Journal of Kerbala University

In recent year, the use of methods to investigate the correlation between trace elements and pathologies has increased in prevalence. The samples of serum from 201 people in Karbala, Iraq were collected. The group included healthy individuals and those with hypertension. Their ages ranged from 16 to 75 years. The lead levels were quantified using inductively coupled plasma optical emission spectrometry (ICP-OES). The relationship between blood lead levels and various factors, including health, gender, and smoking, was examined and discussed using the F-test and a two-tailed t-test. The lead concentration (Pb) in blood serum of people with hypertension (11.123 ± 16.898)μg/g was significantly higher than that in healthy people (1.662 ± 0.442) μg/mL, with p=0.01. Although the difference in lead (Pb) concentration between males and females was not statistically significant (p=0.059), the average Pb concentration in male serum (7.306 ± 12.069) μg/mL was higher than in female serum (4.422 ± 12.300) μg/mL. In contrast, the difference in lead concentrations between smokers (5.28 ± 9.76) μg/mL and non-smokers (6.65 ± 14.64) μg/mL was not statistically significant, with ρ=0.231. In summary, a study of 201 people in Karbala, Iraq, showed that lead (Pb) levels in serum were significantly lower in healthy individuals compared to those with high blood pressure. Also, males had higher lead levels than females. However, smoking did not seem to have a significant effect.

Background: Since wheat and barley are vital food supplies for both people and animals, it is critical for public health to check their radiological safety. Long-term health hazards may result from exposure to naturally occurring radionuclides, especially radon (Rn) and radium (Ra). The purpose of this study was to assess the radon concentration (CRn), effective radium content (CRa), radon emission rate (RER), alpha index (Iα), and effective dose rate (EDR) in wheat and barley crops grown throughout Iraq.
Materials and Methods: A total of 22 samples of wheat and barley were collected from 11 Iraqi governorates. . CR-39 detectors were used for alpha-track detection in order to quantify CRn, CRa, RER, Iα, and EDR.
Results: A wheat sample from the Basra Governorate had the greatest values of CRn, CRa, RER, Iα, and EDR (79.328 Bq/m³, 4.070 Bq/kg, 0.003 Bq•m⁻³•h⁻¹, 0.020, and 150.500 µSv/year, respectively), whereas samples from Anbar had the lowest values. The southern governorates showing the highest radiological impact. Northern regions showed intermediate values, while central regions recorded the lowest amounts. In certain samples, improper application of agricultural fertilizers was also linked to elevated radioactivity. Every measured value stayed below the suggested safety limits set by the WHO and ICRP.
conclusion: Despite regional variations, radiological evaluations show that Iraqi wheat and barley meet international safety criteria. These results emphasize the need for ongoing agricultural product monitoring to guarantee food safety and identify possible radioactive environmental sources.

Oral bacteria can be exhibited by antagonistic activity of Lactobacillus spp. isolated from different sources. The current study antagonistic activities was investigated in vitro of Lactobacillus acidophilus and Lactobacillus plantarum isolated from against bacterial oral cavity . One hundred fifty samples were collected from patients with problems of oral cavity from September 2024 to December 2024. The isolates were cultured and identified biochemically. Molecular identification for Lactobacillus spp. was performed by PCR technique. Thereafter, cell-free supernatant (CFS) of those species was prepared from MRS culture and used in an antimicrobial assay that was conducted by agar well diffusion method. Results showed that both species revealed a significant (P < 0.05) antibacterial effect against tested pathogens including Sphingomonas spp., Streptococcus spp., Staphylococcus aureus, Pseudomonas aeruginosa, and E. coli. In conclusions, cell-free supernatant produced by isolated Lactobacillus spp. exhibited antagonistic activity against a wide range of Gram-positive and Gram-negative oral cavity pathogens suggesting its potential application in new functional probiotic products for oral healthcare.

The complexity of the pathophysiology of autoimmune diseases, along with the evasive mechanisms of these conditions, makes the diagnosis and treatment of these conditions highly challenging. The use of "Artificial Intelligence" in scientific and medical research has opened new ways to investigate, diagnose, forecast, and treat autoimmune diseases. Recently, "AI" has become a valuable tool in medical immunology and rheumatology, providing creative solutions to the underlying intricacy of certain conditions. Research is currently demonstrating Artificial Intelligence's potential in sickness prognosis, risk classification, treatment reaction prediction and evaluation. Through the use of sophisticated machine learning and deep learning algorithms, "AI" systems can analyze large-scale medical, genetic, and imaging facts. They can also identify diffused styles that people usually miss. These findings improve diagnosis accuracy and make it easier to create individualized, focused treatment plans. Additionally, "AI" applications in autoimmune disease research improve clinical decision-making, allow for more accurate and customized treatments, reduce trial-and-error in drug selection, eventually enhance long-term patient outcomes and disease management.

