Objective: This study proposes a clustering framework for Chinese materia medica (CMM) based on a large language model (LLM), aiming to explore potential compatibility patterns among CMMs from the semantic perspective of CMM property theory. Methods: First, a CMM property knowledge base was constructed based on Chinese Materia Medica, including 567 commonly used CMMs characterized by four properties, five flavors, and meridian tropism. Then, 49 CMMs derived from 10 prescriptions for Zangdu (脏毒, pathogenic toxins) recorded in Waike Zhengzong (《外科正宗》, Orthodox Manual of External Medicine) and Yangke Xinde Ji (《疡科心得集》, Collected Insights on Ulcer Medicine) were selected as the experimental dataset. Five semantic representation methods—One-Hot, Word2Vec, Bidirectional Encoder Representations from Transformers (BERT), Beijing Academy of Artificial Intelligence General Embedding (BGE), and Qwen—were applied to encode CMM property information into vector representations. Subsequently, t-distributed Stochastic Neighbor Embedding (t-SNE) was used for nonlinear dimensionality reduction on high-dimensional semantic vectors, followed by k-means clustering (k = 7). Clustering performance was evaluated using the Silhouette Score (SS), Davies-Bouldin Index (DBI), and Calinski-Harabasz Index (CHI). Results: The Qwen-based clustering method, CMM-EmbedCluster, achieved the highest SS (0.607 4) and CHI (158.057 2), as well as the lowest DBI (0.499 5), indicating improved cluster separation and compactness compared with other methods. Visualization of CMM clustering results showed that the clusters were well separated in the low-dimensional space, with strong inter-cluster discrimination and high intra-cluster functional consistency. Further interpretability analysis of CMM clustering results revealed stable structural differences among clusters in terms of four properties, five flavors, and meridian tropism, forming functional partitions consistent with CMM property theory. Conclusion CMM-EmbedCluster utilizes an LLM to achieve semantic-level representation and clustering of CMMs within the framework of CMM property theory, providing support for exploring potential compatibility patterns among CMMs from the perspective of CMM property semantics.

Against the backdrop of digital transformation in higher education， the traditional teaching model has become inadequate to meet the demands for cultivating new-quality nursing talents. Nursing ethics teaching confronts several dilemmas， such as unbalanced regional development， disconnection between theory and practice， scattered teaching resources， and fragmented knowledge points. With the iterative update of modern digital technologies， the knowledge graph， due to its structured， semantic， scalable， and multi-source integrative characteristics， offers a technological pathway to address these issues. Based on the construction foundation of the Ministry of Education’s first batch of virtual teaching and research platforms for humanistic nursing courses， the nursing ethics artificial intelligence（Al）course was developed around the“thinking-research-teaching-learning”framework. By strengthening ideological and political guidance， constructing a curriculum knowledge graph， and promoting the deep integration of humanistic care and professional knowledge，the course realized intelligent support for research，learning，and teaching. This innovative education model provides a replicable paradigm for cultivating outstanding nursing talents equipped with humanistic literacy， digital competence， and advanced ethical thinking， while also offering references for the digital transformation of higher nursing education.

Objective:To explore the genotype and the clinical phenotype of SCN2A-related developmental delay in children. Methods:A case series study was adopted. Collect clinical data from 10 cases of children with SCN2A gene variants diagnosed with global developmental delay/intellectual disability who were admitted to the Children′s Hospital between July 2019 and March 2023. Summarize the clinical phenotype and genotype based on clinical data such as general information, clinical manifestations, imaging examinations, laboratory tests, genetic testing results, and comprehensive pediatric neuropsychological development assessment. Results:A total of 10 patients were recruited, including 7 males and 3 females, with an age range of 27 days to 5 years and 9 months. 9 patients underwent children′s neuropsychological and behavioral assessments, and the results were consistent with global developmental delay, including 2 mild cases, 4 moderate cases, and 3 severe cases. 3 cases had autism spectrum disorder, and 2 cases had epilepsy. 6 patients underwent complete head MRI examination, and 4 of them showed abnormalities, including delayed myelination, widening of the local extra brain space in the frontal lobe, and abnormal frontal lobe morphology. All 10 cases had point variants. Among them, 9 cases are de novo and 1 case is maternal inheritance. Out of 10 cases, there were 5 cases with copy number variations, but all of them were of unknown significance. Among the 10 variants, 8 have been reported and 2 have not been reported, namely c.4145A>T(p.N1382I) and c.4937T>A(p.I1646N). In this study, 4 out of 10 patients with SCN2A variants had variation sites located in the S4 segment of domain which constitute Nav1.2, the sodium ion channel encoded by SCN2A. The developmental quotient level was lower when the variation sites were located in the S4 segment of domain, and the difference was statistically significant ( t=-3.101, P=0.017), indicating that the severity of developmental delay may be related to the localization of amino acids corresponding to variant sites within the protein domain. Conclusion:SCN2A mutations are strongly associated with diverse neurodevelopmental disorders. In this study, the phenotypic spectrum of SCN2A variants encompassed epilepsy, global developmental delay, and autism spectrum disorder. Affected individuals exhibited early-onset developmental delays, predominantly moderate to severe in severity. Voltage-sensing domain dysfunction in sodium channels may constitute a critical pathomechanism underlying neurodevelopmental impairments. Further electrophysiological characterization and molecular mechanistic studies are warranted todelineate the genotype-phenotype correlations between specific variant loci and clinical severity.

