Autophagy is a conserved cellular clearance pathway that supports homeostasis by removing damaged or superfluous intracellular components. Within microglia, autophagy is emerging as a regulator of key processes that modify neurodegeneration, including phagocytosis, cytokine secretion, and senescence. Many studies that have examined the effect of disrupted autophagy on microglial functions have used genetic knockouts of the machinery required to conjugate microtubule-associated light chain 3 (LC3) to the autophagic membrane. However, much of this molecular machinery is also required for a set of distinct but related pathways known as the conjugation of ATG8s to single membranes (CASM). CASM includes processes of particular importance in microglia, such as LC3-associated phagocytosis and LC3-associated endocytosis. It is thus not clear which of the effects of the disruption of LC3 conjugation in microglia are attributable to the loss of autophagy or the loss of CASM function. In this review, we describe the mechanisms of autophagy and CASM and highlight the effects of the loss of these pathways on key microglial processes relevant to brain ageing and neurodegenerative diseases. We discuss recent literature that has revealed the effects of ageing and neurodegeneration on microglial autophagy, and the effects of microglial autophagy and/or CASM disruption on key microglial functions such as phagocytosis, cytokine secretion, and senescence. Finally, we discuss the potential therapeutic implications of these findings for neurodegeneration and highlight key unanswered questions for future research.

Aims: To develop a robust and clinically feasible framework for cancer survival prediction using only histopathology images while leveraging transcriptomic knowledge during training.

Aims: To develop a robust and clinically feasible framework for cancer survival prediction using only histopathology images while leveraging transcriptomic knowledge during training.

Methods: The study proposed Adaptive Multi-modality Knowledge Distillation (AMKD), a framework designed to transfer complementary molecular-level information from transcriptomic data to pathology-based models. The AMKD framework consists of two essential elements. First, a gene-guided pathology enhancement module is designed to inject genomics-aware information from a multimodal teacher into pathology features. Second, an adaptive redundancy reduction loss is introduced to regulate knowledge distillation by accounting for prediction discrepancies between teacher and student models. This design allows the student model to retain biologically meaningful knowledge during training and remain effective with only histopathology data at inference.

Results: Comprehensive experiments on four The Cancer Genome Atlas (TCGA) cancer cohorts demonstrate that AMKD achieves state-of-the-art survival prediction performance, with an average concordance index (C-index) of 0.669, surpassing both unimodal pathology-based and fully multi-modal approaches. External validation on the independent Clinical Proteomic Tumor Analysis Consortium head and neck squamous cell carcinoma (CPTAC-HNSC) cohort further confirms the robustness and cross-dataset generalization ability of the proposed framework. Ablation studies further confirm the effectiveness of each proposed component in enhancing cross-modal knowledge transfer and improving generalization under incomplete modality scenarios.

Conclusion: The proposed AMKD framework provides a clinically practical solution for robust cancer survival analysis when transcriptomic data are unavailable. By adaptively distilling multi-modal knowledge into a pathology-based model, AMKD bridges the gap between research and clinical applicability, enabling scalable and cost-effective prognostic prediction in real-world settings.

Self++ is a conceptual design framework for human–Artificial Intelligence (AI) symbiosis in extended reality (XR) that preserves human authorship while still benefiting from increasingly capable AI agents. Because XR can shape both perceptual evidence and action, apparently ‘helpful’ assistance can drift into over-reliance, covert persuasion, and blurred responsibility. Self++ grounds interaction in two complementary theories: Self-Determination Theory (autonomy, competence, relatedness) and the Free Energy Principle (predictive stability under uncertainty). It operationalises these foundations through co-determination, treating the human and the AI as a coupled system that must keep intent and limits legible, tune support over time, and preserve the user’s right to endorse, contest, and override. These requirements are summarised as the co-determination principles (T.A.N.): Transparency, Adaptivity, and Negotiability. Self++ organises augmentation into three concurrently activatable overlays spanning sensorimotor competence support (Self: competence overlay), deliberative autonomy support (Self+: autonomy overlay), and social and long-horizon relatedness and purpose support (Self++: relatedness and purpose overlay). Across the overlays, it specifies nine role patterns (Tutor, Skill Builder, Coach; Choice Architect, Advisor, Agentic Worker; Contextual Interpreter, Social Facilitator, Purpose Amplifier) that can be implemented as interaction patterns, not personas. The contribution is a role-based conceptual map that generates testable design propositions for XR-AI systems that grow capability without replacing judgment, enabling symbiotic agency in work, learning, and social life and resilient human development.

