Abstract: Traditional Operations Research (OR) models incorporate structured numerical inputs, such as demand parameters, costs, emissions, processing times, facility and transport mode capacities, to simulate stochastic behaviors and optimize objective-functions. There is however a vast number of OR-relevant parameters that can be derived from textual and semi-structured sources, such product reviews, social media, news, contracts, procurement documents, ESG reports, policy texts, patents, maintenance records, technical manuals and others. Under this context, the purpose of this study is to develop an artifact and parameter-centered framework for explaining how such textual inputs are transformed into model-ready OR components and incorporated into forecasting, simulation, optimization, logistics, inventory, sustainability, procurement, network analysis, and decision-support models. The main insights derived from the employed framework reveal that: (i) the OR value of textual information lies not in text analysis itself, but in its transformation into validated model-ready artifacts, such as covariates, parameters, constraints, scenarios, rules, weights, graph relations, simulation triggers, and solver inputs; (ii) different textual sources and language-processing methods can generate distinct OR artifacts that enter models through different integration mechanisms, including covariate augmentation, parameter updating, constraint generation, scenario definition, objective-function weighting, graph construction, retrieval support, and solver-code generation; (iii) the same text-derived artifact may play different roles across OR model types, for example a disruption event may update a simulation scenario, increase a lead-time parameter, remove a routing arc, or modify a supplier-risk penalty; and (iv) evaluation must extend beyond NLP accuracy to include artifact validity, parameter validity, model feasibility, mathematical consistency, solver correctness, deployment reliability, and downstream decision usefulness.

Abstract: A fully coupled three-dimensional (3D) thermo-magneto-elastic layer-wise formulation is developed for the analysis of multilayered flat and curved panels used in aerospace and aeronautical applications. The model relies on a system of coupled second-order differential equations along the thickness coordinate z, formulated in a mixed orthogonal curvilinear reference system. The governing equations combine the three-dimensional equilibrium equations with the magnetic induction divergence equation and the heat conduction equation, providing a unified multifield framework for thermo-magneto-elastic analyses. Through a suitable definition of the curvature parameters, the same formulation can be directly applied to plates, cylinders, cylindrical panels, and shells with constant radii of curvature. The governing equations are analytically solved by adopting harmonic expansions in the in-plane directions together with the exponential matrix method along the thickness coordinate. The harmonic representation naturally satisfies simply-supported boundary conditions along the panel edges. The multilayered structure is modeled according to a layer-wise strategy, where the continuity of the selected mechanical, magnetic, and thermal variables is enforced across the interfaces between adjacent layers. Different loading boundary conditions can be assigned at the external surfaces by prescribing pressure loads, magnetic potential, transverse magnetic induction, and over-temperature. The numerical investigation is divided into two stages. First, the accuracy of the proposed formulation is verified through comparisons with thermo-magneto-elastic solutions available in the literature. Then, a comprehensive set of new benchmark results is presented by considering different geometries, thickness ratios, and loading boundary conditions. Both tabulated values and through-the-thickness distributions are reported for the most significant field variables. These benchmark results provide useful reference data for the assessment and validation of future two-dimensional and three-dimensional analytical and numerical formulations devoted to coupled thermo-magneto-elastic problems.

Abstract: Background: Fragrance-free policies are increasingly adopted to improve indoor air quality and reduce fragrance exposure linked to adverse health outcomes and accessibility barriers. This scoping review mapped the evidence on fragrance-free policies across sectors. Methods: Following Arksey and O’Malley’s framework and PRISMA-ScR guidelines, we searched five peer-reviewed databases and grey literature sources. Documents describing formal scent-free or fragrance-free policies or guidance in workplaces, healthcare, educational, or public settings were included. Data were charted on policy characteristics, enforcement mechanisms, implementation supports, and governance features. Results: Sixty-three documents were included. Findings revealed substantial variability in terminology, scope, and enforcement. Many policies relied on voluntary compliance and awareness-based strategies, with limited integration of structural supports such as procurement controls, staff training, and evaluation frameworks. Enforcement mechanisms were predominantly reactive, responsibilities were inconsistently defined, and consequences for non-compliance were limited. Although many policies referenced multiple chemical sensitivity or fragrance sensitivity, accommodation pathways and accountability structures were frequently lacking. Four key themes indicate that many fragrance-free policies function as accommodation tools rather than integrated environmental health interventions. Conclusion: Strengthening definitional clarity, institutional accountability, and structural integration may enhance effectiveness and support more consistent, equitable, and evidence-informed approaches to reducing fragrance-related exposures in indoor environments.

