ABSTRACT VGG‐16 and Inception are widely used CNN architectures for image classification, but they face challenges in target categorization. This study introduces B4‐GraftingNet, a novel deep learning model that integrates VGG‐16's hierarchical feature extraction with Inception's diversified receptive field strategy. The model is trained on the OCT‐CXR dataset and evaluated on the NIH‐CXR dataset to ensure robust generalization. Unlike conventional approaches, B4‐GraftingNet incorporates binary particle swarm optimization (BPSO) for feature selection and grad‐CAM for interpretability. Additionally, deep feature extraction is performed, and multiple machine learning classifiers (SVM, KNN, random forest, naïve Bayes) are evaluated to determine the optimal feature representation. The model achieves 94.01% accuracy, 94.22% sensitivity, 93.36% specificity, and 95.18% F1‐score on OCT‐CXR and maintains 87.34% accuracy on NIH‐CXR despite not being trained on it. These results highlight the model's superior classification performance, feature adaptability, and potential for real‐world deployment in both medical and general image classification tasks.

BACKGROUND AND OBJECTIVE: In everyday clinical practice, medical decision is currently based on clinical guidelines which are often static and rigid, and do not account for population variability, while individualized, patient-oriented decision and/or treatment are the paradigm change necessary to enter into the era of precision medicine. Most of the limitations of a guideline-based system could be overcome through the adoption of Clinical Decision Support Systems (CDSSs) based on Artificial Intelligence (AI) algorithms. However, the black-box nature of AI algorithms has hampered a large adoption of AI-based CDSSs in clinical practice. In this study, an innovative AI-based method to compress AI-based prediction models into explainable, model-agnostic, and reduced decision support systems (NEAR) with application to healthcare is presented and validated. METHODS: NEAR is based on the Shapley Additive Explanations framework and can be applied to complex input models to obtain the contributions of each input feature to the output. Technically, the simplified NEAR models approximate contributions from input features using a custom library and merge them to determine the final output. Finally, NEAR estimates the confidence error associated with the single input feature contributing to the final score, making the result more interpretable. Here, NEAR is evaluated on a clinical real-world use case, the mortality prediction in patients who experienced Acute Coronary Syndrome (ACS), applying three different Machine Learning/Deep Learning models as implementation examples. RESULTS: NEAR, when applied to the ACS use case, exhibits performances like the ones of the AI-based model from which it is derived, as in the case of the Adaptive Boosting classifier, whose Area Under the Curve is not statistically different from the NEAR one, even the model's simplification. Moreover, NEAR comes with intrinsic explainability and modularity, as it can be tested on the developed web application platform (https://neardashboard.pythonanywhere.com/). CONCLUSIONS: An explainable and reliable CDSS tailored to single-patient analysis has been developed. The proposed AI-based system has the potential to be used alongside the clinical guidelines currently employed in the medical setting making them more personalized and dynamic and assisting doctors in taking their everyday clinical decisions.

MOTIVATION: Biomarker discovery is one of the most frequent pursuits in bioinformatics and is crucial for precision medicine, disease prognosis, and drug discovery. A common challenge of biomarker discovery applications is the low ratio of samples over features for the selection of a reliable not-redundant subset of features, but despite the development of efficient tree-based classification methods, such as the extreme gradient boosting (XGBoost), this limitation is still relevant. Moreover, existing approaches for optimizing XGBoost do not deal effectively with the class imbalance nature of the biomarker discovery problems, and the presence of multiple conflicting objectives, since they focus on the training of a single-objective model. In the current work, we introduce MEvA-X, a novel hybrid ensemble for feature selection (FS) and classification, combining a niche-based multiobjective evolutionary algorithm (EA) with the XGBoost classifier. MEvA-X deploys a multiobjective EA to optimize the hyperparameters of the classifier and perform FS, identifying a set of Pareto-optimal solutions and optimizing multiple objectives, including classification and model simplicity metrics. RESULTS: The performance of the MEvA-X tool was benchmarked using one omics dataset coming from a microarray gene expression experiment, and one clinical questionnaire-based dataset combined with demographic information. MEvA-X tool outperformed the state-of-the-art methods in the balanced categorization of classes, creating multiple low-complexity models and identifying important nonredundant biomarkers. The best-performing run of MEvA-X for the prediction of weight loss using gene expression data yields a small set of blood circulatory markers which are sufficient for this precision nutrition application but need further validation. AVAILABILITY AND IMPLEMENTATION: https://github.com/PanKonstantinos/MEvA-X.

The Josephin Domain (JD), i.e. the N-terminal domain of Ataxin 3 (At3) protein, is an interesting example of competition between physiological function and aggregation risk. In fact, the fibrillogenesis of Ataxin 3, responsible for the spinocerebbellar ataxia 3, is strictly related to the JD thermodynamic stability. Whereas recent NMR studies have demonstrated that different JD conformations exist, the likelihood of JD achievable conformational states in solution is still an open issue. Marked differences in the available NMR models are located in the hairpin region, supporting the idea that JD has a flexible hairpin in dynamic equilibrium between open and closed states. In this work we have carried out an investigation on the JD conformational arrangement by means of both classical molecular dynamics (MD) and Metadynamics employing essential coordinates as collective variables. We provide a representation of the free energy landscape characterizing the transition pathway from a JD open-like structure to a closed-like conformation. Findings of our in silico study strongly point to the closed-like conformation as the most likely for a Josephin Domain in water.