Genomic instability (GI), characterized by the progressive failure of mechanisms that maintain genome integrity, serves as a fundamental link between aging and cancer at the molecular level. It not only drives the aging process but also promotes tumorigenesis through multiple pathways: on one hand, GI can induce cellular senescence and create a pro-inflammatory and tissue remodeling microenvironment via the senescence-associated secretory phenotype; on the other hand, GI can bypass senescence, directly facilitating tumor progression through mechanisms such as aneuploidy, the expansion of pre-malignant clones, and chronic inflammation mediated by DNA damage-associated molecular patterns. The decline in physiological functions accompanying aging and the increased risk of cancer are closely associated with the accumulation of GI, while aging itself may exert anti-cancer effects through irreversible cell cycle arrest in specific contexts. Therefore, a thorough investigation of GI’s dual role in aging and cancer can help reveal the shared biological basis of both processes and provide new strategies for the precise prevention and treatment of age-related tumors.

Accumulating evidence has indicated that the normal tissue adjacent to tumor (NAT) is distinct from both healthy and tumor tissues. It is suggested that the crosstalk between NAT and tumor tissues helps shape the tumor microenvironment and promotes cancer progression, but the molecular and cellular evidence for this crosstalk is scarce. In this perspective, we propose that NAT tissue constitutes a unique “peritumoral microenvironment (Peri-TME)” mostly conditioned by the tumor. Furthermore, cellular senescence is identified as a key characteristic of Peri-TME, which accelerates tumor growth, as illustrated in our recent studies in colorectal cancer (CRC). Finally, strategies to target the senescent Peri-TME may represent an effective means to disrupt the vicious interaction between Peri-TME and Tumor and enhance therapeutic efficacy.

Aging is a heterogeneous, multi-system process driven by the interplay between accumulating molecular damage and the progressive erosion of resilience. While damage accumulates in a ubiquitous and homogeneous fashion, resilience is finite and unequally distributed across physiological systems and individuals, yielding distinct biological trajectories that diverge early in life, giving rise to the individual-specific decline in physiological function and the manifestation of a diverse spectrum of organ-specific diseases, and converge only when multisystem dysregulation overwhelms compensatory capacity. Early deviations in mitochondrial function, proteostasis, inflammation, and metabolic regulation often remain clinically silent, detectable only through gerodiagnostics, longitudinal, sensitive biomarkers of aging. Yet most biomarkers were developed to detect disease rather than quantify aging biology, and their mechanistic specificity declines with advancing multimorbidity. Precision geromedicine therefore requires the integration of gerodiagnostics that capture system-level resilience and stress responsiveness with measures of functional reserve, behavior, physiology, and the exposome, enabling the identification of individualized aging trajectories and the biological pathways that drive them. This combined approach clarifies causal pathways, enables earlier detection of vulnerability and supports individualized gerointerventions that modify aging trajectories rather than specifically but narrowly focusing on individual age-related diseases.

Knots and links play a fundamental role across a wide range of physical fields, from classical to quantum physics. In optics, structured light fields with multiple controllable degrees of freedom provide a versatile experimental platform for investigating topological properties. Knot theory underpins the topological control of high-dimensional structured light, thereby giving rise to fundamental physical effects and applications. Furthermore, topologically structured light has attracted significant attention for light-matter interaction. Here we review the recent advances in topologically structured light from the perspective of knot theory. Starting from the basics of knots and related braids, we introduce the generation and manipulation of topologically structured light from the purely spatial domain across to the spatiotemporal domain. Moreover, we demonstrate that the particle-like structured light, such as photonic skyrmions and hopfions, can host the topologies of high dimensional space, followed by brief discussions on potential applications as well as an outlook and future trends and challenges in this field.

Lymphatic endothelial cells (LECs) and the lymphatic vasculature have evolved from being viewed as passive conduits for fluid drainage and metastatic dissemination to active, dynamic regulators of inflammation and tumor immunity. In solid tumors, both tumor-associated and lymph node (LN)–resident LECs engage in complex interactions with their environment to orchestrate immune processes, including antigen transport and presentation to T cells, leukocyte recruitment and trafficking via chemokine gradients, and local immune modulation through the expression of co-inhibitory ligands such as programmed death-ligand 1 (PD-L1). These multifaceted roles enable LECs to either amplify effector responses or induce tolerance, profoundly influencing the efficacy of cancer immunotherapies depending on their activation state, tissue context, and molecular programming. This minireview synthesizes and discusses recent advances in tumor lymphangiogenesis, the role of LECs and their intensive crosstalk with the immune compartments, in the coordination of anti-tumor immune responses, with particular focus on LEC-autophagy as a lipid metabolic checkpoint controlling lymph node T cell egress, and its far-reaching implications for optimizing immunotherapy outcomes in solid tumors.

