Smoking remains a major public health concern, with smoking-related mortality increasing by approximately 3 ν = 0.95 provided the best fit to the observed data, achieving R^2 = 0.9678 , outperforming the corresponding integer-order model. A Chebyshev spectral collocation method was employed because of its high accuracy and rapid convergence in solving fractional differential equations. Numerical simulations showed that increasing recovery and quit rates significantly reduces smoking prevalence, while sensitivity analysis identified the smoking initiation rate as the most influential parameter. Cost-effectiveness analysis for public awareness campaigns and Nicotine Replacement Therapy (NRT) revealed that NRT alone was the most cost-effective intervention. A limitation of this study is that validation relies on a single historical dataset from 2001–2009.
Ovarian cancer is one of the most aggressive gynaecological cancers due in large part to its late detection and absence of highly specific biomarkers that would allow for reliable early diagnosis. Weighted Gene Co expression Network Analysis (WGCNA) and competing endogenous RNA (ceRNA) provides a well-established integrative approach for ovarian cancer biomarker discovery, however identified candidates have limited experimental verification. In this review, we evaluated four ovarian cancer studies employing integrated WGCNA–ceRNAapproaches. Several recurrent biomarkers namely MALAT1, H19, HOTAIR, NEAT1, XIST, LINC00152, the miR-200 family, and miR-21 are supported more strongly by targeted functional studies than by integrated network analyses, while their roles as ceRNA components remain unverified. Conversely, TP53, BRCA1, and BRCA2 are established, clinically translated biomarkers detected by genomic analysis that predates the WGCNA ceRNA strategy. Although these biomarkers possess robust cohort or functional support, experimentally confirmed ceRNA interactions remain rare, indicating that their biomarker relevance lacks a verified underlying ceRNA mechanism. Furthermore, limited hub gene overlap across independent cohorts raises concerns regarding WGCNA reproducibility. This review evaluates published WGCNA–ceRNA studies, highlighting data preprocessing, parameter selection, cohort variability, and experimental validation as major determinants of biomarker reliability. Future studies should prioritize standardized analytical pipelines, multicenter validation cohorts, experimental confirmation of predicted ceRNA mechanisms, and integration with single-cell, spatial transcriptomic, and epitranscriptomic data to improve biomarker reproducibility and accelerate their clinical translation.
Viral outbreaks combined with rapid emergence of mutated viruses have highlighted an urging need for accelerating antiviral drug discovery pipelines. Unfortunately, current drug discovery remains stuck to conventional methods which are slow especially during pandemics. In this article, we present a literature-based synthesis of an integrative machine learning (ML) guided QSAR framework that unifies ligand-based, structure-based, and systems biology approaches towards the aim of generating a host directed antiviral repurposing strategy. Moreover, a modern ML enhanced QSAR modeling strategy is proposed to target host directed therapeutics (HDTs), particularly the host kinase enzymes. The proposed framework integrates molecular descriptor modeling, ensemble learning methods (e.g., RF, gradient boosting), graph neural networks (GNNs), and multi-omics target prioritization to outline a predictive antiviral repurposing model. This structured workflow encompasses dataset assembly, descriptor generation, model training, virtual screening, and experimental validation as sequential stages to guide, rather than as a pipeline that has itself been built or independently validated here, translational deployment. The review is illustrated through a retrospective narrative synthesis of four independently published, clinically relevant repurposed HDTs, namely Baricitinib, Lapatinib, Bemcentinib, and Sunitinib. These published case studies, drawn from the primary literature, exemplify how AI/ML-enhanced QSAR and network-based approaches have been used elsewhere to identify active antiviral kinase inhibitors; they are presented here as illustrative evidence of feasibility of such a computational pipeline. Thus, the AI guided repurposing of host kinase inhibitors offers a systematically accelerated strategy to bridge the drug design gap, with the potential for faster therapeutic deployment against viral threats pending prospective, harmonized validation. This review describes a framework that combines artificial intelligence (AI), machine learning (ML), and Quantitative Structure-Activity Relationship (QSAR) modeling to speed up the search for new antiviral drugs. Instead of targeting the virus directly, the framework targets host cell proteins such as kinases, which many viruses hijack during infection, an approach also known as host-directed therapy (HDT). To show how this approach could work, we review four drugs that were originally developed for other diseases and later found to also fight viral infections: Baricitinib, Lapatinib, Bemcentinib, and Sunitinib. Each case was reported independently in the published literature, and we present them here as examples of what AI-assisted drug repurposing can achieve, not as proof that our specific framework has itself been built and tested. Accelerated therapeutic antiviral drug discovery pipelines are being a critical need due to viral outbreaks and rapid emergence of mutated viruses. Host Directed Therapeutics (HDTs) are new drug discovery strategies that can modulate specific host pathways essential for viral multiplication. The AI-HDT Framework is proposed to bridge the gap between the computational chemical prediction and clinical real-life application. The integration of multi-omics data such as phosphoproteomic data and transcriptomic data using the GNN models will help scientist to identify uniquely expressed host genes during the various episodes of viral infection.
