets, the most common targets became clear: ALK, BCL-2, mTOR, DNA and androgen axis. These results may help to define future therapies against castration-sensitive prostate cancer. The use of the Vini computer model to explore therapeutic combinations represents an innovative approach in the search for effective treatments for castration-sensitive prostate cancer, which, if clinically validated, could potentially lead to new breakthrough therapies.

Background: Traditional publishing models, open access and major publishers, cannot adequately address the key challenges of academic publishing today: Speed of peer review, recognition of work and incentive mechanisms, transparency and thrust of the system. Methods: To address these challenges, the authors propose Decentralised Academic Publishing (DAP), which is based on the novel HashNET DLT platform. The DAP introduces several innovative components: tracking the activities of all participants in the peer review process using blockchain and smart contracts, the introduction of the Scholarly Wallet for holding reputation (non-fungible) and reward (fungible) tokens, the use of the Scholarly Wallet as the main interface to the DAP platform, the Virtual Editor that enables automatic discovery of the research area and invitation of reviewers, and finally the global database of evaluated reviewers, ranked by the quality of their previous work. Results: The DAP platform is in the development phase, with the design and functionalities of all modules defined. An exception is the central component of DAP, the Scholarly Wallet module, whose first prototype has already been created, tested and published. The implementation of DAP is planned for the next phase of the HorizonEurope TruBlo project and other research initiatives. The DAP platform will be connected to the publishing ecosystem: 1) as a backend system (distributed blockchain database) for existing publishing platforms and 2) as a standalone publishing platform with its own API interface. Conclusions: The authors believe that DAP has the potential to significantly improve academic peer review and knowledge dissemination. It is expected that the use of blockchain technology, the fast HashNET consensus platform and tokens for reward (fungible) and reputation/ranking (non-fungible) will lead to a more efficient and transparent way of rewarding all participants in the peer review process and ultimately advance scientific research.

Spike glycoprotein is essential for the reproduction of the SARS-CoV-2 virus, and its inhibition using already approved antiviral drugs may open new avenues for treatment of patients with the COVID-19 disease. Because of that we analyzed the inhibition of SARS-CoV-2 spike glycoprotein with FDA-approved antiviral drugs and their double and triple combinations. We used the VINI in silico model of cancer to perform this virtual drug screening, showing HIV drugs to be the most effective. Besides, the combination of cobicistat-abacavir-rilpivirine HIV drugs demonstrated the highest in silico efficacy of inhibiting SARS-CoV-2 spike glycoprotein. Therefore, a clinical trial of cobicistat-abacavir-rilpivirine on a limited number of COVID-19 patients in moderately severe and severe condition is warranted.

3D garment printing is a technology that is experiencing rapid development, and more and more garments are being printed on 3D printers. Advanced processes such as Selective Laser Sintering and PolyJet 3D printing are used. However, 3D printing of clothes still experiences some disadvantages, of which the biggest is the slow 3D printing speed and usually a small printing area of 3D printers. The novel, large multi-head printers can significantly increase the printing speed, but such printers require an additional step to prepare given 3D models that increase. However, pre-processing time grows exponentially with the number of printer heads, therefore computer resources that exceed the capabilities of a single workstation are required to prepare a printout. This implies the use of HPC resources, and because of its flexibility, lower cost and ease of use, HPC Cloud is optimal platform for such jobs. We have shown that the structure of these jobs fits perfectly in the HPC Cloud environment.

Virtual drug screening is one of the most widely used approaches for finding new drugs candidates. The process consists in selecting one or more chemical compounds with the highest binding free energy to target proteins. Given that the empirical space of chemical compounds is extremely large and estimated to has over 50 millions of them, finding the most effective drug is computationally challenging. Furthermore, the vast majority of proteins still lack the experimentally obtained 3D structures, making it hard to accurately calculate their binding free energies with chemical compounds. With this in mind, the aim of our study was to investigate the accuracy of the Autodock Vina program in a virtual drug screening on a set of proteins that do not have experimentally determined structures. To do this, we performed a virtual drug screening with the Autodock Vina on a large set of drug-kinase pairs taken from the IDG-Dream Drug-Kinase Binding Prediction Challenge. The results obtained show that the Autodock Vina can be used effectively in such unstructured environments.