Artificial intelligence (AI) is revolutionizing the design of ionizable lipids, the pivotal components of lipid nanoparticles (LNPs) for messenger RNA (mRNA) delivery, enabling efficient exploration of vast chemical space of ionizable lipids beyond the reach of traditional methods. This mini-review explores the burgeoning field of AI-powered design and optimization of ionizable lipids for mRNA delivery. We also discuss the critical role of high-throughput experimental strategies, particularly barcoding coupled with next-generation sequencing, in generating the large-scale in vivo datasets for model training. Finally, we discuss current challenges, including data quality and the necessity for domain-specific modeling strategies, and present a future outlook on the integration of AI with scientific computing for LNP research.

Ferroptosis, a regulated cell death modality, driven by iron-dependent lipid peroxidation, is intrinsically coupled with mitochondrial metabolic turbulence and redox dysregulation. While the mitochondrial sirtuin Sirtuin 3 (SIRT3) is canonically viewed as a master regulator of energy homeostasis, its defensive repertoire against ferroptosis extends far beyond the simplistic activation of antioxidant enzymes. In this review, we synthesize emerging evidence to construct an integrated “metabolic-structural” defense model orchestrated by SIRT3. We first delineate how SIRT3 functions as a metabolic rheostat, rewiring tricarboxylic acid (TCA) cycle flux via the deacetylation of isocitrate dehydrogenase 2 (IDH2) to sustain the nicotinamide adenine dinucleotide phosphate (NADPH)/glutathione (GSH) reservoir. Breaking away from the classical enzymatic paradigm, we highlight a novel non-enzymatic substrate regulatory mechanism where SIRT3 stabilizes the glutamate transporter SLC25A22 through specific deacetylation-ubiquitination crosstalk, thereby limiting ferroptotic susceptibility. Furthermore, we expand the SIRT3 signaling landscape by proposing a “SIRT3-nuclear factor erythroid 2-related factor 2 (Nrf2) deacetylation axis” that bridges mitochondrial stress signals to nuclear transcriptional defense, and by detailing its control over iron entry via the iron regulatory protein 1 (IRP1)-transferrin receptor 1 (TfR1) pathway. At the organelle level, we examine how SIRT3 remodels mitochondrial dynamics, upregulating optic atrophy-associated protein 1 (OPA1) while suppressing dynamin-related protein 1 (DRP1), to construct a “fusion network barrier” that dilutes ROS toxicity. We also posit a critical hypothesis: SIRT3 safeguards the integrity of mitochondria-associated endoplasmic reticulum membranes, preventing structural decoupling and calcium overload that triggers ferroptotic sensitization. Finally, we reconcile the context-dependent duality of SIRT3 in cancer and degenerative diseases, outlining future therapeutic strategies that leverage these multidimensional vulnerabilities.

Background: Glioblastoma multiforme (GBM) progression relies on active dialog between tumor cells and infiltrating immune cells, mainly including macrophages. Macrophages are closely linked to iron handling and reactive oxygen species (ROS) regulation, suggesting that ferroptosis may play a pivotal role in macrophage behavior.

Methods: Bulk RNA-seq GBM datasets, The Cancer Genome Atlas (TCGA)-GBM and gene set enrichment 4412 (GSE4412) were used to quantify immune infiltration (xCell), identify differentially expressed genes (limma), and extract ferroptosis-related markers (FerrDb V2). Ferroptosis-associated transcriptional programs were visualized using the ferroviz R package. Single-cell RNA-sequencing GBM datasets (GSE189650) were processed to characterize ferroptosis-related signatures and explore the dynamic regulation of ferroptosis-related genes in M2 tumor-associated macrophages (M2-TAMs). Deep learning neural networks were trained to predict recurrence based on ferroptosis hallmarks.