Bacterial toxins are master manipulators of host immunity, serving as crucial virulence factors that enable pathogens to evade, dysregulate, and ultimately overcome host defenses. This article review aimed to explore how bacterial toxins, including exotoxins and endotoxins, change immunological activities and cause diseases. Exotoxins, which include super antigens, membrane-living toxins, and A-B type toxins, disrupt the communication between immune cells, increase cytokine production, leading to cell lysis or immunological paralysis. Endotoxins, especially lipopolysaccharide from Gram-negative bacteria, activate macrophage and causes systemic immune reaction, leading to septic shock. Moreover, TA systems, which regulate bacterial persistence and biofilm formation, thereby promoting chronic infections that evade immune clearance and antibiotic treatment, biofilm, and multidrug resistance, allowing them to stay alive, even under stress conditions. Finally, cyto-lethal distending toxins or CDTs, act as genotoxin and damage host’s DNA, modulate the cellular cycle, and suppress adaptive immunity, for example, by inhibiting T and B cell proliferation. Recently, progress in the development of a cell-coated or CMN nanoparticles represents a novel approach to toxin neutralization, which anchors bacterial toxins away from their physiological receptors by mimicking host cells. Therefore, a comprehensive understanding of how toxins, TA systems, and CDTs subvert immunity is crucial for developing targeted anti-virulence strategies and novel therapeutics to counteract antibiotic-resistant infections

Background:Procalcitonin (PCT) is used to differentiate between bacterial and viral infections; nevertheless, its effectiveness in the context of COVID-19 is unknown. The inflammatory response associated with COVID-19, mediated by IL-6 cytokine activation, may contribute to increased levels of PCT regardless of bacterial coinfections.
Methods:This prospective case-control study recruited 200 patients during the 2025 Arbaeen pilgrimage. Serum levels of PCT, IL-6, CRP, WBC count, and bacterial cultures were used to assess their value in detecting bacterial coinfections among cases with or without COVID-19.
Results:Significantly, PCT levels were not increased in cases with or without bacterial coinfections. The incidence of bacterial coinfections was significantly reduced in cases with COVID-19 compared with cases without COVID-19. On the other hand, levels of IL-6 were substantially increased in cases with COVID-19 compared with cases without COVID-19. The ROC curve showed that PCT was of limited value in detecting bacterial coinfections (AUC = 0.72; 95% CI = 0.65 to 0.79). The sensitivity of PCT at an optimal cutoff of 0.12 ng/mL was 71%, with specificity of 68%. Multivariable logistic regression identified IL-6 and WBC count as independent predictors of bacterial coinfections.
Conclusion:Procalcitonin is an inaccurate marker for detecting bacterial coinfections in the context of COVID-19 because of its relationship with IL-6 levels. The effectiveness of IL-6 levels and WBC count in detecting bacterial coinfections makes them more useful than PCT levels.

Complete blood count (CBC) and iron studies parameters are the core markers of health and disease. This retrospective research examined red blood cell (RBC) complete blood count indices, such as RBC HGB HCT, MCV MCH MCHC, RDW, and ferritin levels, in male and female patients as well as in the patients from 1570 clinical records who had and did not have a diagnosis of thalassemia (DX). We have illustrated the patterns by doing descriptive statistics, correlation analyses, and comparative tests (t tests ANOVA). The data collection size is 1570 sample. The obtained results are in a good agreement with the earlier reported physiological differences like the significantly higher levels of RBC, HGB, and HCT in males as compared to females (p<0. 001). Additionally, the DX=1 group was characterized by higher RDW and ferritin levels, which could reflect the body's inflammatory response or dyserythropoiesis. This kind of data, initiated investigation not only highlights the necessity of the demographic context for the interpretation of the routinely done blood tests but also points out certain markers that are especially worrying in a specific group of patients. This paper is an example of the use of retrospective data mining for producing new clinical research hypotheses.

