Clinical code prediction maps unstructured discharge summaries to ICD-10-CM leaf codes in a large, sparse, and deeply hierarchical label space. Most systems treat the task as flat multi-label classification, scoring codes independently and providing limited training signal for rare labels. We propose a graph-constrained traversal policy that formulates ICD prediction as a finite-horizon decision process over a pruned code hierarchy. A single language model descends the graph level by level, selecting valid child nodes until billable leaf codes are reached. This converts extreme multi-label prediction into sparse, hierarchy-aware subset decisions while guaranteeing structurally valid outputs. On MIMIC-IV discharge summaries, our best supervised policy, SFT-1+, achieves 0.709 micro-F1 on a curated 50-code subset and 0.527 micro-F1 on the full 15,761-code space, outperforming flat baselines including CAML, LAAT, and PLM-ICD. In the full setting, SFT-1+ improves over the strongest flat baseline by 0.044 micro-F1 and 0.157 macro-F1, suggesting that graph-constrained decomposition mitigates the rare-code bottleneck. A controlled factorial study evaluates architecture, training algorithm, and data budget. Across both scales, one shared policy matches a three-specialist cascade while avoiding its context-window overflow on 28-32

Synthetic clinical data are increasingly used for healthcare machine-learning development, model validation, data sharing, and predeployment testing, yet such data often claim to be trustworthy after passing a limited collection of realism tests. A synthetic dataset may indeed claim statistical similarity while leaking training membership, erasing rare subgroups, failing on held-out real patients, or lacking sufficient artifacts for reproduction. We introduce SynTrustBench, an evidence-gated and executable benchmark for evaluating trustworthiness claims across five non-compensable dimensions: fidelity, clinical utility/validity, privacy, equity, and robustness/generalization. Its Evidence Assessment component audits published reports and produces a five-element Evidence Maturity Profile (EMP) together with a separate evaluability gate. Its executable structured-tabular protocol accepts frozen real training data, held-out real test data, a synthetic table, and a declarative configuration; computes dimension-specific metrics and uncertainty; and produces subgroup results, failure flags, benchmark cards, and provenance manifests. In a frozen pilot audit of 30 reports, 17 of 30 quantitatively evaluated privacy, 2 of 30 documented a formal privacy guarantee to the audit threshold, 2 of 30 evaluated equity, 12 of 30 evaluated robustness, and only 4 of 30 passed the evaluability gate. The executable implementation operationalizes the same dimensions through distribution and dependency checks, frozen train-on-real/test-on-real (TRTR) and train-on-synthetic/test-on-real (TSTR) utility, empirical privacy attacks, subgroup analysis, perturbation testing, and a controlled failure-injection harness. SynTrustBench does not certify clinical safety or collapse trustworthiness into a single score. Instead, it provides an inspectable predeployment contract for identifying what was evaluated, what failed, what remains unknown, and whether evidence is sufficiently complete and reproducible for comparison or downstream healthcare AI use.

Covariate shift across training-data partitions biases model selection and parameter estimation in cross-validation, lifelong learning, and federated learning. We propose Partition-Induced Covariate-shift Correction (), a Fisher information-based regularization framework that mitigates distribution mismatch between data partitions and a reference distribution. approximates partition divergence using the Fisher Information Matrix (FIM) and incorporates the resulting statistic as a regularizer during optimization. The same formulation applies to both centrally partitioned datasets (batches or cross-validation folds) and inherently distributed data (federated clients or decentralized nodes), requiring only partition-local gradient statistics rather than raw data. We further introduce a conditional adaptation mechanism that combines FIM shift with KL divergence to detect significant distribution shifts and activates regularization only when necessary. Experiments on more than 40 datasets demonstrate consistent improvements under both natural and synthetic covariate shift. On fragmented batch and fold settings, reduces fragmentation-induced performance degradation by more than 20% and 25%, respectively. On seven federated learning benchmarks, it consistently outperforms FedAvg, FedProx, and SCAFFOLD by 3 -5 percentage points without requiring client-specific personalization. These results demonstrate that Fisher information provides an effective and unified mechanism for mitigating partition-induced covariate shift across both centralized and distributed learning.

