GPR126, also known as ADGRG6, is one of the most deeply studied aGPCRs. Initially, GPR126 was thought to be a receptor associated with muscle development and was primarily expressed in the muscular and skeletal systems. With the deepening of research, it was found that GPR126 is expressed in multiple mammalian tissues and organs, and is involved in many biological processes such as embryonic development, nervous system development, and extracellular matrix interactions. Compared with other aGPCRs proteins, GPR126 has a longer N-terminal domain, which can bind to ligands one-to-one and one-to-many. Its N-terminus contains five domains, a CUB (complement C1r/C1s, Uegf, Bmp1) domain, a PTX (Pentraxin) domain, a SEA (Sperm protein, Enterokinase, and Agrin) domain, a hormone binding (HormR) domain, and a conserved GAIN domain. The GAIN domain has a self-shearing function, which is essential for the maturation, stability, transport and function of aGPCRs. Different SEA domains constitute different GPR126 isomers, which can regulate the activation and closure of downstream signaling pathways through conformational changes. GPR126 has a typical aGPCRs seven-transmembrane helical structure, which can be coupled to Gs and Gi, causing cAMP to up- or down-regulation, mediating transmembrane signaling and participating in the regulation of cell proliferation, differentiation and migration. GPR126 is activated in a tethered-stalk peptide agonism or orthosteric agonism, which is mainly manifested by self-proteolysis or conformational changes in the GAIN domain, which mediates the rapid activation or closure of downstream pathways by tethered agonists. In addition to the tethered short stem peptide activation mode, GPR126 also has another allosteric agonism or tunable agonism mode, which is specifically expressed as the GAIN domain does not have self-shearing function in the physiological state, NTF and CTF always maintain the binding state, and the NTF binds to the ligand to cause conformational changes of the receptor, which somehow transmits signals to the GAIN domain in a spatial structure. The GAIN domain can cause the 7TM domain to produce an activated or inhibited signal for signal transduction, For example, type IV collagen interacts with the CUB and PTX domains of GPR126 to activate GPR126 downstream signal transduction. GPR126 has homology of 51.6%-86.9% among different species, with 10 conserved regions between different species, which can be traced back to the oldest metazoans as well as unicellular animals.In terms of diseases, GPR126 dysfunction involves the pathological process of bone, myelin, embryo and other related diseases, and is also closely related to the occurrence and development of malignant tumors such as breast cancer and colon cancer. However, the biological function of GPR126 in various diseases and its potential as a therapeutic target still needs further research. This paper focuses on the structure, interspecies differences and conservatism, signal transduction and biological functions of GPR126, which provides ideas and references for future research on GPR126.

Membrane proteins are integral components of cellular membranes, accounting for approximately 30% of the mammalian proteome and serving as targets for 60% of FDA-approved drugs. They are critical to both physiological functions and disease mechanisms. Their functional protein-protein interactions form the basis for many physiological processes, such as signal transduction, material transport, and cell communication. Membrane protein interactions are characterized by membrane environment dependence, spatial asymmetry, weak interaction strength, high dynamics, and a variety of interaction sites. Therefore, in situ analysis is essential for revealing the structural basis and kinetics of these proteins. This paper introduces currently available in situ analytical techniques for studying membrane protein interactions and evaluates the characteristics of each. These techniques are divided into two categories: label-based techniques (e.g., co-immunoprecipitation, proximity ligation assay, bimolecular fluorescence complementation, resonance energy transfer, and proximity labeling) and label-free techniques (e.g., cryo-electron tomography, in situ cross-linking mass spectrometry, Raman spectroscopy, electron paramagnetic resonance, nuclear magnetic resonance, and structure prediction tools). Each technique is critically assessed in terms of its historical development, strengths, and limitations. Based on the authors’ relevant research, the paper further discusses the key issues and trends in the application of these techniques, providing valuable references for the field of membrane protein research. Label-based techniques rely on molecular tags or antibodies to detect proximity or interactions, offering high specificity and adaptability for dynamic studies. For instance, proximity ligation assay combines the specificity of antibodies with the sensitivity of PCR amplification, while proximity labeling enables spatial mapping of interactomes. Conversely, label-free techniques, such as cryo-electron tomography, provide near-native structural insights, and Raman spectroscopy directly probes molecular interactions without perturbing the membrane environment. Despite advancements, these methods face several universal challenges: (1) indirect detection, relying on proximity or tagged proxies rather than direct interaction measurement; (2) limited capacity for continuous dynamic monitoring in live cells; and (3) potential artificial influences introduced by labeling or sample preparation, which may alter native conformations. Emerging trends emphasize the multimodal integration of complementary techniques to overcome individual limitations. For example, combining in situ cross-linking mass spectrometry with proximity labeling enhances both spatial resolution and interaction coverage, enabling high-throughput subcellular interactome mapping. Similarly, coupling fluorescence resonance energy transfer with nuclear magnetic resonance and artificial intelligence (AI) simulations integrates dynamic structural data, atomic-level details, and predictive modeling for holistic insights. Advances in AI, exemplified by AlphaFold’s ability to predict interaction interfaces, further augment experimental data, accelerating structure-function analyses. Future developments in cryo-electron microscopy, super-resolution imaging, and machine learning are poised to refine spatiotemporal resolution and scalability. In conclusion, in situ analysis of membrane protein interactions remains indispensable for deciphering their roles in health and disease. While current technologies have significantly advanced our understanding, persistent gaps highlight the need for innovative, integrative approaches. By synergizing experimental and computational tools, researchers can achieve multiscale, real-time, and perturbation-free analyses, ultimately unraveling the dynamic complexity of membrane protein networks and driving therapeutic discovery.