OBJECTIVE To provide help for further exploring the mechanism of action of traditional Chinese herb by constructing a complex network of traditional Chinese herb-gene-protein,optimizing the mining method of potential associated genes of traditional Chinese herb and improving the mining efficiency of traditional Chinese herb system biology information.METHODS A graph neural network model HERBGAT with an attention mechanism was proposed.A small amount of traditional Chinese herb-related gene data in the public data platform was used as input,and deep mining was performed in the traditional Chinese herb-gene-protein complex net-work to output potential traditional Chinese herb-related genes.The prediction results were analyzed by disease association analysis and KEGG signaling pathway analysis on the bioinformatics platform to clarify their mechanism of action,and the prediction results were verified by the literature retrieval platform.RESULTS The training results showed that the average prediction accuracy of the HERB-GAT model could reach 94%.Compared with the other two advanced complex network mining methods,HERBGAT showed better per-formance in the three indicators of ACC,AUC and AUPR.In the literature verification stage,the model prediction results were verified by TCM clinical literature and modern pharmacology literature,showing the good effect of HERBGAT in practical application.At the end of this paper,taking the HERBGAT model and the improved EMOGI model to explore the mechanism of action of Pinellia ternata in treating lung cancer as an example,199 potential associated genes of Pinellia ternata in treating lung cancer were found,and these potential associated genes were preliminarily analyzed and discussed with the help of bioinformatics methods.CONCLUSION The HERBGAT model can effectively mine potential traditional Chinese herb-associated genes,improve the mining efficiency of traditional Chinese herb-gene-protein complex networks,provide new ideas and references for the optimization of traditional Chinese herb system biology information mining methods,and provide data basis and experimental direction for exploring the mechanism of action of tradi-tional Chinese herb.

Objective:To explore the application value of radiomics models based on MRI of vertebrae and paravertebral muscles in identifying potential vertebral fragility fracture risk in osteoporosis and osteopenia.Methods:This cross-sectional study collected data from patients who underwent both dual-energy X-ray absorptiometry (DXA) and lumbar MRI at the Third Affiliated Hospital of Southern Medical University between January 2014 and December 2023,retrospectively. Based on DXA results, patients were categorized into osteoporosis group ( n=302) and osteopenia group ( n=264), with fracture and non-fracture patients matched at 1∶1 ratio by propensity score matching based on age, gender, and body mass index. The fourth lumbar vertebra was selected as the region of interest (ROI) for the vertebral body, and the bilateral psoas major, erector spinae, and multifidus muscles were selected as the ROIs for the paraspinal muscles. A total of 7 259 radiomics features were extracted from these ROIs. The dataset was divided into a training set and a test set in an 8∶2 ratio by simple random sampling (osteoporosis group 241 and 61 cases, osteopenia group 211 and 53 cases). The T-score was used to establish the clinical model. After feature normalization and dimensionality reduction, logistic regression was applied to build three radiomics models: vertebral model, paraspinal muscle model, and vertebral-paraspinal muscle model. The T-score was then combined with the radiomics model that achieved the highest area under the receiver operating characteristic curve (AUC) in the test set to construct a clinical-radiomics combined model. Model performance was evaluated using the AUC. The DeLong test was used to compare the diagnostic efficacy between models. Results:In the test set, the vertebral-paravertebral muscle model achieved the highest AUC among radiomics models and was selected for combination with the T-score. In identifying potential vertebral fragility fractures of osteoporosis group, the AUC (95% CI) of the clinical model, vertebral model, paraspinal muscle model, vertebral-paraspinal muscle model, and clinical-radiomics model were 0.523 (0.373-0.672), 0.869 (0.779-0.959), 0.608 (0.464-0.752), 0.876 (0.791-0.961), and 0.860 (0.769-0.952), respectively. For osteopenia group, the corresponding AUC(95% CI) were 0.625 (0.467-0.783), 0.696 (0.547-0.845), 0.706 (0.563-0.848), 0.816 (0.702-0.930), and 0.820 (0.710-0.930). The DeLong test showed that the vertebral model for identifying the potential vertebral fracture risk in osteoporosis group had better performance than the paraspinal muscle model ( Z=3.28, P=0.001). While for osteopenia group, there was no significant difference in diagnostic performance between the vertebral model and the paraspinal muscle model ( Z=0.09, P=0.932). The recognition efficacy of the clinical model and the vertebral-paraspinal muscle model was significantly different ( Z=3.69, 1.98; P<0.001, P=0.047), while there was no significant difference between the clinical-radiomics combined model and the vertebral-paraspinal muscle model ( Z=1.51, 0.12; P=0.131, 0.904). Conclusion:The MRI-based vertebral-paraspinal muscle radiomics model can effectively identify osteoporosis or osteopenia patients with potential fragility fracture risk. In osteopenia group, the efficacy of the MRI radiomics models based on the vertebra and paraspinal muscles in identifying potential vertebral fragility fracture risk is comparable.

Objective: To investigate the mechanism of Yishen Tongluo Formula in ameliorating renal pyroptosis in diabetic nephropathy mice by regulating the toll-like receptor 4 (TLR4)/myeloid differentiation factor 88 (MyD88)/nuclear factor-κB (NF-κB) signaling pathway. Methods: Sixty C57BL/6 male mice were randomly divided into control (10 mice) and intervention groups (50 mice) using random number table method. The diabetes nephropathy model was established by intraperitoneally injecting streptozotocin(50 mg/kg). After modeling, the intervention group was further divided into model, semaglutide (40 μg/kg), and high-, medium-, and low-dose Yishen Tongluo Formula groups (15.6, 7.8, and 3.9 g/kg, respectively) using random number table method. The high-, medium-, and low-dose Yishen Tongluo Formula groups were administered corresponding doses of medication by gavage, the semaglutide group received a subcutaneous injection of semaglutide injection, and the control group and model groups were administered distilled water by gavage for 12 consecutive weeks. Random blood glucose levels of mice in each group were monitored, and the 24-h urinary protein content was measured using biochemical method every 4 weeks; after treatment, the serum creatinine and urea nitrogen levels were measured using biochemical method. The weight of the kidneys was measured, and the renal index was calculated. Hematoxylin and eosin, periodic acid-Schiff, periodic Schiff-methenamine, and Masson staining were used to observe the pathological changes in renal tissue. An enzyme-linked immunosorbent assay was used to detect urinary β2-microglobulin (β2-MG), neutrophil gelatinase-associated lipocalin (NGAL), and kidney injury molecule-1 (KIM-1) levels. Western blotting and real-time fluorescence PCR were used to detect the relative protein and mRNA expression levels of nucleotide-binding domain leucine-rich repeat and pyrin domain-containing receptor 3 (NLRP3), Caspase-1, gasdermin D (GSDMD), interleukin-1β (IL-1β), and interleukin-18 (IL-18) in renal tissue. Immunohistochemistry was used to detect the proportion of protein staining area of the TLR4, MyD88, and NF-κB in renal tissue. Results: Compared with the control group, the random blood glucose, 24-h urinary protein, serum creatinine, urea nitrogen, and renal index of the model group increased, and the urine β2-MG, NGAL, and KIM-1 levels increased. The relative protein and mRNA expression levels of NLRP3, Caspase-1, GSDMD, IL-1β, and IL-18 in renal tissue increased, and the proportion of TLR4, MyD88, and NF-κB protein positive staining areas increased (P<0.05). Pathological changes such as glomerular hypertrophy were observed in the renal tissue of the model group. Compared with the model group, the Yishen Tongluo Formula high-dose group showed a decrease in random blood glucose after 12 weeks of treatment (P<0.05). The Yishen Tongluo Formula high- and medium-dose groups showed a decrease in 24-h urinary protein, creatinine, urea nitrogen, and renal index, as well as decreased β2-MG, NGAL, and KIM-1 levels. NLRP3, Caspase-1, GSDMD, IL-1 β, and IL-18 relative protein and mRNA expression levels were also reduced, and the proportion of TLR4, MyD88, and NF-κB protein positive staining areas was reduced (P<0.05). Pathological damage to renal tissue was ameliorated. Conclusion Yishen Tongluo Formula may exert protective renal effects by inhibiting renal pyroptosis and alleviating tubular interstitial injury in diabetic nephropathy mice by regulating the TLR4/MyD88/NF-κB signaling pathway.

