Introduction: Recent advancements in biomedical research have been significantly accelerated by Artificial Intelligence (AI), Machine Learning (ML), and computational modeling, enabling breakthroughs that were previously hidden or unattainable with conventional approaches. AI and ML have evolved from supportive technologies to central components of biomedical discovery. Methods: A comprehensive literature review was conducted in PubMed, Scopus, Web of Science, IEEE Xplore, and Google Scholar, covering studies published between January 2015 and March 2026. Relevant articles focusing on AI, machine learning, deep learning, computational modeling, precision medicine, medical imaging, and clinical diagnostics were systematically screened and analyzed. Results: The data and information reviewed confirm that ML has reached expert-level benchmarks in image-heavy diagnostics and individualized risk stratification. However, high accuracy in a laboratory setting does not always translate to clinical success. Discussion: The reviewed evidence demonstrates that AI and ML are increasingly being applied in healthcare by enhancing diagnosis, prognosis, and personalized treatment. However, challenges related to interpretability, data quality, clinical validation, and ethical implementation must be addressed to ensure reliable, transparent, and widespread adoption in routine clinical practice. Conclusion: The findings indicate that for these systems to transform healthcare delivery, the emphasis should shift from solely predictive accuracy to transparency and clinical validation. The future of this field depends on creating interpretable models that clinicians trust, ensuring that technological advancements improve patient outcomes rather than merely achieving higher accuracy in laboratory tests.
Introduction: Circular RNAs (circRNAs) are crucial regulators in Gastric Cancer (GC). Although RNA editing affects circRNA biogenesis, its specific role in GC-associated circRNA dysregulation remains unclear. This study characterizes the interaction between RNA editing and circRNAs, elucidating its impact on GC progression. Methods: Integrating multiple transcriptome datasets, we analyzed RNA editing's impact on circRNAs via correlation assessments, RNA-Binding Protein (RBP) motif analysis, and miRNA interaction profiling. Furthermore, we constructed RNA editing-mediated competing endogenous RNA (ceRNA) networks and evaluated drug resistance to identify novel candidate biomarkers and therapeutic targets. Results: The data showed that GC samples have significantly higher A-to-I RNA editing levels but much lower circRNA expression when compared to matched normal tissues. We found that ADAR is overexpressed in GC and appears to drive this dysregulation. Furthermore, RNA editing events in the introns flanking circRNAs may influence how these molecules are made by changing RBP interactions. Using ceRNA network analysis together with drug resistance data, we pinpointed six hub genes and eight A-to-I editing sites that could serve as prognostic markers or therapeutic targets. Discussion: Elevated ADAR likely represses circRNA biogenesis, rewiring downstream regulatory networks. The identified editing sites offer high stability for noninvasive biomarker detection, though in vivo and in vitro experimental validations are required to confirm these computational predictions. Conclusion: This study systematically elucidates the link between RNA editing and circRNAs in GC, providing valuable insights into disease pathophysiology and highlighting novel targets for developing more effective treatment strategies.
Introduction: RNA-Binding Proteins (RBPs) play critical roles in post-transcriptional gene regulation by modulating RNA splicing, transport, stability, and translation. Dysfunction of RBPs is associated with various diseases, including cancer, neurodegenerative disorders, and viral infections. However, existing computational methods for RBP prediction often rely on single-modality features or direct feature concatenation, limiting their ability to capture the complex interactions between RNA and proteins. Methods: We propose AttentionRBP, a multimodal framework that integrates protein-language-model embeddings, structural information, and physicochemical properties for RBP prediction. To improve cross-modal feature fusion, a Dropout-based sparse cross-attention mechanism was designed to suppress noisy interactions and enhance representation robustness. Results: Experimental results demonstrated that AttentionRBP outperformed several baseline methods in terms of Accuracy, MCC, AUC, and F1-score. Ablation studies confirmed the effectiveness of each modality and the contribution of the sparse cross-attention strategy. In addition, SHAP-based interpretability analysis indicated that cross-modal features contributed substantially to model predictions, while structural information complemented sequence-based embeddings. Discussion: The proposed multimodal framework effectively integrates heterogeneous biological information, thereby improving the robustness and interpretability of RBP prediction. The sparse crossattention mechanism also reduces the influence of irrelevant cross-modal interactions during feature fusion. Conclusion: AttentionRBP provides an effective framework for multimodal RBP prediction and shows promising generalization performance on unseen data. A publicly accessible web server was also developed to support batch submission and multi-species prediction for efficient RBP identification.
