Miha Moškon

The increasing availability of publicly accessible data has created new opportunities for data-driven and AI-supported water research. Such data are particularly attractive due to their transparency, cost-effectiveness, and ability to support reproducible scientific studies. Moreover, publicly available datasets offer strong potential for methodological transferability across geographical locations, enabling comparative analyses, model generalization, and knowledge sharing across regions with diverse environmental and socio-economic conditions. In this work, we explore both the benefits and limitations of conducting AI-driven research using publicly available datasets, emphasizing their role in accelerating innovation and enabling broader participation in scientific inquiry. We specifically focus on water-related data, including environmental monitoring datasets that capture various aspects of water quality, availability, and ecosystem dynamics. These datasets typically rely on measurements such as chemical composition, temperature, flow rates, and pollutant levels, often collected through governmental or international monitoring initiatives. We introduce WaterDataHub 1.0, a curated discovery resource for publicly available data sources relevant to water research and artificial intelligence applications. The principal contribution of this work is WaterDataHub 1.0 itself: a centralized, cross-domain, and AI-oriented discovery resource that organizes fragmented water-related data sources within a common framework and facilitates their identification and reuse for data-driven research.

This study proposes a data acquisition, preprocessing and machine learning pipeline to enhance the accuracy of travel time estimates for the public bus network in Ljubljana, Slovenia. Existing prediction services often rely on vehicle location and may insufficiently capture dynamic congestion and contextual conditions. Using data from five Ljubljana Passenger Transport (LPP) routes, we evaluate LPP and Google Routes estimates against observed travel times. We develop a multisource model combining real-time variables with contextual predictors, including bus line, time of day, day type, and weather. The approach demonstrates how multisource integration of local transit feeds, external routing, and environmental indicators can improve prediction accuracy, reliability, and overall performance.

Circadian rhythm disruptions are increasingly recognized in disorders such as obstructive sleep apnea (OSA), yet analysis of 24 h gene expression patterns remains challenging due to the lack of reliable reference genes for normalization. Even commonly used housekeeping genes may exhibit circadian oscillations, which can confound rhythmic gene expression analyses and hinder biomarker identification. To address this limitation, we evaluated the gene expression stability of 11 commonly used housekeeping genes in blood collected every 6 h over 24 h period from 40 adults with varying OSA severity and controls. Stability ranking by analytical tools RefFinder and EndoGeneAnalyzer identified ACTB (β-actin) and RPL13A (ribosomal protein L13a) as the most consistent reference genes, with minimal intra- and inter-individual variability across sampling times and disease groups. Their suitability was assessed by personalized cosinor analysis of core clock genes (BMAL1, PER2, CRY1), demonstrating that appropriate normalization enables detection of circadian oscillations in clinical samples. Using the optimal normalization, CosinorPy analysis of the core clock genes revealed significant circadian oscillations of at least one clock gene in the studied participants. These findings establish ACTB and RPL13A as robust reference genes for blood-based circadian studies of OSA and provide an important methodological framework for future circadian biomarker research.

Circadian rhythm disruptions are increasingly recognized in disorders such as obstructive sleep apnea (OSA), yet analysis of 24-hour gene expression patterns remains challenging due to the lack of reliable reference genes for normalization. Even commonly used housekeeping genes may exhibit circadian oscillations, which can confound rhythmic gene expression analyses and hinder biomarker identification. To address this limitation, we evaluated the gene expression stability of 11 commonly used housekeeping genes in blood collected every 6 hours over 24-hour period from 40 adults with varying OSA severity and controls. Stability ranking by analytical tools RefFinder and EndoGeneAnalyzer identified ACTB (β-actin) and RPL13A (ribosomal protein L13a) as the most consistent reference genes, with minimal intra- and inter-individual variability across sampling times and disease groups. Their suitability was assessed by personalized cosinor analysis of core clock genes (BMAL1, PER2, CRY1), demonstrating that appropriate normalization enables detection of circadian oscillations in clinical samples. Using the optimal normalization, CosinorPy analysis of the core clock genes revealed significant circadian oscillations of at least one clock gene in the studied participants. These findings establish ACTB and RPL13A as robust reference genes for blood-based circadian studies of OSA and provide an important methodological framework for future circadian biomarker research.

Obstructive sleep apnea (OSA), one of the most common breathing disorders during sleep, is associated with circadian rhythm disruption, yet circadian genes expressed in blood remain largely unexplored as biomarkers. Reliable circadian gene expression analysis requires stable reference genes for qRT-PCR normalization, but no clear guidance exists for blood-based circadian studies in OSA. We evaluated the expression of 11 candidate reference genes ( GAPDH, ACTB, RPL13A, PPIB, TBP, HPRT1, PPIA, SDHA, TUBB2A, UBC ) in whole blood collected every 6 hours over 24 hours from 40 participants who underwent overnight respiratory polygraphy. Gene stability was assessed with RefFinder and EndoGeneAnalyzer across OSA severity levels, intra-individual variability, and time-point variability. ACTB and RPL13A consistently emerged as the most stable reference genes under all conditions. Their robustness was confirmed by personalized cosinor analysis of core clock genes ( BMAL1, PER2, CRY1 ), showing that reliable normalization enables detection of circadian oscillations in clinical samples. While population-level analysis revealed no significant rhythmicity, individual profiles showed oscillations in 25 participants, independent of OSA severity. These findings identify ACTB and RPL13A as strong candidates for circadian transcriptomic studies in blood and provide a methodological foundation for biomarker discovery in sleep medicine, molecular medicine, and translational chronobiology.

Urban environments are increasingly challenged by rapid urbanisation and climate change, demanding strategic responses that are both adaptable and sensitive to local context. Typological classification offers a structured approach to understanding diverse urban contexts, enabling targeted interventions that support climate neutrality and livability. While global pressures are shared, their impacts differ widely across cities, highlighting the need for context-aware urban analytics to guide effective transformation. This paper presents a methodological perspective on a computational framework and workflow based on open source data, designed to support the classification and optimisation of urban environments across different urban contexts; it explores the framework’s potential and limitations, grounded in a review of relevant literature and available datasets. We propose a workflow encompassing four main steps: (1) classifying urban environments based on quantifiable characteristics, (2) identifying key performance indicators (KPIs) differentiated by urban typology, (3) proposing interventions to optimise urban environments according to underlying typological classification, and (4) validating the proposed solutions in simulated environments. The framework prioritises open data sources provided by public authorities as well as open science and citizen science initiatives. A more streamlined integration of data is proposed, facilitating both the classification and assessment of urban environments aligned with their primary typological designation.

This paper concerns non-overlapping codes, block codes motivated by synchronisation and DNA-based storage applications. Most existing constructions of these codes do not account for the restrictions posed by the physical properties of communication channels. If undesired sequences are not avoided, the system using the encoding may start behaving incorrectly. Hence, we aim to characterise all non-overlapping codes satisfying two additional constraints. For the first constraint, where approximately half of the letters in each word are positive, we derive necessary and sufficient conditions for the code’s non-expandability and improve known bounds on its maximum size. We also determine exact values for the maximum sizes of polarity-balanced non-overlapping codes having small block and alphabet sizes. For the other constraint, where long sequences of consecutive equal symbols lead to undesired behaviour, we derive bounds and constructions of constrained non-overlapping codes. Moreover, we provide constructions of non-overlapping codes that satisfy both constraints and analyse the sizes of the obtained codes.

Constraint-based modelling and genome-scale metabolic models (GEMs) have been used extensively to analyze omics data, providing a mechanistic perspective on complex metabolic systems and networks