Many people believe that vitiligo, an acquired depigmenting sickness marked by the selective loss of melanocytes, is an immune-mediated illness. Immunological markers may improve diagnostic accuracy and offer insight into disease activity and development, despite the fact that clinical examination is still the mainstay of diagnosis. Vitiligo is a long-term, acquired depigmenting condition that causes distinct white macules and patches due to a focal loss of functioning melanocytes. Even though vitiligo's clinical characteristics are widely known, its complicated pathophysiology entails a complex interaction between immunological dysregulation, oxidative stress, and genetic predisposition. There is mounting evidence that vitiligo is primarily an autoimmune condition brought on by innate and adaptive immune responses that target melanocytes. Important immunological processes include increased reactive oxygen species that lead to cellular stress, aberrant activation of innate immunity cells including natural killer cells and dendritic cells, and the resulting generation of cytotoxic T lymphocytes that are specific to melanocytes. Cytokine networks, including interferon-γ and the CXCL9/10 chemokine axis, which promote T cell recruitment and melanocyte death, further contribute to the breakdown of immunological tolerance. Chronicity of the disease is also influenced by altered antigen presentation and dysfunctional regulatory T cells. Diagnostic methods for vitiligo such as imaging techniques like ultraviolet light photography, reflectance confocal microscopy, computer-assisted imaging analysis, optical coherence tomography, visible light and digital photography, colorimetry, spectrometry and immunological diagnosis. We conclude that the comprehension and clinical assessment of vitiligo have been greatly improved by recent developments in diagnostic tools.

The necessity of creating efficient techniques for identifying offensive language, especially hate speech, has increased due to the quick expansion of user-generated material on social media platforms. Even though machine learning and deep learning models have made significant strides in this area, accuracy and related quantitative metrics are frequently the only criteria used to evaluate them. Despite their usefulness, these metrics fall short in capturing the intricacies of learning dynamics or the interpretability of model predictions. By including training curve analysis and Local Interpretable Model-agnostic Explanations (LIME) into the assessment of hate speech recognition models, with an emphasis on the Iraqi Arabic dialect, this study overcomes this problem.
We experimented with three different methods: first, using training FastText and BiLSTM separately, second, BiLSTM with FastText embeddings, and third combining FastText embeddings with lexicon-based features. A closer examination of the second method showed significant subtleties, even though it had the greatest reported accuracy of 97.3%. While LIME interpretations offered insight into the linguistic elements underlying classifications, training curves revealed problems with overfitting and stability. These revelations highlighted the significance of interpretability and training dynamics by exposing misclassifications that accuracy alone was unable to account for.
The results emphasize how important it is to implement thorough assessment frameworks that strike a balance between qualitative comprehension and quantitative performance. These methods guarantee the creation of hate speech detection models for Arabic dialects that are more resilient, trustworthy, and context-sensitive.

In this study, we suggest an enhanced and systematic model for achieving sustainability in the redevelopment projects of Iraqi cities. Through complex decision-making tools such as Dijkstra’s algorithm, A* search, and linear integer programming models. The research project aims to perfect the most effective ways to invest in urban infrastructure. The results suggest the approach to combine cost functions and evaluation heuristics is viable, generating solutions that are capable of serving dynamic urban requirements while respecting time and finance limitations. Nowadays, the best financial distribution is considered a problem in urban development projects, particularly those for post-conflict Iraqi cities, and answers are not always sufficient due to budget or time constraints. This paper provides a results-oriented, cost-effective decision-making method by integrating traditional optimisation and AI-driven search algorithms. This includes that of Dijkstra's algorithm, A* search, and hill-climbing heuristics involving linear integer programming for providing quality-wise candidate project prioritization and scheduling suggestions. The models developed are verified with otherwise entirely known, researcher-generated data derived from artificial scenarios that model realistic city planning scenarios (although these costs, implementation times , and delay penalties are up to the discretion of the researchers designing each synthetic scenario). The data is for proof-of-concept and decision support, not empirical generalisations at scale. The results indicate that the use of math-based cost models with heuristic assessment schedules can effectively reduce total project costs in order to minimise delay penalty and increase screening efficiency as compared with traditional scheduling methods. The lab presented is a tangible proof-of-concept to enhance the resilience and transparency of urban development processes in environments where resources are scarce.