Clinician burnout threatens patient safety, care quality, and workforce sustainability, especially in high-acuity ICUs. Existing detection approaches rely on retrospective surveys or coarse EHR metadata, limiting their ability to capture the evolution of burnout-related stress. We analyze 10,000 ICU discharge summaries from the MIMIC-IV database and propose a narrative-driven, weakly supervised framework for provider-level surveillance of burnout risk. Our approach integrates BioBERT-based sentiment modeling, lexical stress cues, latent topic structure, structured workload proxies, and temporal dynamics. In the absence of survey ground truth, we use a quantile-based ordinal labeling strategy to distinguish low, medium, and high burnout risk. A logistic regression classifier achieves an F1 score of 0.84 for conservative high-risk screening, while temporal features enable trajectory-based monitoring without degrading point-in-time performance. Specialty-specific analysis reveals elevated narrative stress indicators among Radiology, Psychiatry, and Neurology providers. ICU clinical narratives encode actionable, longitudinal signals for scalable burnout surveillance beyond static sentiment or metadata-only approaches.

Medical world models aim to learn a latent state of patient or organ physiology and a transition function that forecasts how that state evolves under interventions, supporting downstream tasks from imaging-based diagnosis to digital-twin treatment planning. Two failure modes threaten the reliability of such models in clinical deployment: (i) covariate shift, because training data are fragmented across hospitals, scanners, and time, so the feature distribution seen by the latent-dynamics predictor differs across fragments and from the distribution at deployment; and (ii) confidence misalignment, because multi-step forecasts are often overconfident exactly where clinical risk is highest. We argue that both problems admit a unified treatment via a single lightweight regularisation objective, CalTwin, which combines a Fisher-Information-based shift penalty adapted from our prior work on fragmented covariate-shift remediation with a Confidence Misalignment Penalty adapted from our prior work on calibrated vision-language classification , applied here to a GRU-based medical world model's latent transition predictor. We derive the combined objective, establish which proof steps transfer from the classification setting without modification and which require adaptation, and evaluate it on the PhysioNet 2019 Sepsis Challenge, treating the two hospital systems as sequential training fragments and the unseen system as an out-of-distribution test. CalTwin reduces OOD next-step latent-state MSE by 9.1% relative to the no-penalty baseline (FIM penalty alone accounts for 7.0%); the ECE reduction from the Confidence Misalignment Penalty is real but small (0.7% for CalTwin, 1.3% for CMP alone).

Metabolic disorders, particularly type 2 diabetes mellitus (T2DM), represent a significant global health burden, disproportionately impacting genetically predisposed populations such as the Pima Indians (a Native American tribe from south central Arizona). This study introduces a novel machine learning (ML) framework that integrates predictive modeling with gene-agnostic pathway mapping to identify high-risk individuals and uncover potential therapeutic targets. Using the Pima Indian dataset, logistic regression and t-tests were applied to identify key predictors of T2DM, yielding an overall model accuracy of 78.43

The global burden of Type II diabetes demands innovative strategies that combine predictive tools with targeted therapies. This study applies machine learning to the PIMA Indian dataset, identifying glucose, BMI, and age as key predictors, and integrates these with biological pathway mapping to support precision medicine. A novel methodological contribution is pathway mapping without gene-level data, linking clinical features to diabetes-related mechanisms like insulin signaling and PPAR pathways. This approach is particularly valuable for population datasets lacking molecular detail. In addition to established therapies (e.g., PPAR-based treatments, AMPK activators), the study explores emerging options such as dual GLP-1/GIP receptor agonists, novel AMPK activators, and sirtuin-related targets. Phytochemicals with multi-target effects are also evaluated. By bridging predictive modeling and biological insight, this research presents a framework for early detection and therapeutic innovation in T2DM, with special relevance for genetically predisposed populations such as the PIMA Indians.

Knowledge graphs (KGs) serve as powerful tools for organizing and representing structured knowledge. While their utility is widely recognized, challenges persist in their automation and completeness. Despite efforts in automation and the utilization of expert-created ontologies, gaps in connectivity remain prevalent within KGs. In response to these challenges, we propose an innovative approach termed “Medical Knowledge Graph Automation (M-KGA)". M-KGA leverages user-provided medical concepts and enriches them semantically using BioPortal ontologies, thereby enhancing the completeness of knowledge graphs through the integration of pre-trained embeddings. Our approach introduces two distinct methodologies for uncovering hidden connections within the knowledge graph: a cluster-based approach and a node-based approach. Through rigorous testing involving 100 frequently occurring medical concepts in Electronic Health Records (EHRs), our M-KGA framework demonstrates promising results, indicating its potential to address the limitations of existing knowledge graph automation techniques.