Abstract: Soil moisture is a key forecast variable of land surface models. Direct assimilation of microwave brightness temperature data to optimize soil moisture initial fields is an effective approach to improve simulation accuracy of soil moisture. However, most existing direct assimilation methods adopt physical radiative transfer models as observation operators, and their complex parametric errors greatly restrict the improvement of assimilation performance. This study introduces a high-precision MLP-based surrogate radiative transfer model as the observation operator. Combined with the Simplified Extended Kalman Filter (SEKF), it develops a direct radiance data assimilation system for the Common Land Model (CoLM). Assimilation experiments are conducted using brightness temperature data from the Microwave Radiation Imager (MWRI) onboard the FY-3D satellite. Their performance over China's land areas is systematically assessed through comparison with the assimilation scheme based on the Community Microwave Emission Model (CMEM). The results show that the MLP-based assimilation scheme can effectively improve soil moisture simulation accuracy, yet the improvement varies across vegetation types: grassland areas achieve the largest error reduction (10.2%), while semidesert areas present the most prominent increase in correlation coefficient (53.9%). Compared with the CMEM scheme, the MLP scheme exhibits better error stability and produces generally improved assimilation effects—specifically, in semidesert areas, the error decreases by 9.4% and the correlation coefficient increases by 62.8%. This study demonstrates that deep learning-based observation operators have strong application potential for land surface data assimilation under complex physical mechanisms.

Abstract: Mitogen-activated protein kinase (MAPK) cascades, including the ERK1/2, JNK, p38, and ERK5 subfamilies, are central regulators of cellular processes such as proliferation, differentiation, apoptosis, and inflammation. Dysregulation of these signaling pathways is a hallmark of human diseases, most notably cancer, where they drive tumorigenesis, metastasis, and drug resistance, as well as chronic inflammatory and neurodegenerative disorders. The therapeutic potential of targeting MAPKs has spurred the development of numerous small-molecule inhibitors. This review provides a focused analysis of key MAPK inhibitors in oncology, including JNK inhibitors, p38 inhibitors, BIRB-796, ERK1/2 inhibitors and ERK5 inhibitors. We detail their mechanisms of action, preclinical efficacy across diverse cancer models, and progress in clinical trials. Despite promising preclinical data and the entry of several compounds into clinical evaluation, challenges such as low kinase selectivity, off-target effects, and the emergence of resistance mechanisms have limited success, mirroring the difficulties encountered when developing p38 inhibitors for inflammatory diseases. However, emerging strategies, including rational combination therapies, computer-assisted structure-based design for isoform-specific inhibition, and the exploration of novel chemical scaffolds, offer renewed promise. This review concludes by discussing these future directions and the ongoing potential of MAPK inhibitors as a cornerstone of targeted therapy for cancer and other diseases.

Abstract: This paper tries to deal with the Extended, Generalized, and Grand Riemann Hypotheses under a unified framework based on the general properties of L-functions. Specifically, the divisibility properties of entire functions expressed as absolutely and uniformly convergent infinite products with irreducible real polynomial factors (as a result of pairing complex conjugate zeros in the Hadamard product), combined with the uniqueness of zero multiplicities and the symmetric functional equation, force all zeros of the completed L-functions in the critical strip onto the critical line. Consequently, the existence of Landau-Siegel zeros is excluded, thereby confirming the Landau-Siegel zeros conjecture. As to the Davenport-Heilbronn counterexample, since it possesses no Euler product-the fundamental structural property that confines zeros to the critical strip, it is not in the scope of this paper's methods and conclusions.