Accurate preoperative assessment of parotid and breast tumors is critical for effective cancer management. However, the development of a unified diagnostic framework remains challenging due to complex anatomical contexts and the heterogeneous appearance of lesions in ultrasound imaging. These challenges motivate a unified framework that jointly models tumor delineation and malignancy stratification, allowing the two tasks to inform each other. We propose MulGM-Net, a unified multitask framework that incorporates contrastive learning to improve category-level feature representation in medical image analysis. The framework introduces a decision-based category memory with a gating mechanism that models the accumulation of category-level representations. This memory module enables the adaptive sharing of generic category information between segmentation and classification tasks, and is optimized using the r-IBS similarity metric to enhance feature discrimination. Performance was evaluated using a clinical parotid gland tumor dataset collected from a Stomatological Hospital and the public BUSI dataset. MulGM-Net achieved competitive performance relative to representative single-task and multi-task baselines, with the most pronounced gains in the parotid classification task, where soft-tissue contrast is limited. Specifically, on the clinical parotid dataset, it achieved a classification accuracy of 0.986 and a Dice similarity coefficient (DSC) of 0.85. On the BUSI dataset, it yielded competitive results with 0.94 accuracy and 0.84 DSC. Overall, MulGM-Net provides a multitask learning framework for addressing challenging cancer diagnosis scenarios. Experiments on two datasets suggest the potential utility of MulGM-Net for joint tumor delineation and malignancy assessment in head-and-neck and breast imaging, while the limited dataset size and the absence of external validation indicate that broader confirmation is still required.
Fatigue detection in elderly individuals is essential for preventing health complications, particularly fall-related incidents, in residential and home care settings. Electrocardiogram (ECG) signals offer a non-invasive means of capturing fatigue-related physiological changes, yet existing solutions often lack the lightweight, interpretable, and deployable design required for real-world elderly care. This study proposes a portable, low-cost ECG-based system for real-time fatigue detection, using an enhanced AD8232 module integrated with an ESP32 microcontroller, with the aim of supporting fall-risk prevention. A dataset of 6,304 ECG segments was collected from 100 elderly participants. SMOTE and Gaussian noise injection were applied to address class imbalance and improve robustness. GRU, LSTM, and RNN models were trained and validated using subject-wise splitting. The GRU model achieved the best performance, with 97.96
Metabolic syndrome (MetS) is recognized as an important clinical syndrome influencing population health. The rising prevalence of MetS globally highlights the need for accurate early prediction tools to identify at-risk individuals before disease onset. Machine learning (ML) can enhance predictive performance by identifying complicated clinical data correlations that cannot be identified by traditional statistical methods. This study aims to systematically assess the recent literature regarding the application of machine learning in predicting metabolic syndrome and to inform clinical implementation and guide the development of future models. This systematic review was performed following PRISMA guidelines. Eight databases including PubMed, CINAHL, Medline, IEEE Xplore, Science Direct, Scopus, ERIC, and Web of Science were comprehensively searched to identify relevant studies published between January 2020 and March 2026. Two independent researchers examined the articles, extracted data, and assessed the risk of bias utilizing the Prediction Model Risk of Bias Assessment Tool (PROBAST). Out of 9,249 identified articles, 44 studies met the inclusion criteria and were included in the review. The MetS prevalence in the population investigated was between 7.1
Hybrid deep learning in this review refers to a breast cancer diagnostic pipeline in which a deep neural network is the central predictor and at least one non-routine component materially changes representation learning, model selection, fusion, or inference. Such components include architectural coupling, feature fusion, decision-level ensembles, knowledge-guided reasoning, optimization-assisted search, or multimodal integration of imaging with clinical, textual, pathological, or omics data. This systematic review synthesizes hybrid deep learning methods across mammography, ultrasound, MRI, histopathology/whole-slide imaging, and clinically oriented multimodal settings. Thermography was part of the search scope, but no eligible journal article satisfied the inclusion criteria. A PRISMA 2020-guided protocol was used to search Scopus, PubMed, IEEE Xplore, and ScienceDirect (January 2019 to February 2026), followed by de-duplication, title/abstract screening, and full-text assessment. The final review included 62 journal studies (2020–2026). Of these, 42 met the reporting-completeness requirements for numerical summaries, while the remaining 20 contributed only to the qualitative synthesis. Quantitative percentages and comparisons therefore describe the reporting-complete subset and may not fully represent all eligible studies. A six-family taxonomy groups hybrid mechanisms into feature-level fusion, model-level hybrids, decision-level ensembles, neuro-fuzzy/knowledge-guided hybrids, optimization-assisted hybrids, and multimodal hybrids. Model-level hybrids appear most often (45.2