The need to address emerging challenges presented by the ever-expanding capabilities of genomics, proteomics, and medical imaging research introduces new challenges to clinical machine learning (ML). Examples of such variables include inefficiency in computations and overfitting. To address these issues, in this study, feature selection (FS) paradigm is proposed, which integrates predictive performance and feature parsimony, grounded on the Ant Colony optimization (ACO). The framework effectively searches big feature spaces in order to locate small useful subsets. The evaluation of candidate features subsets chosen by ACO is done using three classifiers (RF, SVM and GBM). With as few as 300 features, dimensionality reduction (by up to 92 percent) is observed without losing accuracy on four high-dimensional biological datasets, with Leukemia achieving 96.4. Enrichment analysis of Gene Ontology proves that our method is more accurate, more stable, and biologically relevant than filter-based methods and metaheuristics. The method is quite suitable to a clinical environment having limited resources due to its competitive running times and a fast convergence (around 70 iterations). Developing feature sets, which are miniature, comprehensible, and physiologically pertinent, the results demonstrate that the proposed framework can enhance clinical decision assistance frameworks and tailored therapy.

As a key paradigm in movie recommender systems, model-based collaborative filtering (CF) addresses systemic weaknesses of memory-based approaches, such as poor scalability and data sparsity. Many surveys document CF algorithms but often lack a taxonomic framework and a critical analysis of trade-offs among accuracy, scalability, and interpretability, especially for movie recommendations. This survey attempts to fill this gap by (1) proposing an original five-tier taxonomy that classifies model-based CF into: (i) matrix factorization (SVD, ALS, SVD++), (ii) association rule mining (Apriori, FP-Growth), (iii) probabilistic models (e.g., Bayesian networks), (iv) deep learning (NCF, GNNs), and (v) hybrid architectures; (2) conducting a comparative analysis of 23 recent works (2019–2025) evaluated on MovieLens, Netflix, and TMDB under varying sparsity conditions; and (3) critically assessing the accuracy–scalability trade-off, highlighting that matrix factorization methods remain competitive with deep learning approaches despite substantially lower computational complexity. According to our findings, hybrid models that combine frequent itemset mining with collaborative filtering improve cold-start performance (F1-score increases of 12-18%). While pure deep learning models often achieve marginally higher accuracy on dense datasets, they suffer from opacity and high computational demands—making hybrid models more practical for real-world deployment, where interpretability and efficiency are prioritized. There are three open challenges that we identify: (i) the ability of a system to adapt dynamically to changing user preferences without retraining, (ii) the ability to create a personalized movie recommender under the GDPR while guaranteeing user privacy, and (iii) the need for a standardized evaluation metric to compare performance systematically on different datasets. This review provides researchers with a structured mapping of CF techniques to dataset characteristics and application requirements, aiding practitioners in selecting relevant CF solutions.

Highly imbalanced classes are devastating to network intrusion detection, where some atypical but high-impact attacks can comprise less than 0.1% of the traffic volume. Conditional GANs have been found to be very successful in sampling minority classes; however, their centralized learning contradicts privacy needs in a distributed setting. Based on previous federated GAN architectures, my architecture, federated conditional GANs (FedGAN-IDS), introduces a quality-weighted form of an aggregation scheme, local discriminator equilibrium, as a proxy to address the weakness of federated conditional GANs, in which minority classes are not locally represented. This incremental approach modifies past client-selection methods in federated GANs by down-weighting erratic clients in non-IID intrusion detection tasks. FedGAN-IDS in IID partitioning (ε ≤ 3.0, δ = 10⁻⁵) was trained on a simulated dataset analogous to CIC-IDS-2017 and yielded a macro-F1 of 0.6704. It also results in an absolute improvement of 0.1049 pp (18.5% relative) on minority classes and 0.1074 pp higher than non-generative federated baselines, such as the federated classifier (macro-F1 0.5630). Ablation experiments prove that the gains are caused by quality-weighted aggregation and that the gains are significant in non-IID cases (p < 0.05 by paired t-tests). Limitations: The method employs synthetic data, which should be tested on real-world network traces to have wider applicability.