Results: High M2-TAM infiltration in poor-prognosis tumors was associated with a deregulated ferroptosis program defined by nine markers, tumor necrosis factor alpha-induced protein 3 (TNFAIP3), arachidonate lipoxygenase 5 (ALOX5), perilipin 2 (PLIN2), spermidine/spermine-N1-acetyltransferase 1(SAT1), ferritin light chain (FTL), cathepsin B (CTSB), poly (ADP-ribose) polymerase 8 (PARP8), heme oxygenase-1 (HMOX1), autophagy-related 7 (ATG7), validated at single-cell resolution and enriched in CD16+/CD163+ myeloid cells. A ferroptosis score independently predicted M2 infiltration in a clinicobiological multivariable model. Deep learning analyses revealed opposite regulation of TNFAIP3 (repressed ferroptosis driver) and FTL (upregulated ferroptosis suppressor) during recurrence, indicating impaired ferroptosis in M2-TAMs during disease progression. TNFAIP3 expression in primary tumors aligned with a pro-inflammatory signature, whereas FTL expression in recurrent tumors correlated with an invasive program involving apolipoprotein E (APO-E), serpin family E member 1 (SERPINE1), thioredoxin-interacting protein (TXNIP), glycoprotein non‑metastatic melanoma protein B (GPNMB), cystatin C (CST3), dihydrofolate reductase (DHFR), β2-microglobulin (B2M), and nuclear protein-1 (NUPR1).

Conclusion: Glioblastoma-associated macrophages display defective ferroptosis program in aggressive and recurrent tumors, linking reduced inflammatory activity to enhanced invasiveness. Restoring ferroptosis activity in these immune cells may represent a promising therapeutic strategy.

Microwave (MW) medicine has emerged as a distinct interdisciplinary field, predicated on the unique capacity of non-ionizing electromagnetic radiation to penetrate deep-seated tissues and interact efficiently with biological dielectrics for diverse therapeutic and diagnostic applications. Despite its clinical establishment in tumor ablation and hemostasis, conventional MW interventions are largely constrained by non-selective macroscopic heating, leaving the intricate potential of non-thermal biophysical modulation underutilized. The integration of engineered biomaterials provides a transformative framework to bridge this gap, enabling the precise modulation of MW-tissue interactions at the micro- and nanoscale. This review systematically elucidates how rational material design via tuning dielectric and magnetic loss, band-gap engineering, and structural polarization expands MW medicine beyond bulk heating toward controlled biological regulation. We discuss mechanisms where biomaterials function as localized energy antennas to sharpen thermal gradients, as MW-dynamic sensitizers to induce reactive oxygen species generation, and as intelligent interfaces to regulate ionic homeostasis. Representative advancements are summarized across antitumor, antibacterial, and anti-inflammatory therapies, alongside innovations in high-fidelity thermoacoustic imaging. Furthermore, emerging frontiers in non-destructive tissue repair and neuromodulation are highlighted. This review critically examines the design principles and translational challenges of MW-based medical technologies by analyzing correlations between physicochemical parameters and specific biological outcomes. It is expected to advance MW medicine from empirically guided thermal interventions toward mechanism-driven, precision-targeted electromagnetic therapeutics.

Aging tissues accumulate DNA damage, while genome instability is a defining feature of cancer. Despite this shared foundation, DNA damage is still largely viewed as a transient lesion that is either faithfully repaired or converted into a mutation. New evidence challenges this binary view, indicating that DNA double-strand breaks (DSBs), even when accurately repaired at the sequence level, can leave durable and heritable alterations in chromatin organization and gene regulation. Accordingly, DSB repair restores DNA integrity but does not necessarily re-establish the original regulatory architecture. The biological consequences of such post-repair regulatory memory remain largely underappreciated, progressively contributing to age-associated tissue dysfunction while simultaneously creating permissive states for malignant transformation and therapy resistance. In this commentary, we argue that reframing DNA damage as a source of heritable regulatory change, rather than solely as a mutational event, reshapes our understanding of aging trajectories and cancer risk.

Caspase-2 is a key genome-surveillance protease that functions to kill or arrest cells with abnormal chromosome content, via PIDDosome-dependent and -independent mechanisms. However, this surveillance function presents a biological paradox in the liver, where physiological polyploidy, is essential for organogenesis, genomic buffering and adaptive stress responses to maintain liver homeostasis. While short-term caspase-2 loss promotes adaptive polyploidy, our recent work demonstrates that prolonged caspase-2 deficiency drives pathogenic hyperploidy, fuelling chronic inflammation, and increased age-associated hepatocellular carcinoma in mice. These findings underscore the need for tight ploidy control in the maintenance of liver homeostasis. They also highlight potential risks for therapeutic targeting of caspase-2 in fatty liver disease and suggest that pathways governing lipid metabolism also influence long-term tumour surveillance mechanisms. This perspective discusses caspase-2 as a guardian of hepatic genome integrity, and the dual, context-dependent roles of polyploidy in liver physiology and metabolic liver disease.