This poster presents an AI-powered intelligent tutoring system to enhance programming education through personalized and adaptive learning experiences. Using Generative AI, Machine Learning, and Natural Language Processing (NLP), our AI Tutor dynamically adjusts instructional content to match student needs, providing real-time feedback and a more interactive learning experience. The system integrates a chat-based interface that fosters seamless interaction, maintains learning context, and delivers personalized responses. Built on LangChain and Retrieval-Augmented Generation (RAG), the AI Tutor improves response accuracy and coherence, making it a powerful tool for programming education.

The rapid spread of fake news across social media and digital platforms poses a significant challenge to the credibility of public information. To address this issue, we proposed a framework that integrates GloVe-based semantic embeddings with additive attention mechanisms to enhance fake news detection (FND). The proposed approach integrates multiple components to capture both semantic and contextual features from news articles. Specifically, we employ pre-trained GloVe embeddings to transform raw textual data into meaningful vector representations that preserve word relationships. A Bidirectional LSTM network is used to learn sequential dependencies, while a scaled additive attention mechanism selectively emphasizes critical information within the text, enhancing model interpretability and accuracy. This multifaceted fusion allows the system to focus on linguistically significant patterns that distinguish real from fake content. We evaluate the proposed model on benchmark datasets validated by Politifact and Reuters, and the results demonstrate that the approach significantly outperforms existing state-of-the-art methods in terms of precision, recall, and overall classification accuracy. The proposed model provides a scalable solution for real-time fake news detection and can be extended to other domains of misinformation.

Metabolic disorders, such as type 2 diabetes mellitus (T2DM), pose substantial global health challenges, particularly affecting genetically predisposed populations like the Pima Indians. This study proposes a novel machine learning (ML) framework that integrates predictive modeling with gene-agnostic pathway mapping to identify high-risk individuals and discover therapeutic targets. Employing the PIMA Indian dataset, logistic regression and t-tests identified key predictors of T2DM, achieving a model accuracy of 78.43%. A pathway mapping strategy links these predictors to insulin signaling, AMPK, and PPAR pathways, facilitating mechanistic insights without molecular data. Therapeutic strategies encompass dual GLP-1/GIP receptor agonists, AMPK activators, SIRT1 modulators, and phytochemicals, validated through pathway enrichment. This framework advances precision medicine by providing interpretable, scalable solutions for early detection and targeted intervention in metabolic disorders and contributions:(1) An ML framework combining logistic regression and principal component analysis (PCA) for T2DM risk prediction; (2) A gene-agnostic pathway mapping method for mechanistic insights;(3) Novel therapeutic targets for high-risk populations.

Intensive Care Unit (ICU) patients often present with complex, overlapping signs of physiological deterioration that require timely escalation of care. Traditional early warning systems, such as SOFA or MEWS, are limited by their focus on single outcomes and fail to capture the multi-dimensional nature of clinical decline. This study proposes a multi-label classification framework to predict Care Escalation Triggers (CETs), including respiratory failure, hemodynamic instability, renal compromise, and neurological deterioration, using the first 24 hours of ICU data. Using the MIMIC-IV database, CETs are defined through rule-based criteria applied to data from hours 24 to 72 (for example, oxygen saturation below 90, mean arterial pressure below 65 mmHg, creatinine increase greater than 0.3 mg/dL, or a drop in Glasgow Coma Scale score greater than 2). Features are extracted from the first 24 hours and include vital sign aggregates, laboratory values, and static demographics. We train and evaluate multiple classification models on a cohort of 85,242 ICU stays (80 percent training: 68,193; 20 percent testing: 17,049). Evaluation metrics include per-label precision, recall, F1-score, and Hamming loss. XGBoost, the best performing model, achieves F1-scores of 0.66 for respiratory, 0.72 for hemodynamic, 0.76 for renal, and 0.62 for neurologic deterioration, outperforming baseline models. Feature analysis shows that clinically relevant parameters such as respiratory rate, blood pressure, and creatinine are the most influential predictors, consistent with the clinical definitions of the CETs. The proposed framework demonstrates practical potential for early, interpretable clinical alerts without requiring complex time-series modeling or natural language processing.