Despite the ongoing development of new targeted cancer drugs, the survival rate of patients with aggressive forms of cancer like lung and pancreatic cancer is still poor. The main reason is the ability of cancer to develop resistance against cancer drugs. One strategy to overcome this resistance is to use cancer therapies with several drugs administered at the same time. This can increase our chances to kill cancer cells before they develop resistance. In order to investigate the effectiveness of such therapies, we let the in silico model of cancer Vini to calculate the most effective 2-drug therapies against non-small cell lung (NSCLC), small cell lung cancer (SCLC), and pancreatic cancer. Vini calculated the combination of vinorelbine with paclitaxel as the most effective against NSCLC, the combination of everolimus with doxorubicin as the most effective against SCLC, and the combination of everolimus with paclitaxel as the most effective against pancreatic cancer. As the existing clinical studies confirm Vini’s calculations, it is justified to let Vini search for the combined cancer therapies with even more drugs. In order to further increase their effectiveness, the next research step will be the personalization of such therapies.

Cancer is a system with thousands of genes and proteins with the complex interactions between them. By examining the cancer drug activity on only part of this system, we do not know in which direction the whole system will evolve, and whether therapy will be useful or not. This is one of the main reasons why cancer therapies still do not meet our expectations. In order to find more effective anticancer therapies, it is important to consider the impact of drugs on the entire cancer system. The second largest eigenvalue plays a key role in complex systems optimization. The algorithms minimizing the second largest eigenvalue of graphs have been already used to speed up processes in computer networks and differential cryptanalysis. Based on the aforementioned, it could be assumed that maximizing the second highest eigenvalue could slow down the processes in metabolic networks that describe processes in cancer. To verify our hypothesis, we have built the in silico model of cancer Vini and run it on a supercomputer. Vini transformed the metabolic pathways of cancer from Kyoto Encyclopedia of Genes and Genomes into the binding energy matrices representing binding energies between the genes and proteins on one side and drugs being investigated on another side. Some matrix elements also represent interactions between proteins and genes. Then, it calculated the second largest eigenvalues of these matrices. In the end, we compared the calculated results against the existing in vitro and in vivo experimental results. The calculated efficacy of cancer drugs was confirmed in 79.31% of in vivo experimental cases, and in 92.30% of in vitro experimental cases. These results show that the second largest eigenvalue plays an important role in metabolic cancer networks and that the Vini model can be an effective aid in finding more effective cancer therapies.

There is a mounting evidence that certain herbal compounds express strong activity against various types of cancer cells. Curcumin from Curcuma longa, resveratrol from Vitis vinifera, and artemisinin from Artemisia annua are few examples from the longer list of more than fifty anticancer herbs known. This is mostly due to the extreme complexity of cancer processes that an exact mechanism of anticancer activity of these compounds is still unknown. However, with the rise of powerful supercomputers and the advanced in silico models of cancer, our chances to understand these processes increased, and now we are in place to deliver more effective therapies than ever before. KEGG (Kyoto Encyclopedia of Genes and Genomes) cancer pathways are example of such in silico models. From KEGG cancer pathways, we were able to identify sixty-two oncogene proteins involved in the development, proliferation, angiogenesis, metastasis and resistance of cancer. From them, we selected those having the largest number of hits in Pubmed database related to each of these cancer hallmarks, and use supercomputer for docking simulations between them and certain anticancer herbal compounds. Significant activity of anticancer herbal compounds against oncogene proteins was found. We hope our research will setup pathways towards more effective therapies.

Despite the various hardware and software improvements in Cloud architecture, there still exists the huge performance gap between the commodity supercomputers and Cloud when running HPC communication intensive applications. In order to find what is preventing them to better scale on Cloud, we evaluated HPL and NAMD benchmarks on HPE Openstack testbed, and NAMD benchmarks on supercomputer located at Rijeka University Supercomputing Center. Our results revealed two major bottlenecks: the throughput of the interconnect, and Cloud orchestration layer, among other responsible for the management of the communication between Cloud instances. We investigated the influence of jittering, but did not find the significant influence on performance. Our conclusion is that by solely increasing the interconnect throughput, one will not improve the scalability of HPC communication intensive HPC applications in Cloud. This is also backed up with NAMD performed at HP Labs, and with HPL benchmark performed at San Diego Supercomputing Center. We propose two possible scenarios of scalability improvements. One with distributed model of Cloud Orchestration layer; another with bare metal containers. Efficient load balancing remains the must if we want to see HPC applications scaling over many million Cloud cores. For this, we propose novel SLEM based load balancing strategy.