Thermal comfort, indoor air quality, lighting comfort, and acoustic comfort are often treated as separate domains, despite interacting trade-offs. The discrete choice experiment (DCE) technique helps to account for explicit and implicit factors influencing choices, giving a more holistic view of indoor comfort. This paper reviews discrete choice experiment studies that include indoor environmental comfort elements. From 1,418 records found in scientific databases, 21 studies were reviewed. By analysing these studies, DCE’s ability to value indoor comfort was evaluated and methodological deficiencies identified. Given the heterogeneity between studies, this review focused on analysing the p-value and the direction of estimates. Thermal comfort in these studies shows positive and statistically significant coefficients, indoor air quality is generally positive but statistically heterogeneous. Evidence for positively valued lighting and acoustic comfort is frequently statistically non-significant. Methodological challenges identified affecting indoor environmental quality (IEQ) valuation in DCEs include proxy dilution, insufficient attribute clarity, embedded cost signals, and demographic heterogeneity. Proxy-based operationalisations may confound comfort valuation with convenience, cost, or multi-domain interaction effects. Additionally, occupant control and agency emerge as meta-attributes that generate utility beyond environmental improvement. Survey-based DCE can reliably elicit thermal comfort preferences from occupants; experiential approaches such as virtual reality-based DCE may allow occupants to better value acoustic and lighting comfort. Overall findings suggest indoor comfort should be modelled as a multi-attribute bundle instead of separate domains. This review shows that DCE is a useful tool for analysing indoor comfort as a bundle of attributes with strong interaction effects.

Construction jobs involve heavy, repetitive loads and awkward postures that are prone to increased risk of musculoskeletal injury. As a result, a recognized demand exists for dependable exoskeleton support to improve workers’ safety and productivity. The main objective of this project was to enhance the controls based on the ability of surface electromyography (sEMG) to obtain the motion intentions of its users, preventing mistimed torque and excessive force. To solve the issues of signal noise and inter-individual variability, this study proposes a hybrid model, which combines nonnegative matrix factorization to extract transferable muscle coordination patterns, a convolutional neural network to learn robust local cross-channel features, and a Transformer encoder to capture long-range temporal dependencies. This approach was tested on a bricklaying dataset that mimics the actual conditions on construction jobsites. The model performance was 87% in action-recognition accuracy, and generalized across subjects, illustrating robustness to signal artifacts along with differences in tools, postures, and workloads. The results support the proposal of sEMG as a non-invasive practical control signal for exoskeletons and indicate that the proposed model can enable timely and precise assistance on a variety of tasks. This study extends the previous research in intent recognition for field-ready exoskeleton manipulation and establishes the foundation for future multi-task adaptation and personalized assistance in the construction process.

Acute respiratory distress syndrome (ARDS) is a life-threatening form of acute respiratory failure characterized by diffuse lung inflammation and edema. Despite increased understanding of the molecular biology underlying ARDS, the complex pathogenesis still limits the development of targeted pharmacologic therapies. Cell death plays a vital role in defending against pathogen infections and triggering tissue inflammation, which can damage the alveolar-capillary barrier and ultimately lead to the development of ARDS. Thus, targeting various cell death pathways may be an attractive entry point for therapeutic intervention in ARDS. Intriguingly, recent genetic and biochemical studies have emphasized the importance of revealing the crosstalk among various cell death pathways and indicated that this connectivity exhibits a considerable degree of plasticity in the molecular regulation of potential therapeutic targets in ARDS. In this review, we summarize the mechanisms of the different types of regulated cell death (RCD) and describe the physiological and pathological processes that contribute to ARDS pathogenesis. We also discuss the emerging crosstalk among various RCD modalities, and highlight that targeting cell death pathways is an effective therapeutic strategy for ARDS.

Spatiotemporal optical vortices (STOVs) possess helical phase singularities distributed jointly in space and time, enabling light to carry transverse orbital angular momentum and offering a fundamentally new degree of freedom for structured light. However, the precise detection of STOVs with fractional topological charges remains highly challenging, as conventional interferometric and diffraction-based techniques suffer from limited resolution and experimental complexity. Here we demonstrate a scattering-assisted detection scheme that enables ultra-high-precision measurement of STOVs with fractional topological charges. By exploiting the sensitivity of random scattering media to the spatiotemporal phase structure of broadband optical fields, we establish a robust mapping between scattered intensity patterns and fractional STOV states. This approach achieves reliable discrimination of fractional topological charges with step sizes down to 0.01, significantly surpassing the resolution of existing methods. Furthermore, we leverage this capability to realize a multi-level free-space optical communication scheme encoded by fractional STOV states, demonstrating enhanced channel capacity within a compact experimental configuration. This work introduces scattering media as a powerful platform for probing spatiotemporal phase singularities, and opens new opportunities for high-dimensional spatiotemporal photonics and optical communications.