Introduction: Colorectal cancer (CRC) is a leading cause of cancer-related mortality worldwide, largely attributable to limitations in early detection and patient stratification. Although N6-methyladenosine (m6A) modification has emerged as a critical epitranscriptomic regulator in cancer biology, its role in driving molecular heterogeneity and immune microenvironment remodelling during colorectal pathogenesis remains poorly characterised. Materials and Methods: This study integrated multi-transcriptomic datasets from normal colon, adenoma, and CRC tissues to identify m6A-related differentially expressed genes (m6A-RDEGs). Using systematic bioinformatics analyses and machine learning approaches, we developed a six-gene signature to distinguish CRC from precancerous stages. Furthermore, consensus clustering based on m6ARDEG expression patterns was applied to define molecular subtypes. Results: A robust six-gene signature effectively differentiated CRC from earlier stages. Consensus clustering revealed two distinct molecular subtypes with markedly different immune microenvironment profiles. Subtype 1 exhibited an immunosuppressive phenotype, characterized by significant M2 macrophage infiltration and enrichment of TGF-β signaling pathways, while Subtype 2 displayed divergent immunological characteristics. Functional analyses linked these subtypes to key pathways, including ferroptosis and HIF-1 signaling. Discussion: Our findings establish a novel link between m6A-mediated molecular subtyping and immune microenvironment remodeling in colorectal pathogenesis. The identified signature and subtypes provide not only diagnostic biomarkers but also insights for developing subtype-specific immunotherapeutic strategies. Conclusion: This study elucidates the role of m6A modification in CRC heterogeneity and immune microenvironment regulation, offering potential tools for improved stratification and targeted immunotherapy in colorectal cancer.
Introduction: The COVID-19 pandemic has highlighted the importance of understanding host-virus interactions, particularly the role of interferon (IFN) signaling pathways in viral pathogenesis. Methods: This study employed a comprehensive machine learning framework to analyze Perturb-seq data from SARS-CoV-2-infected Calu-3 respiratory epithelial cells to identify key host factors associated with viral infection under various IFN signaling conditions. We implemented two complementary feature selection strategies: Raw Feature Ranking using ten algorithms combined with IFS technology, and Autoencoder-based Feature Reconfiguration for dimensionality reduction, followed by ranking with the same algorithms combined with IFS. We compared expression differences between infected and uninfected cells in both IFN-knockout and control conditions to investigate distinct responses to SARS-CoV-2 infection. Results: In control conditions, Raw Feature Ranking identified important genes including FOS, JUN, B2M, PPIA, and MALAT1, known for their roles in viral pathogenesis, antigen processing, and inflammatory response. Under IFN knockout conditions, numerous mitochondrial genes (MT-ND3, MTCO1, MT-ND4L, MT-ND4, MT-CO3, MT-ATP6, MT-CO2, MT-CYB) emerged as key discriminatory features, suggesting mitochondrial dysfunction as a crucial mechanism when IFN signaling is compromised. Autoencoder-based methods showed close alignment with raw feature ranking, identifying 63 and 64 features in the IFN and Control groups, respectively, with 46 and 32 overlapping genes. Additional key factors included EEF1A1, KRT8, KRT19, MIF, SPINT2, and S100A10. Discussion: Enrichment analysis revealed consistent enrichment of oxidative phosphorylation and thermogenesis pathways across conditions, with IFN knockout cells specifically enriched for coronavirus disease pathways. Conclusion: This comprehensive approach provided robust biomarker identification and molecular insights into SARS-CoV-2 immune evasion mechanisms.