Bacterial biofilms are highly complex structures. We are more and more recognizing that bacterial biofilms are predominant forms of the bacterial existence against the planktonic one. Under certain circumstances, bacteria starts to build biofilm and forms 3D structure, so called extracellular matrix. Hulled within this matrix, bacteria becomes more prone to host defense mechanisms and most antibiotics, thus expressing considerably higher virulence and antibiotic resistance than its planktonic form. Because of their importance, numerous researchers investigated bacterial biofilms in the last decades, using numerous methods, like electron microscopy, mass spectroscopy and nuclear magnetic resonance. However, neither of these methods is able to reveal an exact structure of extracellular matrix. Exploring dynamics of extracellular matrix is even more complex, and out of the reach for known analysis methods. For these reasons, there is a need for more effective method, and this could be computer driven simulation. In order to check if it could be a method of choice, we estimated the computational resources needed to simulate the bacterial biofilm. We found that possibility of performing this simulation in the reasonable time on fastest supercomputers today does not exists, and will not be available until at least 2028. For this reason, we explored possibilities of running NAMD based bacterial biofilms simulations on Cloud, and landed with the same conclusion. Besides, we found that for both approaches NAMD has to extend its scalability from about current 500.000 cores to many millions of cores in the future.

Healthy and stable state of complex biological systems lies in the narrow region between chaos and frozen state. The cause of chaos is often small initial change in starting conditions, quickly spreading and leading to system wide changes. For example, few microbial pathogens can quickly lead the host to the serious illness and even death. By the definition, an attractor in a certain dynamical system is a set of physical properties toward which this system tends to evolve, regardless of the starting conditions. It is not easy to control chaos and move system from attractor back to the stable state. This is even more difficult in case of hyperchaos with more attractors. Existing control strategies are passive; modify control parameters and wait until system lends in a future, hopefully stable state. At there is a clear need for more efficient control strategy, we propose the optimization of the matrix representing an evolution operator of the system under the control. Matrix elements are system state functions, and function variables are control parameters. Eventually, optimizing this matrix with respect to some goals can help to drive system back to stability. On the case of Samoyed dog lady attacked by methicillin-resistant and multi-species bacterial pathogens, we were able to identify the chaotic system with at least three attractors; two of them leading the system into the illness, and one driving it back to the healthy state. Consecutively, we established the matrix of its evolution operator and discussed some ways of the matrix optimization, in a hope our approach might help to develop therapies that are more efficient. Not of the less importance, we hope our experience with integrative therapy including standard therapeutics, herbal extracts and homeopathic remedies could help a clinician confronted with a similar case.

High Performance Linpack ( HPL) is an industry standard benchmark used in measuring the computational power of High Performance Clusters. In contrary to HPC clusters consisting of equal computational nodes, running HPL on heterogeneous HPC clusters, built up of a computing nodes with different computational power, showed in most cases poor efficiency. In such type of clusters, efficiency of HPL further decreases if the speed of interconnect links between computing nodes is different. In order to improve HPL efficiency on such a clusters, one needs to optimally balance HPL workload on computing nodes accordingly to their computational power, and at the same time, take into the consideration the speed of communication links between them. Our thesis is that the problem of efficiently running HPL on heterogeneous HPC cluster is solvable, and that one can formulate it as a Semidefinite Optimization of Second Eigenvalue in Magnitude (SLEM) matrix describing data-flow of HPL in a cluster. In order to test a validity of such an approach, we run a series of HPL benchmarks on Isabella HPC cluster, both optimized respective to SLEM and non-optimized. By comparing results obtained with SLEM optimization of HPL against non-optimized HPL, we were able to identify a huge improvement in HPL efficiency when using SLEM. Moreover, by taking into the consideration memory sizes of computational nodes, we were able to improve SLEM optimization of HPL further.

In the last years a strong interest of the HPC (High Performance Computing) community raised towards cloud computing. There are many apparent benefits of doing HPC in a cloud, the most important of them being better utilization of computational resources, efficient charge back of used resources and applications, on-demand and dynamic reallocation of computational resources between users and HPC applications, and automatic bursting of additional resources when needed. Surprisingly, the amount of HPC cloud solutions related to standard solutions is still negligible. Some of the reasons for the current situation are evident, some not so. For example, traditional HPC vendors are still trying to exploit their current investment as much as possible, thus favoring traditional way of doing HPC. The next, although virtualization techniques are developing in ever increasing rate, there are still open questions, like scaling of HPC applications in virtual environment, collaboration between physical and virtual resources, and support of the standard HPC tools for virtualized environments. At last but not the least, the advent of GPU computing has raised difficult questions even in the traditional and well developed HPC segment, like scalability, development of HPC GPU based applications. In addition, because many HPC vendors are speaking about already having a fully functional HPC cloud solution, there is a need to provide answers on the following questions: Are they fulfilling virtualization promises, both physical and virtual? Which type of clouds do they support: private, public or both, i.e. hybrid clouds? How well HPC applications scale on their cloud solutions? This paper is an overview of the current HPC cloud solutions, for sure not complete and solely a view of authors, however intended to be a helpful compass for someone trying to shift from standard HPC to large computations in cloud environments.

