1. Cheminformatic analysis of protein surfaces provides binding site insights andinforms identification strategiesAndrej Milisavljević, Jure Pražnikar, Urban Bren, Marko Jukič, 2025, original scientific article Abstract: Aims: Understanding protein–ligand binding site behavior is central to structure-based drug design. Weanalyzed amino acid composition and interactions in protein–ligand small-molecule binding sites anddeveloped a novel method for binding site prediction.Materials and methods: We analyzed the PDBBind+ database, which contains the largest protein–ligand binding site dataset known to us, using existing cheminformatics packages and in-house code.We used the resulting data to train a binding site prediction model.Results: Within solvent-accessible binding regions, tryptophan, phenylalanine, tyrosine, methionine,and glycine, were enriched. Interaction analysis revealed hydrophobic contacts as the most frequent,followed by hydrogen bonds, water-bridged hydrogen bonds, salt bridges, π–π, π–cation, and occa-sional halogen interactions. We introduced the amino acid binding site enrichment index (ABSE), tosupport small-molecule binding site detection, and developed a model that discriminates binding sitesequences from protein surface patches with 0.91 accuracy.Conclusions: This work offers interpretable composition–interaction relationships and practical tool forbinding site characterization. To facilitate application, we provide a free, open-source, fast, bindingsiteidentification tool (AABS), available at https://gitlab.com/Jukic/aabs. We anticipate that these findingsand tool will advance binding site prediction and accelerate computationally intensive drug discoverywithin medicinal chemistry.
Keywords: protein surface analysis, small-molecule binding site detection, machine learning, cheminformatics, amino acidindex, binding site, mall-molecule–protein interactions, in-silico drug design
Published in DKUM: 08.12.2025; Views: 0; Downloads: 1
 Full text (7,50 MB) |
2. |
3. Temporal and statistical insights into multivariate time series forecasting of corn outlet moisture in industrial continuous-flow drying systemsMarko Simonič, Simon Klančnik, 2025, original scientific article Abstract: Corn drying is a critical post-harvest process to ensure product quality and compliance with moisture standards. Traditional optimization approaches often overlook dynamic interactions between operational parameters and environmental factors in industrial continuous flow drying systems. This study integrates statistical analysis and deep learning to predict outlet moisture content, leveraging a dataset of 3826 observations from an operational dryer. The effects of inlet moisture, target air temperature, and material discharge interval on thermal behavior of the system were evaluated through linear regression and t-test, which provided interpretable insights into process dependencies. Three neural network architectures (LSTM, GRU, and TCN) were benchmarked for multivariate time-series forecasting of outlet corn moisture, with hyperparameters optimized using grid search to ensure fair performance comparison. Results demonstrated GRU’s superior performance in the context of absolute deviations, achieving the lowest mean absolute error (MAE = 0.304%) and competitive mean squared error (MSE = 0.304%), compared to LSTM (MAE = 0.368%, MSE = 0.291%) and TCN (MAE = 0.397%, MSE = 0.315%). While GRU excelled in average prediction accuracy, LSTM’s lower MSE highlighted its robustness against extreme deviations. The hybrid methodology bridges statistical insights for interpretability with deep learning’s dynamic predictive capabilities, offering a scalable framework for real-time process optimization. By combining traditional analytical methods (e.g., regression and t-test) with deep learning-driven forecasting, this work advances intelligent monitoring and control of industrial drying systems, enhancing process stability, ensuring compliance with moisture standards, and indirectly supporting energy efficiency by reducing over drying and enabling more consistent operation.
Keywords: advanced drying technologies, continuous flow drying, time-series forecasting, LSTM, GRU, TCN, deep learning, statistical analysis, optimization of the drying process
Published in DKUM: 03.11.2025; Views: 0; Downloads: 3
Full text (3,02 MB)
This document has many files! More... |
4. Predicting relative density of pure magnesium parts produced by laser powder bed fusion using XGBoostKristijan Šket, Snehashis Pal, Janez Gotlih, Mirko Ficko, Igor Drstvenšek, 2025, original scientific article Abstract: In this work, Laser Powder Bed Fusion (LPBF), an additive manufacturing (AM) process, was optimised to produce pure magnesium components. The focus of the presented work is on the prediction of the relative product density using the machine learning model XGBoost to improve the production process and thus the usability of the material for practical use. Experimental tests with different parameters, laser power, scanning speed and layer thickness, and fixed parameters, track overlapping and hatching distance, were analysed and resulted in relative material densities between 89.29% and 99.975%. The XGBoost model showed high predictive power, achieving an R2 test result of 0.835, a mean absolute error (MAE) of 0.728 and a root mean square error (RMSE) of 0.982. Feature importance analysis showed that the interaction of laser power and scanning speed had the largest influence on the predictions at 35.9%, followed by laser power × layer thickness at 29.0%. The individual contributions were laser power (11.8%), scanning speed (10.7%), scanning speed × layer thickness (9.0%) and layer thickness (3.6%). These results provide a data-based method for LPBF parameter settings that improve manufacturing efficiency and component performance in the aerospace, automotive and biomedical industries and identify optimal parameter regions for a high density, serving as a pre-optimisation stage.