Sequence-to-sequence (seq2seq) text generation is a fundamental problem in natural language processing, which underlies applications such as chat-bots, machine translation and creative writing. This research investigates the fine-tuning of pre-trained Text-to-Text Transfer Transformer models, such as T5-base and T5-large on a conversational dataset, concerning their ability to produce contextually coherently and semantically relevant textual completions. The models were trained on the DailyDialog dataset with a maximum input length of 256 tokens. They were evaluated using both automatic metrics (accuracy, perplexity and BLEU scores) and diversity and fluency in qualitative evaluations.
The larger T5-large model outperformed the smaller one. It achieved a next-sentence prediction accuracy of approximately 97.9%, at a final loss in the range between 0.7 and 1.2, and produced responses that were nearly indistinguishable from human naturalness judging with perplexity scores but had good variety and conveyed interesting, conversational content or knowledge. T5-base; however, showed competitive performance with less precision and diversity. That made it a good fit for situations that have limited computing resources. These findings demonstrate the importance of trade-offs between model size and resource constraints for seq2seq models in text completion applications.
Our results are in line with other work that highlights the high generalization capabilities of T5 across varied NLP tasks, and suggest that it has strong potential for efficient language synthesis, given adequate optimization. This work is part of ongoing efforts to enable better dialogue systems and provides practical guidance for applying T5 variants for conversational AI.

Intrusion Detection Systems (IDSs) are representing a core of component of organizational network security, as they serve as the first line of defence against a wide range of cyber threats. They are responsible for effectively handling potential intrusions on the network. Internet attacks severely damage a website or server, make them inaccessible to other users. Numerous IDS systems employ flow-based network traffic analysis to identify potential threats. In this paper, we proposed a LightGBM(Light Gradient Boosting Machine) machine learning model that designed to create a flow-based anomaly detection intrusion detection system for multi-class classification. The model is evaluated using the updated CICIDS2017 dataset) CICIDS2017_Cleaned( for multi-attack classification. The model achieved an accuracy of 0.9984,F1-score of 0.9591and precision of 0.9356. The model has excelled in accuracy, detection rate, and low false alarm rate.Because they often use machine learning techniques for rather than intrusion detection systems. The study shows that the results are difficult with multi-class classification on imbalanced data sets, and lacks effective dynamic detection mechanism. In order to remedy this research gap, this article proposes a lightGBM-based dynamic intrusion detection system blending the CICIDS2017 dataset with SMOTE.

The rapid expansion of short-form video platforms has increased the demand for intelligent systems capable of understanding human emotions expressed across multiple modalities. Multimodal sentiment analysis (MSA) addresses this challenge by integrating textual, acoustic, and visual information within unified deep learning frameworks. This survey provides a structured and analytical overview of recent multimodal sentiment analysis approaches, emphasizing early, late, and hybrid fusion strategies. Representative architectures, including transformer-based models, convolutional neural networks, and recurrent neural networks, are comparatively examined, with particular focus on cross-modal attention mechanisms that enhance interaction modeling and predictive performance. Benchmark datasets such as CMU-MOSI, CMU-MOSEI, MELD, IEMOCAP, CH-SIMS, and TikTok-10M are analyzed to highlight their importance for training and evaluating multimodal systems across diverse linguistic and cultural contexts. Comparative findings suggest that hybrid transformer-based architectures generally outperform unimodal approaches, especially in detecting subtle emotional expressions such as sarcasm and mixed affect. Evaluation practices are reviewed across both classification metrics (Accuracy, Precision, Recall, F1-score) and regression metrics (MAE, RMSE), emphasizing metric selection under class imbalance. Despite progress, challenges persist, including limited annotated datasets.

Chaotic dynamical systems present a major problem to the numerical simulation because small errors introduced in the discretization process can be rapidly magnified and this restricts the ability to predict long-horizon trajectories. In many applications, this is not precise pointwise long-time agreement that is desired, but the long-time reproduction of qualitative long-time dynamics and long-time statistical behaviour. The paper explores the stability of high-order time integration schemes of chaotic ordinary differential equations with time-adaptive step control. An effective framework of stability is embraced which measures (i) boundedness and sound time evolution, (ii) attractor geometry by phase-space projections, and (iii) consistency of long-time statistical clues to a rigid reference arrangement. The analysis presents a comparison of fixed-step baselines and high- order adaptive algorithms on selected chaotic benchmarks, cost metrics based on function assessments and diagnostics based on both short window differences and long-time summaries. The findings indicate high-order adaptive integration is very stable and has high statistical fidelity and has good computational efficiency. Next, step-size histories also indicate that adaptive control selectively deploys the resolution of dynamics to adapt to the changing dynamics locally, which is why it is also a factor in contributing to stability-relevant processes in chaotic simulation. On the whole, the results give a viable foundation to the choice of integrator order and tolerance parameters when a dependable long-run behavior and assembly properties are needed.