AIM: To investigate the potential inhibitory effect of pterostilbene on the endothelial-to-mesenchymal transition(EndMT)induced by high glucose conditions in human retinal microvascular endothelial cells(HRMECs).METHODS: The optimal concentration of pterostilbene for treating HRMECs was determined using the CCK-8 assay, with 12.5 and 25 μmol/L concentrations selected for subsequent experiments. Four experimental groups were established: control group, high glucose group, high glucose combined with 12.5 μmol/L pterostilbene treatment group, and high glucose combined with 25 μmol/L pterostilbene treatment group. The expression levels of HDAC7 and EndMT-associated markers were detected via Western blot analysis. Cell migration ability was assessed using Transwell migration assays and scratch wound healing tests, while vasculogenic capability was evaluated through tube formation assays.RESULTS: The CCK-8 assay revealed that pterostilbene at a concentration of 22.07 μmol/L inhibited 50% of cell viability in HRMECs. Western blot analysis demonstrated that compared with the control group, the expression levels of HDAC7, ZEB1, Vimentin, and Snail were significantly upregulated in HRMECs cultured in high glucose(all P<0.01), while the expressions of VE-cadherin and CD31 were significantly reduced(all P<0.01). Compared to the high glucose group, the treatment with 12.5 and 25 μmol/L pterostilbene significantly reduced the expression of HDAC7, ZEB1, Vimentin, and Snail under high glucose conditions(all P<0.01). Notably, 25 μmol/L pterostilbene enhanced the expression of VE-cadherin and CD31(all P<0.01). Scratch wound healing tests revealed that HRMECs treated with high glucose exhibited a significantly increased cell migration rate compared to the control group(P<0.05), while the application of 25 μmol/L pterostilbene significantly suppressed HRMECs migration under high glucose conditions(P<0.01). Transwell migration assays demonstrated that the cell migration rate in the high glucose group was significantly higher than that in the control group(P<0.01), with cell migration rate markedly reduced following treatment with both of 12.5 and 25 μmol/L pterostilbene(all P<0.01). The tube formation assay revealed that the ability of HRMECs to form tubular structures was significantly enhanced under high glucose conditions(P<0.01), and both 12.5 and 25 μmol/L of pterostilbene effectively inhibited this effect(all P<0.01).CONCLUSION: Pterostilbene can inhibit HDAC7 expression, suppress EndMT-mediated migration of HRMECs, and impair tube formation under high-glucose conditions.

AIM: To investigate the potential inhibitory effect of pterostilbene on the endothelial-to-mesenchymal transition(EndMT)induced by high glucose conditions in human retinal microvascular endothelial cells(HRMECs).METHODS: The optimal concentration of pterostilbene for treating HRMECs was determined using the CCK-8 assay, with 12.5 and 25 μmol/L concentrations selected for subsequent experiments. Four experimental groups were established: control group, high glucose group, high glucose combined with 12.5 μmol/L pterostilbene treatment group, and high glucose combined with 25 μmol/L pterostilbene treatment group. The expression levels of HDAC7 and EndMT-associated markers were detected via Western blot analysis. Cell migration ability was assessed using Transwell migration assays and scratch wound healing tests, while vasculogenic capability was evaluated through tube formation assays.RESULTS: The CCK-8 assay revealed that pterostilbene at a concentration of 22.07 μmol/L inhibited 50% of cell viability in HRMECs. Western blot analysis demonstrated that compared with the control group, the expression levels of HDAC7, ZEB1, Vimentin, and Snail were significantly upregulated in HRMECs cultured in high glucose(all P<0.01), while the expressions of VE-cadherin and CD31 were significantly reduced(all P<0.01). Compared to the high glucose group, the treatment with 12.5 and 25 μmol/L pterostilbene significantly reduced the expression of HDAC7, ZEB1, Vimentin, and Snail under high glucose conditions(all P<0.01). Notably, 25 μmol/L pterostilbene enhanced the expression of VE-cadherin and CD31(all P<0.01). Scratch wound healing tests revealed that HRMECs treated with high glucose exhibited a significantly increased cell migration rate compared to the control group(P<0.05), while the application of 25 μmol/L pterostilbene significantly suppressed HRMECs migration under high glucose conditions(P<0.01). Transwell migration assays demonstrated that the cell migration rate in the high glucose group was significantly higher than that in the control group(P<0.01), with cell migration rate markedly reduced following treatment with both of 12.5 and 25 μmol/L pterostilbene(all P<0.01). The tube formation assay revealed that the ability of HRMECs to form tubular structures was significantly enhanced under high glucose conditions(P<0.01), and both 12.5 and 25 μmol/L of pterostilbene effectively inhibited this effect(all P<0.01).CONCLUSION: Pterostilbene can inhibit HDAC7 expression, suppress EndMT-mediated migration of HRMECs, and impair tube formation under high-glucose conditions.