Keywords: additive manufacturing, machine learning, XG Boost, magnesium, relative density
Published in DKUM: 03.11.2025; Views: 0; Downloads: 5
Full text (1,60 MB)
This document has many files! More... |
5. A machine vision approach to assessing steel properties through spark imagingGoran Munđar, Miha Kovačič, Uroš Župerl, 2025, original scientific article Abstract: Accurate and efficient evaluation of steel properties is crucial for modern manufacturing. This study presents a novel approach that combines spark imaging and deep learning to predict carbon content in steel. By capturing and analyzing sparks generated during grinding, the method offers a fast and cost-effective alternative to conventional testing. Using convolutional neural networks (CNNs), the proposed models demonstrate high reliability and adaptability across different steel types. Among the tested architectures, MobileNet-v2 achieved the best performance, balancing accuracy and computational efficiency. The findings highlight the potential of machine vision and artificial intelligence in non-destructive steel analysis, providing rapid and precise insights for industrial applications.
Keywords: carbon content prediction, convolutional neural networks, deep learning, machine vision, spark imaging, steel analysis
Published in DKUM: 03.11.2025; Views: 0; Downloads: 5
Full text (1,84 MB)
This document has many files! More... |
6. Automated speech analysis in depressive disorder: enhancing diagnosis and monitoringAljaž Neuberg, 2025, master's thesis Abstract: This paper investigates the automatic recognition of depression by integrating acoustic, linguistic and emotional features extracted from clinical interviews in the DAIC-WOZ dataset. A total of six classical machine learning classifiers such as Decision Tree, Random Forest, SVM, Gradient Boosting, AdaBoost and XGBoost were systematically evaluated under different class balancing methods (such as SMOTE, SMOTETomek and Random Undersampling) and feature selection strategies. The best model, a decision tree classifier with SMOTE-based balancing and a feature selection technique, achieved a weighted F1 score and accuracy of 0.78 with only eight selected features. These features included all three modalities, demonstrating the
added benefit of a multimodal approach. The results suggest that even relatively simple models, when supported by careful preprocessing and dimensionality reduction, can provide accurate and interpretable predictions. This work emphasizes the importance of feature engineering and balancing techniques in clinical machine learning tasks and lays the foundation for future research on scalable and explainable depression detection systems.
Keywords: Depression, Classification, Machine Learning, Data Balancing, Feature Selection
Published in DKUM: 03.11.2025; Views: 0; Downloads: 2
Full text (4,28 MB) |
7. Artificial intelligence-based approaches for advance care planning : a scoping reviewUmut Arioz, Matthew John Allsop, William D. Goodman, Suzanne Timmons, Kseniya Simbirtseva, Izidor Mlakar, Grega Močnik, 2025, review article Abstract: Background Advance Care Planning (ACP) empowers individuals to make informed decisions about their future healthcare. However, barriers including time constraints and a lack of clarity on professional responsibilities for ACP hinder its implementation. The application of artificial intelligence (AI) could potentially optimise elements of ACP in practice by, for example, identifying patients for whom ACP may be relevant and aiding ACP-related decision-making. However, it is unclear how applications of AI for ACP are currently being used in the delivery of palliative care. Objectives To explore the use of AI models for ACP, identifying key features that influence model performance, transparency of data used, source code availability, and generalizability. Methods A scoping review was conducted using the Arksey and O’Malley framework and the PRISMA-ScR guidelines. Electronic databases (Scopus and Web of Science (WoS)) and seven preprint servers were searched to identify published research articles and conference papers in English, German and French for the last 10Â years’ records. Our search strategy was based on terms for ACP and artificial intelligence models (including machine learning). The GRADE approach was used to assess the quality of included studies. Results Included studies (N = 41) predominantly used retrospective cohort designs and real-world electronic health record data. Most studies (n = 39) focused on identifying individuals who might benefit from ACP, while fewer studies addressed initiating ACP discussions (n = 10) or documenting and sharing ACP information (n = 8). Among AI and machine learning models, logistic regression was the most frequent analytical method (n = 15). Most models (n = 28) demonstrated good to very good performance. However, concerns remain regarding data and code availability, as many studies lacked transparency and reproducibility (n = 17 and n = 36, respectively). Conclusion Most studies report models with promising results for predicting patient outcomes and supporting decision-making, but significant challenges remain, particularly regarding data and code availability. Future research should prioritize transparency and open-source code to facilitate rigorous evaluation. There is scope to explore novel AI-based approaches to ACP, including to support processes surrounding the review and updating of ACP information.