This study proposes a fractional model to enhance the understanding of Ebola transmission dynamics by integrating key socio-behavioural and environmental factors. Many existing Ebola models fail to incorporate memory and non-local effects in disease propagation. Although some fractional-order models have been developed, they mostly rely on assumed parameter values and omit some key epidemiological compartments. These limit the predictive reliability and comprehensiveness of the model. Hence, there is a gap in literature on data-driven fractional-order Ebola models that incorporate parameter estimation with a more extensive compartmental framework, which this study aims to fill. Specifically, we focus on the roles of fear, quarantine, isolation, and environmental contamination in shaping the trajectory of Ebola outbreaks. The proposed model is fitted to real data on reported cases of Ebola. We examined the positivity of the model and observed that it is well-posed. The Ebola reproduction number is obtained, which contains the effective contact rate between susceptible individuals and the pathogens in the environment; the effective contact rate between susceptible and infected individuals that causes Ebola infection; the percentage of quarantined individuals who move from this class to the susceptible class; the rate at which infectious individuals are isolated; and the recuperation rate of contagious individuals. The model is shown to be locally stable, and by employing the concepts of Ulam-Hyers, we notice that the fractional model also exhibits stability. According to the numerical simulation, the fractional model indicates that quarantine and isolation measures are crucial components of outbreak control strategies aimed at limiting contact between infected and susceptible individuals.
Infectious hematopoietic necrosis virus (IHNV) and viral hemorrhagic septicemia virus (VHSV) are major rhabdoviruses that cause severe economic losses in aquaculture. To address this, we employed a comprehensive immunoinformatics strategy to computationally design a chimeric multi-epitope DNA vaccine construct targeting both pathogens. Conserved regions of the viral glycoproteins (G) were analyzed to predict cytotoxic T lymphocyte (CTL), helper T lymphocyte (HTL), and B-cell epitopes with high antigenicity, non-allergenicity, and non-toxicity. The selected epitopes were assembled into a single construct incorporating β-defensin and PADRE as adjuvants to enhance potential immunogenicity. The designed construct exhibited favorable physicochemical properties. Structural modeling and validation supported the reliability of the 3D construct, and molecular docking suggested stable interactions with Toll-like receptors. Immune simulation analyses predicted the potential induction of humoral and cellular immune responses, including antibody production, cytokine release, and memory cell formation. Codon optimization further indicated the construct’s suitability for expression in both E. coli and eukaryotic systems. To the best of our knowledge, this is the first computationally designed chimeric multi-epitope DNA construct targeting both IHNV and VHSV simultaneously. Collectively, these findings provide a theoretical framework for the development of a dual-target vaccine. However, experimental validation in vitro and in vivo is required to evaluate safety, immunogenicity, and protective efficacy in fish models.
Combination therapy is an important clinical strategy for treating complex malignant diseases such as cancer, yet accurately identifying drug combinations with synergistic potential remains challenging. Existing computational models still face severe challenges in characterizing the deep spatial structures of drug molecules, filtering redundant noise in biological networks, and modeling complex entity interactions. To address these issues, this paper proposes MKASynergy, an adaptive drug synergy prediction method based on a mixture-of-experts kernel mechanism. Specifically, the method first constructs a dual-view drug feature enhancement mechanism that fuses the topological and fully connected graphs of drug molecules via a gating network, thereby effectively capturing long-range spatial dependencies. Simultaneously, a dynamic Top-K key protein screening strategy precisely retrieves the critical protein context utilizing the joint drug-cell line environment to avoid noise interference from large-scale networks. Building upon this, we introduce a mixture-of-experts (MoE) kernel mechanism as the interaction engine. It utilizes a dual-layer gated routing mechanism to adaptively invoke expert paths, achieving deep decoupling of multi-dimensional synergistic patterns. Extensive benchmark experiments demonstrate that MKASynergy achieves competitive predictive performance. Furthermore, visualization analysis confirms the model’s effectiveness in feature decoupling and helps interpret latent drug synergistic mechanisms.