Background: Cystic fibrosis (CF) is a progressive genetic disease characterized by defective ion transport, mucus accumulation, chronic infection, and inflammation that drive airway damage and ultimately end-stage lung failure. Previous studies show that high levels of proteolytic enzymes in the sputum of CF patients correlate with declining lung function, but the related effects on distal lung extracellular matrix (ECM) and immune responses are unclear.

Methods: To address this gap, induced pluripotent stem cell (iPSC) lines from healthy donors and CF patients were differentiated into macrophages, and stimulated with lipopolysaccharide (LPS) to compare their inflammatory responses. Bulk RNA sequencing, functional assays, and secreted protein profiling revealed key differences between healthy and CF-derived macrophages, providing insight into how these cells may contribute to inflammatory responses in CF patients. Further, human lung ECM from distal CF lung tissue was isolated, used to generate ECM biomaterials, and combined with iPSC-derived macrophages from healthy and CF donors in vitro. Macrophage phenotype was evaluated through cytokine profiling and RNA sequencing.

Results: CF macrophage inflammation was dysregulated, with elevated baseline IL-8, IL-18, and MCP-1 expression, and a blunted inflammatory response to CF ECM compared to healthy macrophages. By using CF ECM and healthy macrophages, we characterized how healthy cells may be altered in a persistent CF milieu after anticipated CFTR modulator therapy.

Conclusion: These findings reveal altered innate immune behavior in CF and demonstrate the utility of iPSC-derived macrophages for modeling extrinsic immune-ECM interactions in disease.

Developing new quality productive forces (NQPFs) in the building and real estate industry is essential for sustainable economic growth. Utilizing panel data on Chinese A-share listed building and real estate companies from 2011 to 2024, this study empirically investigates the impact of digital transformation (DT) on corporate NQPFs and the underlying mechanisms. Our baseline findings confirm that DT exerts a significantly positive effect on NQPFs, effectively overcoming the “Solow Paradox” in this capital-intensive sector. Moving beyond this initial assessment, the study systematically unpacks the mechanisms underlying digital value creation. We reveal that DT promotes the enhancement of NQPFs through three core mechanisms: fostering technological innovation, upgrading the human capital structure, and improving asset operational efficiency. Furthermore, the multidimensional heterogeneity analysis demonstrates that this digital empowerment is highly context-dependent. The enabling effect is significantly more pronounced in the Eastern and Central regions and is amplified by stronger intellectual property protection, more advanced digital infrastructure, state ownership, and larger firm size. By elucidating these specific pathways, this research offers targeted guidelines for sector-specific digital policy-making and strategies to optimize corporate NQPFs.

Aims: Since distal enhancers are involved in regulating target genes through physical contacting with proximal promoters, identifying enhancer-promoter interactions (EPIs) is critical to deepening our understanding of gene expression. However, high-throughput experimental methods for identifying EPIs are time-consuming and expensive. Therefore, computational methods for predicting EPIs would be valuable and important, but also face a lot of challenges.

Methods: In this paper, we propose a novel deep learning-based method, namely EPIPAM, to predict EPIs only using genomic sequences. EPIPAM firstly uses a deep convolutional neural network to extract high-level sequence features, and then uses a position attention mechanism to compute the positional correlation coefficients of two subregions separately coming from enhancers and promoters, aiming to focus on important regions of them.

Results: Benchmarking comparisons on six different cell lines show that EPIPAM performs better than the state-of-the-art methods in the task of EPIs prediction. More importantly, we notice that, almost without exception, the predictive performance of all methods is really poor once applying a strategy of splitting training and test data by chromosome. Therefore, we explain the possible reason that leads to this situation by systematically exploring the structure of EPI datasets, and indirectly analyze the difficulty of predicting EPIs only using genomic sequences through ChIA-PET contact datasets.

Conclusion: This study presents a novel deep learning-based method to predict EPIs only using genomic sequences. Although the proposed method achieves higher predictive accuracy, it suffers from several limitations, such as highly selective matching bias, negative sample selection issues, and constraints of pre-trained vectors.