Abstract: Background: To assess temporal trends in psychiatric hospitalizations among adolescents admitted to a general pediatric ward in Forlì, Italy, between 2016 and 2025, and to examine diagnostic patterns and their association with sex and age, using the individual patient as the primary unit of analysis. Methods: Single-center retrospective study including all hospitalizations for psychiatric disorders in patients aged 10–17 years over a 10-year period. Each discharge episode was assigned to a single dominant diagnostic category. The primary analyses were conducted at the level of the individual patient (first admission), while episode-level analyses were retained as a pre-specified sensitivity analysis reflecting inpatient burden. Results: A total of 165 adolescents (210 hospitalization episodes) were included; mean (SD) age 14.47 (1.73) years, females 77% of patients (80.5% of episodes). Admissions increased significantly over time (patients: IRR 1.20 per year, 95% CI 1.13–1.27; p< 0.001), with 124/165 (75.2%) patients first admitted in 2021–2025. Suicidal ideation/attempt (27.3%) and eating disorders (15.2%) were the most frequent diagnoses and both showed a significant upward trend. Overall, diagnosis was associated with sex (p=0.004): most diagnoses, including the two most frequent, were female-predominant, whereas psychomotor agitation was over-represented in males. Age at admission increased modestly over time and was higher in 2021–2025. Twenty-seven patients accounted for 45 readmissions, concentrated in the most severe diagnoses. Conclusions: Psychiatric hospitalizations of adolescents in a general pediatric ward rose substantially over the decade, especially from 2021, driven mainly by suicidal ideation/attempt and eating disorders. Findings were robust to analysis at the patient level and support the role of the general pediatric ward as a sentinel setting for severe adolescent mental distress.

Abstract: (1) Background: Horsenettle (Solanum carolinense) is a noxious weed widely distributed across North America and increasingly invasive in other regions. Its strong environmental adaptability, complex defense strategies, and distinctive reproductive traits make it an important model for studying plant-herbivore coevolution. However, the absence of high-quality genomic resources has limited deeper investigation into its adaptive evolutionary mechanisms. (2) Methods: In this study we generated a chromosome-level reference genome assembly for S. carolinense using an integrated approach combining PacBio HiFi long-read sequencing, Illumina second-generation sequencing, and Hi-C chromosome-mapping; (3) The final genome assembly has a total length of 915.40 Mb, with a contig N50 of 51.06 Mb and a scaffold N50 of 73.17 Mb; 96.05% of the sequences were successfully mapped to 12 pseudo-chromosomes. The genome is characterized by a high proportion of repetitive sequences (73.64%) and substantial heterozygosity (1.13%), consistent with a highly repetitive and highly heterozygous genome. BUSCO analysis indicates a completeness of 94.7%. A total of 32,206 protein-coding genes were annotated, of which 97.95% received functional annotations; (4) Conclusions: This reference genome provides a valuable resource for advancing research on the adaptive evolution of Solanaceae weeds, supports the development of more effective management strategies for this troublesome species, and offers a technical reference for assembling other highly heterozygous weed genomes.

Abstract: Abstract: Research on exceptional points (EPs) as singularities, where eigenvalues and eigenvectors coalesce, has sparked a revolution in non-Hermitian physics [1–7], offering unprecedented sensitivity and wave manipulation. Previous studies have predominantly focused on isolated points in the complex plane [8–12], often relying on nonphysical complex parameters or active gain–loss modulation. However, such approaches introduce significant system complexity and hinder scalability, leaving the realization of continuous EP structures in purely passive, real-world systems an open challenge [13–18]. To address this challenge, this study reports the discovery of an EP surface within a purely passive, real-parameter, two-degree-of-freedom (2DOF) damped system. A hidden physical landscape was unveiled, termed the L-surface, representing clusters of loci of innumerable EPs. Leveraging the L-surface derived from exact analytical solutions, this study identified and validated fundamental topological phenomena, including topological jump, imprint, and nucleation, all of which were previously obscured by numerical noise [19–23]. The comprehensive analysis and precise identification of this manifold required a physical parameter precision of 100 decimal places. This regime has been conventionally dismissed as mere numerical noise in standard 64-bit double-precision floating-point formats. The unveiled L-surface provides a robust foundation for next-generation ultrasensitive sensing and perfect energy absorption across frontiers, ranging from quantum computing [24–26] to advanced biosensing [27, 28], extending its transformative impact to the broader realms of electronics [29, 30] and optics [31–33].