The identification of immunogenic peptides is essential for the development of effective vaccines and immunotherapies. However, accurately identifying these peptides remains a significant challenge, as it involves a complex interplay of multiple biological factors. In this study, we present a comprehensive framework for immunogenic peptide identification using a dataset of 492 peptides collected from different sources. To capture diverse peptide characteristics, we apply feature engineering based on sequence composition and physicochemical properties. Our results show that the modified ν -SVM model, combined with L1 feature selection, achieves the best performance on an independent test set, with an AUC of 0.991 and an accuracy of 0.949. Further statistical analysis reveals that leucine-rich motifs are significantly enriched in immunogenic peptides. To enhance transparency, we integrate explainable AI techniques, including SHAP and LIME, to identify and interpret the most influential features contributing to model predictions. Overall, this study highlights the importance of combining robust feature extraction, effective feature selection, and interpretable machine learning to accurately predict peptide immunogenicity. The proposed framework provides biologically meaningful insights and offers a practical approach to support rational vaccine design.
Ensuring the reliability and safety of medical devices is critical for effective healthcare delivery, yet maintenance practices in many hospitals remain static and insufficiently adapted to device-specific risk and usage conditions. This study proposes a Knowledge Graph–driven framework for cognitive hospital equipment maintenance decision-making. The approach integrates a quantitative Preventive Maintenance Prioritization Rate Index with a formally structured ontology, the Medical Device Preventive Maintenance Ontology, to enable semantic knowledge representation and automated reasoning. The ontology models device characteristics, failure modes, risk factors, and maintenance actions, while rule-based inference and semantic queries support dynamic prioritization of preventive interventions. The framework was implemented using semantic web technologies and validated through a case study involving 659 medical devices at a university hospital. Results demonstrate the system’s ability to correlate risk levels, maintenance costs, device age, failure frequency, and operational context, thereby supporting transparent and explainable maintenance decisions. By combining quantitative prioritization with semantic reasoning, the proposed framework enhances maintenance scheduling, optimizes resource allocation, and improves equipment availability and patient safety. The study highlights the potential of knowledge graph–based systems to advance risk-aware and explainable maintenance decision-support strategies in healthcare environments.
GABA aminotransferase (GABA-AT) is a pyridoxal 5′-phosphate (PLP)-dependent enzyme that regulates γ-aminobutyric acid catabolism and remains an important therapeutic target for neurological disorders. Although porcine GABA-AT crystal structures have contributed to mechanistic understanding of PLP-dependent catalysis, the absence of a high-resolution experimental structure of the active human GABA-AT homodimer continues to limit direct structure-based inhibitor design. This mini-review critically examines how computational strategies have been used to address this structural gap in GABA-AT inhibitor discovery. Particular attention is given to the progression from static monomer-based docking toward dynamics-aware and dimer-resolved modelling approaches. Molecular docking, molecular dynamics simulations, MM/PBSA analysis, pharmacophore modelling, and emerging structure-prediction approaches are discussed in relation to their specific contributions and limitations for identifying GABA-AT inhibitors. The review emphasizes that reliable inhibitor design requires careful treatment of PLP-associated catalytic geometry, key recognition residues such as Arg192, Lys329, Ser53 and Gln301, and the functional relevance of the homodimeric architecture. Current evidence indicates that future GABA-AT inhibitor discovery should integrate validated human dimer models, receptor flexibility, PLP-dependent conformational dynamics, and experimentally supported computational predictions rather than relying solely on rigid monomeric docking models.
