Background Non-alcoholic fatty liver disease (NAFLD) is rapidly becoming the most prevalent liver disease worldwide. The accurate identification of NAFLD patients and understand the molecular mechanism of initiating and driving disease progression remain limited. Objectie This study aims to develop a robust and accurate computational model for identifying NAFLD patients using mRNA expression profiles, and to explore the underlying molecular mechanisms that drive disease initiation and progression. Methods We introduce an adversarial deep neural network (ADNN) to identify NAFLD patients based on mRNA expression profiles, which uses adversarial learning techniques to take advantage of existing mRNA expression profiles in the GEO database. We compare the ADNN model with some traditional machine learning methods, such as random forest (RF), adaptive boosting (AdaBoost). Results The area under the receiver operating characteristic curve (AUROC) of the ADNN model is 0.982, which is much higher than the machine learning models such as RF, SVM and deep neural networks. Meanwhile, by combining the ADNN model with the DeepLIFT algorithm, we calculate the importance score of genes and find that the DeepLIFT algorithm can identify genes that don’t differ at the level of mRNA, but interact with other important genes. Such as GSTM1, with the greatest importance score and has no difference between control and NAFLD, but interacts with IGF1 and MEP1B. Conclusion Therefore, the ADNN model can not only identify NAFLD patients from healthy controls, but also help us better understand the disease mechanism of NAFLD.
Background Diabetes is a chronic disease also known as a non-communicable disease that cannot be completely cured but can be controlled through medication, or by maintaining a balanced diet and regular physical activity. Many drugs are available to control diabetes, but these drugs need to be carefully prescribed by the experienced physicians because the drugs may have side effects as well as adverse reactions. Objective We propose an ontology-based safe drug prescription method for diabetes that assists physicians in prescribing appropriate drugs by considering their side effects and interactions with other drugs, diseases and foods. Methods The proposed framework was implemented by creating an ontology by integrating drug-related information which includes side effects, drug interactions with other drugs, diseases and foods from different sources. We developed a Java-based user interface through which information on drug side effects, drug interactions and uses of the drugs can be retrieved using SPARQL queries. Results The inferred results using SWRL rules and SPARQL queries were converted to a questionnaire that was validated by four physicians, and achieved the highest agreement score of 92.8%. The ontology-based model was also evaluated for consistency on HermiT and Pellet reasoners. Conclusion Patients with diabetes frequently have comorbidities requiring multiple medications, which increases the risk of adverse drug reactions. The constructed ontology provides a knowledge base to support physicians in identifying drug side effects and interactions. The Ontology incorporates knowledge related to causes of diabetes including impaired insulin secretion, insulin resistance and other causes based on obesity-related factors.
Background Cardiovascular disease (CVD) is one of the most common causes of death globally. Early detection of CVD and timely treatment for preventing further health complications depend on several factors. Those variables include clinical data, redundant features, and existing machine learning models or deep learning models that do not achieve reliable prediction performance. Objective The primary objective of this research work is to develop an intelligent framework for predicting cardiovascular diseases that classifies the diseases accurately, reduces feature redundancy and enhances efficiency through optimizing feature selection. Methods The proposed Piranha-Based Feature Subset Selection with Constructive Sequential Neural Network (PFSS-CSNet) framework first applies Min-Max normalization to standardize the clinical data. The PFSS algorithm employs feature subset selection to identify the most informative clinical features. In addition, Bayesian optimization is used to determine the optimal hyperparameters of CSNet. The proposed framework is evaluated using the publicly available UCI Heart Disease dataset. Results The accuracy of the proposed PFSS-CSNet framework is 99.70%, the precision estimate is 99.50%, the recall is 99.80%, the specificity assessment is 99.60%, the F1-score is 99.60%, the G-measure is 99.70%, and the AUC is evaluated as 0.997. In comparative experiments, the proposed framework consistently outperforms existing machine learning classifiers in predicting cardiovascular diseases. Conclusion The incorporation of PFSS, Bayesian hyperparameter optimization, and CSNet provides a powerful and computationally efficient framework for predicting cardiovascular diseases. The proposed method achieves accurate prediction performance on the benchmark dataset and can aid clinical decision-making. Nonetheless, routine clinical adoption requires further verification using multi-center external clinical datasets.