Abstract: Geopolymers represent a promising material platform for extrusion-based 3D printing; however, current research remains largely focused on mix design, rheology, printability, buildability, and the relationship between process parameters and the resulting microstructure (1–7). This article therefore compares the behaviour of two 3D-printed geopolymer composite truss beams with reference cementitious composite beams developed within the 3D STAR project (8). The geopolymer elements, 2932 mm long, were designed for the same material volume and target geometry as the reference CC element; however, because of mixture spreading, they reached cross-sections of only approximately 140/250 mm and 170/250 mm. The first beam was printed without setting acceleration, while the second was locally treated with a hot-air gun. In four-point bending, the geopolymer beams reached maximum forces of 22 and 29 kN, whereas the reference cementitious beams reached 31 and 40 kN. The CC elements failed by rupture of the tensile reinforcement, while the GC elements failed by joint failure followed by deformation and local disintegration of the composite. The study thus shows that the main difference between the two systems lies not only in the achieved load-bearing capacity, but also in stiffness, the shape of the load-displacement diagrams, and the failure mechanism.

Abstract: Eye diseases such as trachoma, allergic conjunctivitis, and dry eye syndrome have shown increasing prevalence in regions experiencing adverse environmental and climatic changes. Factors such as air pollution, dust exposure, humidity variations, and ultraviolet (UV) radiation directly impact ocular health, especially among vulnerable populations. In this study, we develop a deterministic compartmental model to explore the dynamics of environmentally-driven eye disease transmission and progression. The model integrates climate-sensitive variables, such as dust concentration and humidity, into the transmission and recovery rates of the disease. We analyse the model's equilibria, investigate the basic reproduction number R₀, and assess the influence of environmental mitigation Strategies on disease control. Numerical Simulations are provided to illustrate how seasonal and anthropogenic changes in environmental conditions affect disease prevalence over time.

Abstract: A systematic evaluation of salt tolerance at the seedling stage was conducted using 143 maize inbred lines in this study. The results showed that salt stress significantly inhibited seedling growth and enabled the identification of several elite salt-tolerant inbred lines, including B114. Salt tolerance was significantly positively correlated with anti-oxidant capacity. Under salt stress, the highly salt-tolerant line B114 exhibited lower membrane damage, stronger reactive oxygen species scavenging capacity, and a significantly higher survival rate than the salt-sensitive line PHT55. Using random for-est-based machine learning, 50 core salt tolerance-related genes were unbiasedly identified from high-dimensional transcriptomic data. Functional enrichment analysis revealed that these genes were primarily involved in redox regulation, ion homeostasis maintenance, and stress signal transduction pathways. This study established a maize salt tolerance evaluation system closely aligned with field conditions and demonstrated that coordi-nated temporal transcriptional regulation represents a core molecular mechanism under-lying high salt tolerance in maize. The elite salt-tolerant germplasm and key candidate genes identified in this study provide valuable genetic resources and a theoretical founda-tion for molecular breeding of salt-tolerant maize adapted to saline-alkaline soils.

Abstract: Pinus radiata is a commercially valuable forest species whose productivity is increasingly threatened by climate change-related biotic and abiotic stresses such as diseases, soil salinization and rising temperatures. Taking as a starting point a sufficiently optimized somatic embryogenesis propagation protocol for radiata pine, this study aimed to evaluate the morphological and biochemical responses embryogenic cell lines subjected to different elicitors such as methyl jasmonate and tagatose during proliferation or sodium butyrate at maturation stage. The effect of various stresses such as salinity and high temperatures was evaluated at proliferation and germination stages. Then, biochemical analyses (total protein, hydrogene peroxide and malondialdehide contents) were carried out in embryogenic tissues and plants and the growth was also assessed in these tissues and in somatic plants. This latter growth assessment led to a classification of plants depending on their suitability to be planted ex vitro. Several embryogenic cell lines were tested and the response to elicitation or stress was found highly genotype-dependent however some trends could be observed particularly when assessing growth. These results suggest that somatic embryogenesis combined with stress priming or elicitor application may be a viable strategy to enhance biotic and abiotic stress tolerance in radiata pine.

Abstract: Traditional Digital Twins (DTs) in energy sectors lack cyber-threat awareness, while cybersecurity DTs overlook downstream physical impacts. Loosely coupled co-simulations attempt to bridge this gap but introduce computational lags that mask critical cross-domain vulnerabilities. To address these limitations, this paper proposes a unified, tightly coupled DT framework that integrates energy systems and cybersecurity domains into a single environment. The methodology models the precise mathematical, thermal, and electrical constraints of key assets to capture cross-domain feedback loops. Specifically, a power transformer and a microgrid-connected inverter serve as case studies to map cyberattack vectors directly onto physical definitions. Numerical validation evaluates multiple threat scenarios, including supervisory, measurement, and physical-level (harmonic) attacks on the transformer, alongside short-circuit and hybrid phase-harmonic attacks on the inverter. Results demonstrate how subtle digital disruptions propagate past communication layers to induce physical degradation and operational stress. By explicitly detailing the governing equations and providing sensitivity analyses, this work delivers a transparent, high-fidelity methodology for protecting critical cyber-physical infrastructures from asset-destructive manipulations.