Triple-negative breast cancer (TNBC) remains one of the most aggressive breast cancer subtypes, characterized by poor prognosis and limited therapeutic options due to drug resistance. In this study, a network pharmacology-based multi-target drug discovery approach was applied to explore the therapeutic potential of Sclerocarya birrea bioactive moieties against TNBC. Initially, hub genes implicated in TNBC progression and drug resistance, including TP53, BCL2, IL6, CASP3, and JUN, were identified through protein–protein interaction and pathway enrichment analyses. Molecular docking revealed that several bioactive compounds from S. birrea, particularly Pseudolaric Acid derivatives, demonstrated high binding affinities across multiple targets, often exceeding those of the reference controls. Hierarchical clustering and correlation analysis further highlighted structural and functional similarities among the compounds, suggesting a consistent multi-target binding profile. Molecular dynamics simulations confirmed the stability of key hub gene–compound complexes, with reduced RMSD, RMSF, and favorable solvent-accessible surface area (SASA) compared with controls, indicating enhanced conformational stability upon ligand binding. Per-residue energy decomposition pinpointed critical residues contributing to binding, while thermodynamic profiling showed that the binding was primarily driven by van der Waals and electrostatic interactions, with Pseudolaric Acid H (BCL2), DMHCA (CASP3), and Pseudolaric Acid B (JUN) achieving significantly more favorable free energies than controls. These findings suggest that S. birrea bioactive compounds exert their therapeutic effect through a synergistic, multi-target mechanism, enhancing apoptotic regulation while overcoming resistance-associated pathways. This integrative in silico study provides a mechanistic blueprint for repurposing S. birrea phytoconstituents as potential multi-target therapeutics for TNBC. Pseudolaric Acid H, DMHCA, and Pseudolaric Acid B were identified as promising compounds that exhibited favorable interactions with key proteins associated with TNBC drug resistance. These findings suggest their potential to modulate drug resistance-related pathways and warrant further experimental validation. The results not only highlight their capacity to reverse drug resistance but also underscore their promise as scaffolds for further development in anti-TNBC drug discovery.
Predictive modelling of tabular datasets has often been hindered by intrinsic sparsity and distributional skew, which are not well-suited to conventional machine learning (ML) architectures and prevent them from capturing higher-order dependencies across heterogeneous patient profiles. We propose a topology-aware graph synthesis pipeline that converts tabular clinical records into a non-Euclidean relational space. By harnessing cosine-similarity-based edge projection and class-augmented connectivity, the framework successfully encodes both local node attributes and global inter-patient structural semantics. We conduct a comparative study of Graph Convolutional Networks (GCNs) and GraphSAGE against a range of traditional ML baselines across 10 multi-institutional heart and lung disease datasets. Our findings reveal that the inductive learning capability of GraphSAGE achieves a competitive F1- score across diverse clinical datasets, compared with conventional ML models used as benchmarks. In particular, the proposed GNN framework demonstrates substantial improvements in minority-class sensitivity on challenging imbalanced datasets where traditional ML models exhibit severe degradation in recall and F1-score. Furthermore, calibration analysis via Expected Calibration Error (ECE) and t-SNE manifold visualisation confirms that separable learned feature representations improve performance. This work strives to bridge the gap between raw clinical data and actionable, calibrated decision support.
Extreme class imbalance hinders clinical text classification, particularly when rare cases have diagnostic or prognostic significance. We benchmark 336 pipelines varying vectorization (BoW, TF–IDF), resampling (12 methods + baseline), and classifiers (15 models) on 49,035 French radiology reports (164 minority; prevalence ≈ 0.33% ). Oversampling—especially BorderlineSMOTE/SMOTE—consistently improves minority-class F_1 ; naïve undersampling degrades it. The top configuration pairs TF–IDF with Stacking and boundary-cleaning undersampling (OSS/NCR), reaching F_1=0.727 , performing comparably to the best oversampling pipeline (RandomOverSampler + LightGBM, F_1=0.717 ) and a strong no-resampling baseline (BoW + Voting, F_1=0.706 ). Sampler and classifier failures at this prevalence are also documented. These results establish robust starting configurations for extreme-imbalance settings and a transparent evaluation protocol to support future work.