Objective To investigate the relationship between brain white matter lesions (WML) and triglyceride glucose (TyG) index in non-diabetic elderly individuals based on magnetic resonance imaging. Methods A total of 523 non-diabetic elderly individuals aged ≥ 60 years were selected from Jinan, Shandong Province, China from June 2018 to December 2019. According to the quartiles of TyG index, there were 133 participants in the first quartile (Q1) group, 127 in the second quartile (Q2) group, 132 in the third quartile (Q3) group, and 131 in the fourth quartile (Q4) group. All participants underwent brain magnetic resonance imaging to evaluate paraventricular, deep, and total WML volumes, as well as Fazekas scores. Results Compared with Q1, Q2, and Q3 groups, Q4 group showed significant increase in periventricular, deep, and total WML volumes (P < 0.05). The proportion of participants with a Fazekas score ≥ 2 in the periventricular, deep, and total WML was higher in the Q4 group compared with the Q1 and Q2 groups (P < 0.05). The proportion of participants with a Fazekas score ≥ 2 in deep WML was higher in Q4 group than in Q3 group (P < 0.05). TyG index was significantly positively correlated with periventricular, deep, and total WML volumes (r = 0.401, 0.405, and 0.445, P < 0.001). After adjusting for confounding factors, TyG index was still significantly positively correlated with periventricular, deep, and total WML volumes (P < 0.001). Logistic regression analysis showed that compared with Q1 group, the risk of Fazekas score ≥ 2 in periventricular WML was 1.950-fold (95% confidence interval [CI]: 1.154-3.294, P = 0.013) in Q3 group and 3.411-fold (95% CI: 1.984-5.863, P < 0.001) in Q4 group, the risk of Fazekas score ≥ 2 in total WML was 2.529-fold (95%CI: 1.444-4.430, P = 0.001) in Q3 group and 4.486-fold (95%CI: 2.314-8.696, P < 0.001) in Q4 group. The risk of Fazekas score ≥ 2 in deep WML was 2.953-fold (95%CI: 1.708-5.106, P < 0.001) in Q4 group compared with Q1 group. Conclusion Increased TyG index is an independent risk factor for WML in non-diabetic elderly individuals.

Objective: Machine learning (ML) has been reported to have better predictive capability than traditional statistical techniques. The aim of this study was to assess the efficacy of ML algorithms and logistic regression (LR) for predicting depressive symptoms during the COVID-19 pandemic. Methods: Analyses were carried out in a national cross-sectional study involving 21,916 participants. The ML algorithms in this study included random forest (RF), support vector machine (SVM), neural network (NN), and gradient boosting machine (GBM) methods. The performance indices were sensitivity, specificity, accuracy, precision, F1-score, and area under the receiver operating characteristic curve (AUC). Results: LR and NN had the best performance in terms of AUCs. The risk of overfitting was found to be negligible for most ML models except for RF, and GBM obtained the highest sensitivity, specificity, accuracy, precision, and F1-score. Therefore, LR, NN, and GBM models ranked among the best models. Conclusion Compared with ML models, LR model performed comparably to ML models in predicting depressive symptoms and identifying potential risk factors while also exhibiting a lower risk of overfitting.

Objective: Machine learning (ML) has been reported to have better predictive capability than traditional statistical techniques. The aim of this study was to assess the efficacy of ML algorithms and logistic regression (LR) for predicting depressive symptoms during the COVID-19 pandemic. Methods: Analyses were carried out in a national cross-sectional study involving 21,916 participants. The ML algorithms in this study included random forest (RF), support vector machine (SVM), neural network (NN), and gradient boosting machine (GBM) methods. The performance indices were sensitivity, specificity, accuracy, precision, F1-score, and area under the receiver operating characteristic curve (AUC). Results: LR and NN had the best performance in terms of AUCs. The risk of overfitting was found to be negligible for most ML models except for RF, and GBM obtained the highest sensitivity, specificity, accuracy, precision, and F1-score. Therefore, LR, NN, and GBM models ranked among the best models. Conclusion Compared with ML models, LR model performed comparably to ML models in predicting depressive symptoms and identifying potential risk factors while also exhibiting a lower risk of overfitting.