Abstract: Background. Lower limb exoprostheses often lead to stump dermatological pathologies. The mechanisms by which mechanical microtraumas progress to non-healing ulcerative defects due to dysbiosis remain poorly understood. Study Objective. To analyze mechanical, inflammatory, and infectious stump skin complications and justify the role of bacterial and mycological dysbiosis in blocking tissue regeneration. Materials and Methods. Scale for the Assessment of Narrative Review Articles principles were used to quality control the narrative review. A targeted search was conducted in PubMed and Scopus databases. Search dates ranged date in January 1980 to May 2026. Results. Skin damage dynamics were categorized into three stages: adaptation (up to 12 months), chronic reactive changes (12–24 months), and late proliferative-infectious destruction (>24 months). The sealed sleeve space creates 100% humidity and alkalization (pH >6.5). This causes a "fungal shift," where resident Malassezia spp. lose dominance to invasive Candida albicans and non-dermatophyte molds (Aspergillus spp., Fusarium spp.). These pathogens form polymicrobial biofilms with Staphylococcus aureus. At the molecular level, delayed regeneration is driven by “frustrated phagocytosis”: macrophages, unable to engulf large fungal hyphae, continuously release reactive oxygen species and enzymes, trapping the wound in the inflammatory phase. Excessive matrix degradation and suppressed angiogenic factors further block epithelialization. Conclusion. The skin under a prosthesis socket forms a unique pathological biotope. Successful regeneration requires preventive mycobiota correction and targeted management of biophysical parameters (pH, humidity) within the “skin-liner” interface.

Abstract: Exposing foams stabilized by photoswitchable surfactants to UV light induces changes in surface surfactant concentration, leading to significant alterations in foam behaviour such as the generation of Marangoni flow and change in foam drainage patterns. The occurrence of Marangoni flow can be observed when either all elements of the foam or only their films are exposed to UV light. Conversely, changes in foam drainage occur when a macroscale portion of a foam column is exposed to UV light. To explore these phenomena, numerical models are developed and validated using experimental data. These models simulate the scale and profile of Marangoni flow from foam networks to films as well as the drainage flow within the foam network. Microscale findings demonstrate that Marangoni flow can be controlled by adjusting the intensity and duration of UV light exposure. Macroscopically, the drainage profile in exposed foam regions undergoes significant changes with varying UV intensity. Furthermore, beyond a certain threshold, the foam drainage reverses direction, contrary to gravity. The effect of foam interfacial mobility on the reversed drainage of both interior and exterior foams is analyzed. The findings provide a potential tool to control foam drainage behaviour without the need to modify other variables.

Abstract: Multimodal misinformation detection has advanced from text-only classification toward joint processing of language, imagery, and social metadata. Yet much of the literature still treats adaptation as an internal fusion problem rather than a systems-level routing problem, leaving open the question of when multimodal escalation is warranted and how expert usage can be reduced without materially weakening predictive quality. This paper introduces MemANS, an event-driven, in-memory Agentic Name Search architecture for multimodal misinformation detection. MemANS treats inference as a resolution process over named experts: a default text–metadata resolver handles ordinary cases, while a specialist is invoked only when cross-modal disagreement and confidence conditions jointly indicate that additional reasoning is justified. The empirical study is conducted on a balanced, fully image-available Fakeddit benchmark comprising 5400 training instances, 840 validation instances, and 840 test instances across six misinformation classes. On the final test set, MemANS achieves 0.6182 macro-F1, 0.6167 accuracy, 0.5405 MCC, and 0.8726 weighted one-vs-rest AUC, clearly surpassing the always-on fusion baseline (0.5732 macro-F1) with a statistically supported gain of 0.0454 (p < 0.001) while using only 1.7202 average experts versus 2.0000. The results indicate that Agentic Name Search is a viable and practically meaningful multimedia-systems abstraction for adaptive multimodal reasoning.