Kidney stones are a very common urological condition resulting in significant healthcare costs. It requires an accurate and timely diagnosis for the treatment to be effective. While CNNs exhibit high diagnostic accuracy, their clinical deployment is often hindered by high computational overhead and its "black box" nature resulting in lack of interpretability. This study benchmarks five pre-trained CNN models (ResNet50, InceptionV3, MobileNetV2, DenseNet121, EfficientNet-B0) for classifying kidney stone cases versus normal cases. Evaluating across a clinically balanced distribution of 1,787 normal and 1,577 stone-positive scans, InceptionV3 achieved the highest performance across all metrics: accuracy 98.66
The genetic landscape and interplay between Parkinson’s disease (PD) and epilepsy have yet to be discovered. Understanding the complex interplay between the genes involved in these disorders is crucial for developing therapeutic strategies. This study aimed to identify common hub genes of PD and epilepsy to identify appropriate drugs. The GSE202101, GSE32534, GSE7486, GSE7621, GSE20141, and GSE49036 datasets were utilized to obtain microarray data from patients with PD and epilepsy. The data underwent preprocessing and batch effect removal. Weighted gene co-expression network analysis (WGCNA) was used to identify hub genes. Finally, the hub genes were validated using the GSE7142 and GSE20295 datasets. Network analysis was performed on the detected genes and drugs retrieved from the DrugBank. A total of 84 microarray samples, including 26 epilepsy, 31 PD, and 27 normal samples, were analyzed. By applying a module connectivity of more than 0.8 and a clinical trait of more than 0.2, 167 hub genes were identified. Using two datasets for validation highlighted the CBS, RPS6KA5, SDC4, GNA13, PDK3, ARHGEF9, CDKN3, and SGK1 genes as potential hub genes. Using STRING and Drug Bank, associated genes and drugs were retrieved, resulting in 47 FDA-approved and 39 non-approved drugs being identified. Among the proposed drugs, magnesium supplements, memantine, felbamate, imidazole, talampanel, and radiprodil are hypothesized to modulate the pathogenicity of PD and epilepsy. Moreover, further clinical investigations are needed to validate their precise roles and therapeutic potential.
The targeted delivery of anticancer chimeric proteins as a therapeutic agent improves drug effectiveness. Research has concentrated on developing novel treatments by exploiting the properties of cytokines, notably interleukins, to suppress the proliferation of cancerous cells. IL-37 is a cytokine that has anti-inflammatory properties, and it limits the development of hepatocellular carcinoma by impeding cell growth, invasion, and angiogenesis of tumor cells as well as strengthening the body’s immune defenses against cancer while Bombinin (BO1) is a peptide that shows anti-HCC activity by disrupting cell-cycle regulation and inducing apoptosis in liver cancer cells with minimal effects on normal hepatocytes. This study computationally designed a fusion protein IL 37-BO1 by combining the targeting IL-37 with the anti-cancer peptide BO1 via a rigid linker to focus treatment directly on cancerous cells. The secondary structure, tertiary structure, and the physiochemical properties of the engineered fusion protein IL 37-BO1 were estimated through GOR IV, trRosetta, and ProtParam respectively. Validation of protein quality and 3D structure was confirmed using ProSa-web, ERRAT2 and RAMPAGE servers. Next, the IL-37–BO1 fusion protein was docked with the IL18Rα–IL1R8 heterodimer receptor to evaluate its binding characteristics. The docked complex was subsequently analyzed for intermolecular interactions and subjected to molecular dynamics simulation to further investigate its structural stability and dynamic behavior. In-silico study showed that the newly designed possesses a basic composition and a molecular weight of 24 kDa. The ERRAT value of 97.52
Identifying disease-relevant disruptions in gene regulatory networks requires computational frameworks that move beyond differential expression analysis toward systematic modeling of interaction loss across heterogeneous multi-cohort datasets. We present a computational pipeline that systematically detects, filters, and validates co-expression interaction dissolution across four TCGA solid tumor cohorts (BRCA, LUAD, HNSC, and STAD), integrating multi-metric quality control, DEG-constrained network inference, double-threshold Pearson filtering benchmarked against GeneMANIA, and cross-cohort consensus ranking. We implement a four-step pipeline: 1) Network construction using a double-threshold filtering algorithm, optimized through benchmarking with GeneMANIA; 2) Multivariate stratification (molecular subtypes and anatomical regions) in four TCGA cohorts; 3) A hierarchical consensus intersection algorithm to identify conserved lost interactions; and 4) Development of a co-expression score based on z-score products for integration into Cox survival models. The pipeline identified a robust core of 18 conserved lost interactions across solid tumors. Survival analysis suggests that pairwise co-expression scores derived from dissolved regulatory links stratify overall survival with hazard ratios up to 2.01, providing prognostic value beyond what single-gene expression levels capture. The proposed framework is generalizable to any multi-cohort RNA-seq compendium and positions co-expression dissolution as a computationally tractable, clinically informative complement to standard differential expression pipelines.