Objective: Machine learning (ML) has been reported to have better predictive capability than traditional statistical techniques. The aim of this study was to assess the efficacy of ML algorithms and logistic regression (LR) for predicting depressive symptoms during the COVID-19 pandemic. Methods: Analyses were carried out in a national cross-sectional study involving 21,916 participants. The ML algorithms in this study included random forest (RF), support vector machine (SVM), neural network (NN), and gradient boosting machine (GBM) methods. The performance indices were sensitivity, specificity, accuracy, precision, F1-score, and area under the receiver operating characteristic curve (AUC). Results: LR and NN had the best performance in terms of AUCs. The risk of overfitting was found to be negligible for most ML models except for RF, and GBM obtained the highest sensitivity, specificity, accuracy, precision, and F1-score. Therefore, LR, NN, and GBM models ranked among the best models. Conclusion Compared with ML models, LR model performed comparably to ML models in predicting depressive symptoms and identifying potential risk factors while also exhibiting a lower risk of overfitting.

Objective: Machine learning (ML) has been reported to have better predictive capability than traditional statistical techniques. The aim of this study was to assess the efficacy of ML algorithms and logistic regression (LR) for predicting depressive symptoms during the COVID-19 pandemic. Methods: Analyses were carried out in a national cross-sectional study involving 21,916 participants. The ML algorithms in this study included random forest (RF), support vector machine (SVM), neural network (NN), and gradient boosting machine (GBM) methods. The performance indices were sensitivity, specificity, accuracy, precision, F1-score, and area under the receiver operating characteristic curve (AUC). Results: LR and NN had the best performance in terms of AUCs. The risk of overfitting was found to be negligible for most ML models except for RF, and GBM obtained the highest sensitivity, specificity, accuracy, precision, and F1-score. Therefore, LR, NN, and GBM models ranked among the best models. Conclusion Compared with ML models, LR model performed comparably to ML models in predicting depressive symptoms and identifying potential risk factors while also exhibiting a lower risk of overfitting.

Objective: Machine learning (ML) has been reported to have better predictive capability than traditional statistical techniques. The aim of this study was to assess the efficacy of ML algorithms and logistic regression (LR) for predicting depressive symptoms during the COVID-19 pandemic. Methods: Analyses were carried out in a national cross-sectional study involving 21,916 participants. The ML algorithms in this study included random forest (RF), support vector machine (SVM), neural network (NN), and gradient boosting machine (GBM) methods. The performance indices were sensitivity, specificity, accuracy, precision, F1-score, and area under the receiver operating characteristic curve (AUC). Results: LR and NN had the best performance in terms of AUCs. The risk of overfitting was found to be negligible for most ML models except for RF, and GBM obtained the highest sensitivity, specificity, accuracy, precision, and F1-score. Therefore, LR, NN, and GBM models ranked among the best models. Conclusion Compared with ML models, LR model performed comparably to ML models in predicting depressive symptoms and identifying potential risk factors while also exhibiting a lower risk of overfitting.

Aim To investigate the role of metabolites of eicosapentaenoic acid (EPA) in promoting the transdifferentiation of pancreatic α cells to β cells. Methods Male C57BL/6J mice were injected intraperitoneally with 60 mg/kg streptozocin (STZ) for five consecutive days to establish a type 1 diabetes (T1DM) mouse model. After two weeks, they were randomly divided into model groups and 97% EPA diet intervention group, 75% fish oil (50% EPA +25% DHA) diet intervention group, and random blood glucose was detected every week; after the model expired, the regeneration of pancreatic β cells in mouse pancreas was observed by immunofluorescence staining. The islets of mice (obtained by crossing GCG

In recent years, nucleic acid therapy, as a revolutionary therapeutic tool, has shown great potential in the treatment of genetic diseases, infectious diseases and cancer. Lipid nanoparticles (LNPs) are currently the most advanced mRNA delivery carriers, and their emergence is an important reason for the rapid approval and use of COVID-19 mRNA vaccines and the development of mRNA therapy. Currently, mRNA therapeutics using LNP as a carrier have been widely used in protein replacement therapy, vaccines and gene editing. Conventional LNP is composed of four components: ionizable lipids, phospholipids, cholesterol, and polyethylene glycol (PEG) lipids, which can effectively load mRNA to improve the stability of mRNA and promote the delivery of mRNA to the cytoplasm. However, in the face of the complexity and diversity of clinical diseases, the structure, properties and functions of existing LNPs are too homogeneous, and the lack of targeted delivery capability may result in the risk of off-targeting. LNPs are flexibly designed and structurally stable vectors, and the adjustment of the types or proportions of their components can give them additional functions without affecting the ability of LNPs to deliver mRNAs. For example, by replacing and optimizing the basic components of LNP, introducing a fifth component, and modifying its surface, LNP can be made to have more precise targeting ability to reduce the side effects caused by treatment, or be given additional functions to synergistically enhance the efficacy of mRNA therapy to respond to the clinical demand for nucleic acid therapy. It is also possible to further improve the efficiency of LNP delivery of mRNA through machine learning-assisted LNP iteration. This review can provide a reference method for the rational design of engineered lipid nanoparticles delivering mRNA to treat diseases.