Abstract: This research paper examines the critical intersection of traditional ecological knowledge (TEK) protection, dam development, and glacial lake outburst flood (GLOF) mitigation in mountain regions, particularly the Hindu Kush-Karakoram-Himalaya (HKKH) region. Drawing from 77 peer-reviewed studies, this review synthesizes evidence on how indigenous and local knowledge systems can be protected and integrated into large-scale infrastructure projects and disaster risk reduction strategies. Key findings reveal that while TEK offers invaluable insights for GLOF early warning systems and community-based adaptation, significant gaps persist in formal protection mechanisms, particularly in dam planning contexts. The paper identifies successful participatory frameworks, documents persistent challenges in knowledge integration, and proposes policy recommendations for safeguarding TEK while enhancing climate resilience in vulnerable mountain communities. This synthesis is particularly timely given the accelerating impacts of climate change on glacial systems and the expansion of hydropower infrastructure in high-mountain Asia.

Abstract: The reliability of unrelated-donor searches depends on high-resolution HLA typing, yet a large fraction of records in national stem-cell donor registries were generated at low or intermediate resolution and are therefore under-used in modern matching. Here we develop an Extreme Learning Machine (ELM) approach that upgrades low/mid- to high-resolution HLA data by learning the haplotype and diplotype structure of a national donor population and assigning the most probable high-resolution genotypes together with posterior probabilities. The model was trained on the Greek national registry (Hellenic Transplant Organization, established 2002; 117,345 donors, ~20% low-resolution) and validated on two independent Greek cohorts (ORAM, n = 20,100; GRPT, n = 4,353) using accuracy and call-rate metrics. The population-specific ELM achieved a per-locus accuracy of 70–94% (depending on the confidence threshold) with an overall call rate of 98.1%, recovering usable high-resolution information and increasing the proportion of registry donors usable in high-resolution matching. The method is fast, lightweight and population-tailored, complementing established expectation-maximisation imputation tools.

Abstract: Relevance. The solvent accessible surface area (SASA) of amino acid residues is a key characteristic for protein structure analysis, but precise methods for calculating it (e.g., FreeSASA) are computationally expensive. Empirical approximations based on the res-idue interaction network (RIN) graph can provide high speed while maintaining ac-ceptable accuracy. Proposed approach. Three empirical functions for estimating relative SASA are pro-posed: approx_sasa, surface_score, and exp_sasa using the degree of the node in RIN as an argument. We present a comparative study of two approaches to graph construction: the classical Cα-graph (threshold 8 Å) and the graph of heavy atoms (Heavy-Atom Graph, HAG, threshold 5.0 Å). The parameters were calibrated on a sample of 509 protein structures (128,794 residues) from various origins using the true relative SASA calculated by the FreeSASA library. Main results. An extended set of 11 RIN topological features was developed and vali-dated, including basic node characteristics, centrality measures (betweenness, eigen-vector, closeness) and hydrophobic subgraph features. Training ensemble models (Random Forest, XGBoost) with these features made it possible to achieve: Random Forest on HAG: MAE = 0.057 ± 0.033, Pearson r = 0.915 ± 0.080 (best result), Random Forest on Cα graph: MAE = 0.066 ± 0.041, Pearson r = 0.890 ± 0.100. Comparison with GNN. We compared our approach with graph neural networks (GCN, GAT, GraphSAGE). GraphSAGE on HAG showed a result close to Random Forest: MAE = 0.0715, Pearson r = 0.8917, indicating the potential applicability of graph neural net-works when using HAG. GCN and GAT performed significantly worse (MAE = 0.14–0.15, Pearson r = 0.51–0.61). Computational efficiency. Empirical formulas are calculated in 0.008 ms per structure (~26,000× faster than FreeSASA), Random Forest in prediction mode is calculated in 36.5 ms (~6× faster than FreeSASA). HAG construction takes 21 times longer than a Cα graph (279.5 ms vs. 13.3 ms). Practical significance. The proposed empirical features are recommended for large-scale pipelines critical to speed and interpretability. Random Forest on HAG is the optimal choice for tasks that require maximum accuracy (MAE = 0.057, Pearson r = 0.915). GraphSAGE on HAG can be considered as an alternative when using deep learning.