In recent years performance of High Performance Computing Clusters took precedence over their power consumption. However, costs of energy and demand for ecologically acceptable IT solutions are higher than ever before, therefore a need for HPC clusters with acceptable power consumption becomes increasingly important. Consequently, the Green500 list, which takes into account both performance and power consumption of HPC clusters, almost reached the popularity of the Top500 list. Interestingly, the Green500 list is not an opponent to Top500 list; its core idea is to complement the Top500. Therefore, the Top500 list still serves as the basis for the Green500 list, and its numbers regarding measured HPL performance, are a basis for calculating the Green500 list. Indeed, the Green500 is the Top500 list ordered by HPL measured performance per Watt. Rmax numbers gained from High Performance Linpack benchmarks serve as performance input parameters, and total power consumed during execution of HPL on a certain HPC clusters is a power consumption parameter. The critical question remains: how to measure the consumed power correctly? This paper proposes that if it is not possible to measure the consumed power, one can still use maximum power consumption numbers rated from hardware vendors to find at least the lower bound green efficiency of HPC clusters. The main idea behind this approach is that Rmax values found on Top500 list never achieve Rpeak theoretical values, and that even most efficient HPL benchmark can never utilize computing nodes at their maximum. Furthermore by comparing MFLOPS/W results we gained with those found on Green500 list, we noted the excellent efficiency of the new HPC Isabella cluster recently powered on at University Computing Centre in Zagreb, ranking in just behind University of North Carolina KillDevil Top500 super cluster.

A trend is developing in High-Performance Computing with cluster nodes built of general purpose CPUs and GPU accelerators. The common name of these systems is CPUGPU clusters. High Performance Linpack (HPL) benchmarking of High Performance Clusters consisting of nodes with both CPUs and GPUs is still a challenging task and deserves a high attention. In order to make HPL on such clusters more efficient, a multi-layered programming model consisting of at least Message Passing Interface (MPI), Multiprocessing (MP) and Streams Programming (Streams) needs to be utilized. Besides multi-layered programming model, it is crucial to deploy a right load-balancing scheme if someone wants to run HPL efficiently on CPUGPU systems. That means, besides the highest possible utilization rate, both fast and slow processors needs to receive appropriate portion of load, in order to avoid faster resources waiting on slower to finish their jobs. Moreover, in HPC clusters on Cloud, one has to take into account not only computing nodes of different processing power, but also a communication links of different speed between nodes as well. For this reasons we propose a load balancing method based on a semidefinite optimization. We hope that this method, coupled with a multi-layered programming, can perform a HPL benchmark on CPUGPU clusters and HPC Cloud systems more efficiently than methods used today.

E-learning ecosystem based on cloud computing infrastructure constantly gains a popularity in a wide research and consumer popularity. There are multiple reasons for this, including the dynamic adaptability of clouds to the changing demands, their ability to provide resources per need basis and the support for virtualization. By additionally bringing the robust security and chargeback model in it, the cloud becomes a real next generation engine for all e-learning aspects, including database, application and web layers. However, and especially in a hybrid public cloud environments, the right allocation scheme of an e-learning content is critical if the given service level agreement has to be fulfilled. It is by far not enough to distribute content accordingly to the provided amount of processing power of computational nodes and corresponding storage systems. If the content scheduler does not take into account the throughput of the network communication links, the whole system become saturated with waiting times and the response of the cloud based e-learning system degrades. Our approach is to consider both processing power of computational nodes and communication links. The idea is to minimize SLEM, the magnitude of the second largest eigenvalue of the positive symmetric square matrix with elements representing time activities of computational nodes and communication links in parallel processing environment. It shows that this minimization problem can be recast as a semi-definite optimization problem, with ability to find an optimal load distribution of e-learning content.

In the world of digital human figures moving through the virtual space there is a little mathematical information about dynamic properties. In the aim to overcome this limitations, 3D human body virtual model consisting of 14 segments has been developed and validated against camera motion capturing of the real human model. The dynamic mass properties of this virtual model have been expressed via tensor of inertia, and model motion has been correlated to the space-time transformation of this tensor. As this tensor has 9 components, there is a huge amount of data that should be processed when doing transformations. By using evolution operators and their spectral properties, we were able to express 3D human body movements with metadata that fit into the realm of 2D computer interface and are suitable for the analysis.