Keywords: advance care planning, digital tools, palliative care, artificial intelligence, machine learning
Published in DKUM: 29.10.2025; Views: 0; Downloads: 5
Full text (1,52 MB) |
8. Recommender system for computer componentsAljaž Herzog, 2025, master's thesis Abstract: This thesis explores the implementation of a personalized and accurate recommender system for computer components, addressing key challenges within the e-commerce sector. A full-stack solution was developed, featuring a React-based frontend, a Flask backend, and a MongoDB database. The system integrates and evaluates four distinct recommendation algorithms: a basic model, Content-Based Filtering (CBF), Collaborative Filtering (CF), and a hybrid approach. Evaluation metrics revealed that CBF provided the highest accuracy among the individual methods. The hybrid system's performance matched that of the CBF model, primarily due to insufficient user interaction data limiting the effectiveness of the CF component. This outcome underscores the critical need for comprehensive datasets to fully leverage the power of collaborative and hybrid recommendation systems.
Keywords: recommender system, computer components, hybrid recommender system, machine learning
Published in DKUM: 23.10.2025; Views: 0; Downloads: 13
Full text (3,28 MB) |
9. Cascade hydropower plant operational dispatch control using deep reinforcement learning on a digital twin environmentErik Rot Weiss, Robert Gselman, Rudi Polner, Riko Šafarič, 2025, original scientific article Abstract: In this work, we propose the use of a reinforcement learning (RL) agent for the control of a cascade hydropower plant system. Generally, this job is handled by power plant dispatchers who manually adjust power plant electricity production to meet the changing demand set by energy traders. This work explores the more fundamental problem with the cascade hydropower plant operation of flow control for power production in a highly nonlinear setting on a data-based digital twin. Using deep deterministic policy gradient (DDPG), twin delayed DDPG (TD3), soft actor-critic (SAC), and proximal policy optimization (PPO) algorithms, we can generalize the characteristics of the system and determine the human dispatcher level of control of the entire system of eight hydropower plants on the river Drava in Slovenia. The creation of an RL agent that makes decisions similar to a human dispatcher is not only interesting in terms of control but also in terms of long-term decision-making analysis in an ever-changing energy portfolio. The specific novelty of this work is in training an RL agent on an accurate testing environment of eight real-world cascade hydropower plants on the river Drava in Slovenia and comparing the agent’s performance to human dispatchers. The results show that the RL agent’s absolute mean error of 7.64 MW is comparable to the general human dispatcher’s absolute mean error of 5.8 MW at a peak installed power of 591.95 MW.
Keywords: cascade hydropower, reinforcement learning, digital twin
Published in DKUM: 17.10.2025; Views: 0; Downloads: 5
Full text (6,94 MB) |
10. Machine learning in antiviral drug designAnja Kolarič, Marko Jukič, Urban Bren, 2026, review article Abstract: Viral infections pose a significant health threat worldwide. Due to the high mutation rates of many viruses and their reliance on host cellular machinery, the development of effective antiviral therapies is particularly difficult. As a result, only a limited number of antiviral agents is currently available. In parallel to modern vaccines, traditional antiviral drug development is both time-consuming and costly, underscoring the need for faster, more efficient approaches. In recent years, particularly since the beginning of the COVID-19 pandemic, machine learning (ML) together with broader artificial intelligence (AI), have emerged as powerful methodologies for drug discovery and offer the potential to accelerate the identification and development of antiviral agents. This review examines the application of ML in the early stages of antiviral drug discovery, with a particular focus on recent studies where ML methods have successfully identified hit compounds with experimentally demonstrated activity in biological assays. By highlighting these successful case studies, the review illustrates the growing impact of ML in advancing the discovery of urgently needed novel antivirals.
Keywords: machine learning, artificial intelligence, antiviral compounds, biological activity
Published in DKUM: 17.10.2025; Views: 0; Downloads: 5
Full text (10,88 MB) |