Background Accurate perioperative risk prediction remains challenging because of class imbalance, cohort heterogeneity, unstable probability estimates and limited interpretability. Objective To develop and externally evaluate CardioMonoStack, an ensemble architecture integrating heterogeneous base learners, branch-level monotonicity priors, stacked meta-learning and feature-dependent residual logit correction. Methods We conducted a multicentre retrospective cohort study of 1911 patients aged ≥65 years undergoing total hip arthroplasty across five healthcare databases. CardioMonoStack was developed using MIMIC-III and MIMIC-IV, validated in INSPIRE, and externally tested in eICU and MOVER. Performance was assessed using AUROC, AUPRC, Brier score, threshold-dependent classification measures, calibration plots and decision curve analysis. Interpretability was examined using SHAP, interaction and individual conditional expectation analyses. Results CardioMonoStack achieved the highest AUROC point estimates among evaluated models in eICU (0.790; 95% CI 0.711–0.865) and MOVER (0.866; 95% CI 0.761–0.949), with AUPRC values of 0.569 (95% CI 0.415–0.717) and 0.571 (95% CI 0.328–0.792), respectively. Calibration and net benefit varied across cohorts and comparators. Omission of L3 increased Brier-score and log-loss point estimates in both external cohorts; paired confidence intervals supported these differences in MOVER but were inconclusive in eICU. SHAP analyses prioritised blood urea nitrogen, creatinine, age, and metabolic and demographic factors. Conclusion CardioMonoStack is a candidate framework for estimating in-hospital heart-failure risk after total hip arthroplasty. Despite promising external discrimination, prospective validation, local recalibration, and clinical impact evaluation are required before implementation.
Background Anemia remains a major public health concern among adolescent girls in Ethiopia, adversely affecting physical growth, cognitive development, educational attainment, and future reproductive health. Accurate prediction of anemia severity and identification of key risk factors are essential for supporting targeted prevention and intervention strategies. Objective This study aimed to develop and compare tree-based ensemble machine learning models for multi-class prediction of anemia severity among Ethiopian adolescent girls using nationally representative survey data. Methods This study analyzed Ethiopian Demographic and Health Survey (EDHS) data from 2011 and 2016, including 15,952 adolescent females aged 15–19 years. Four tree-based ensemble machine learning algorithms (Random Forest, Extra Trees, CatBoost, and XGBoost) were developed and compared with a baseline Multinomial Logistic Regression model. Following TRIPOD-AI recommendations, data preprocessing included missing value imputation, categorical feature encoding, Recursive Feature Elimination (RFE) for feature selection, and Synthetic Minority Oversampling Technique (SMOTE) to address class imbalance. To prevent data leakage, all preprocessing and SMOTE procedures were applied exclusively within the training folds. Models were trained using stratified 10-fold cross-validation on 70% of the data and evaluated on an independent 30% hold-out test set (n = 4786) using accuracy, precision, recall, F1-score, and macro-average ROC-AUC. Model interpretability was assessed using global and local SHapley Additive exPlanations (SHAP). Results Random Forest achieved the highest predictive performance, with an accuracy of 85.60% and a macro-average ROC-AUC of 94.84%, followed by Extra Trees (85.10% accuracy; 94.48% ROC-AUC) and CatBoost (83.80% accuracy; 93.08% ROC-AUC). The baseline Multinomial Logistic Regression model achieved 71.20% accuracy and an ROC-AUC of 81.50%. SHAP analysis consistently identified geographic region, altitude, educational level, household wealth index, and household size as the most influential predictors of anemia severity. Conclusion Tree-based ensemble machine learning models, particularly Random Forest and Extra Trees, substantially outperformed conventional multinomial logistic regression for multi-class prediction of anemia severity among Ethiopian adolescent girls. Their strong predictive performance and SHAP-based interpretability highlight their potential to support evidence-based screening, risk stratification, and targeted public health interventions in resource-limited settings.