DNA-binding proteins (DBPs) play an important role in various cellular processes, including transcription, replication, and DNA repair. Identifying DBPs accurately from protein sequences is important for understanding gene regulation mechanisms and guiding biomedical research. However, traditional computational models often suffer from limited accuracy, heavy dependence on manual feature engineering, and poor generalization across diverse datasets. To overcome these limitations, this study proposed a unified evaluation framework that systematically compares machine learning (ML) and deep learning (DL) approaches for DBP prediction. Firstly, the procedure initiates with the extraction of discriminative features using k-mer frequency, amino acid composition (AAC), and Bidirectional Encoder Representations from Transformers (BERT). Secondly, we utilize ML algorithms, such as Random Forest (RF), Support Vector Machine (SVM), decision tree (DT), and XGBoost, to train on these feature sets. Thirdly, DL models, such as Long Short-Term Memory (LSTM), Convolutional Neural Networks (CNN), and Deep Neural Networks architecture, are developed to capture hierarchical and sequential patterns in protein sequences. Finally, models are evaluated using standard metrics, including accuracy, precision, recall, F1 Score, ROC AUC, and MCC. Experimental results show that DL models, particularly CNN-LSTM, outperform classical ML methods. A comparative analysis reveals that the proposed framework offers enhanced accuracy, robustness, and biological relevance in DBPs prediction.

In the published publication [...]

The rapid growth of online biomedical content has presented a notable challenge in delivering timely and precise semantic annotations. Semantic annotations play a crucial role in contextually indexing data, thereby enhancing search accuracy. This intricate process involves the utilization of multiple coded ontologies, requiring extensive technical expertise and domain knowledge. While automated ontologies face limitations in balancing accuracy and speed, expert knowledge generation is also scarce and expensive. In response to these conflicting challenges, we propose a socio-technical content generation and sharing approach named ‘Semantically,’ which actively involves biomedical experts and scientists. Additionally, ‘Semantically’ leverages schema.org for incorporating additional semantic tags to enhance search engine performance. The outcome is high-quality, machine-understandable content that not only facilitates fast and accurate searches but also instills trust due to the collaborative nature of the annotation process. ‘Semantically’ generated biomedical content was evaluated in two scenarios: (1) search based solely on initial-level annotations and (2) search incorporating additional expert-recommended annotations. ‘Semantically’ enhances the user search experience when compared to benchmark data. In the first scenario, the unigram to 5-grams strategy efficiently recognizes biomedical terms, resulting in high precision, recall, F1, and accuracy scores, all around 0.9. ‘Semantically’ code is openly accessible at https://github.com/bukharilab/Semantically.

Generative AI (GenAI) and LLMs have started to influence how teachers teach and studentslearn, including the ones in programming languages and tutoring. However, there have been debates onwhether AI could be beneficial to students’ learning or not. One way to see this issue is from the perspectives of thestudents. Therefore, this study explored how students perceive the use of AI in their education. The data wasgathered through interviews with 62 students and other stakeholders, such as instructors and IT specialists. The results showed that the students positively perceived using AI as a tutor. Moreover, this study suggests several things to consider when integrating AI tutors for programming. The findings reveal positive student perceptions regarding AI's potential within the teaching-learning process. Students envision AI tutors offering personalized assistance, adapting to individual learning styles, and providing immediate feedback, potentially augmenting their grasp of programming concepts. We applied Statistical analysis, machine learning, and natural language processing techniques such as PCA, t-SNE, LDA, and sentiment analysis.

The prevalence of diabetes presents a significant global health challenge, necessitating practical prognostic tools for timely intervention. Harnessing machine learning offers a promising avenue for accurate prediction, aiding in early detection and prevention. This study delves into the development of machine learning models for diabetes prognosis, leveraging the PIMA Indian dataset. Emphasizing the importance of early detection, the research underscores diabetes as a modifiable condition through lifestyle adjustments. By analyzing diverse healthcare data, including electronic health records and imagery, machine learning algorithms can unveil latent patterns crucial for timely diagnosis. This project has three objectives: 1) To find the significant features contributing to diabetes, 2) To develop a predictive model for diabetes prediction, and 3) To find an optimal model for diabetes prediction, comparing the performance of models. The research methodology encompasses a range of statistical and machine-learning algorithms applied to the PIMA dataset to identify optimal prediction methods. Our research underscores the pivotal role of glucose levels as the foremost predictor of diabetes across all models examined, reaffirming the significance of this metabolic marker in diabetes prognosis. The findings highlight logistic regression efficacy in feature extraction and prediction accuracy over random forest, K nearest neighbors (KNN), and deep neural (DL) networks, suggesting its potential for automated prognostic tools. Furthermore, omics data integration promises to enhance DL model performance, paving the way for robust diagnostic solutions.