ObjectiveTo investigate the value of different noninvasive diagnostic models in the diagnosis of esophageal and gastric varices since there is a high risk of esophageal and gastric varices in patients with compensated hepatitis B cirrhosis and significant portal hypertension， and to provide a basis for the early diagnosis of esophageal and gastric varices. MethodsA total of 108 patients with significant portal hypertension due to compensated hepatitis B cirrhosis who attended Guangdong Provincial Hospital of Traditional Chinese Medicine from November 2017 to November 2023 were enrolled， and according to the presence or absence of esophageal and gastric varices under gastroscopy， they were divided into esophageal and gastric varices group （GOV group） and non-esophageal and gastric varices group （NGOV group）. Related data were collected， including age， sex， imaging findings， and laboratory markers. The chi-square test was used for comparison of categorical data between groups； the least significant difference t-test was used for comparison of normally distributed continuous data between groups， and the Mann-Whitney U test was used for comparison of non-normally distributed continuous data between groups. The receiver operating characteristic （ROC） curve was plotted to evaluate the diagnostic value of five scoring models， i.e.， fibrosis-4 （FIB-4）， LOK index， LPRI， aspartate aminotransferase-to-platelet ratio index （APRI）， and aspartate aminotransferase/alanine aminotransferase ratio （AAR）. The binary logistic regression method was used to establish a combined model， and the area under the ROC curve （AUC） was compared between the combined model and each scoring model used alone. The Delong test was used to compare the AUC value between any two noninvasive diagnostic models. ResultsThere were 55 patients in the GOV group and 53 patients in the NGOV group. Compared with the NGOV group， the GOV group had a significantly higher age （52.64±1.44 years vs 47.96±1.68 years， t=0.453， P<0.05） and significantly lower levels of alanine aminotransferase ［42.00 （24.00 — 17.00） U/L vs 82.00 （46.00 — 271.00） U/L， Z=-3.065， P<0.05］， aspartate aminotransferase ［44.00 （32.00 — 96.00） U/L vs 62.00 （42.50 — 154.50） U/L，Z=-2.351， P<0.05］， and platelet count ［100.00 （69.00 — 120.00）×10⁹/L vs 119.00 （108.50 — 140.50）×10⁹/L， Z=-3.667， P<0.05］. The ROC curve analysis showed that FIB-4， LOK index， LPRI， and AAR used alone had an accuracy of 0.667， 0.681， 0.730， and 0.639， respectively， in the diagnosis of esophageal and gastric varices （all P<0.05）， and the positive diagnostic rates of GOV were 69.97%， 65.28%， 67.33%， and 58.86%， respectively， with no significant differences in AUC values （all P>0.05）， while APRI used alone had no diagnostic value （P>0.05）. A combined model （LAF） was established based on the binary logistic regression analysis and had an AUC of 0.805 and a positive diagnostic rate of GOV of 75.80%， with a significantly higher AUC than FIB-4， LOK index， LPRI， and AAR used alone （Z=-2.773，-2.479，-2.206， and-2.672， all P<0.05）. ConclusionFIB-4， LOK index， LPRI， and AAR have a similar diagnostic value for esophageal and gastric varices in patients with compensated hepatitis B cirrhosis and significant portal hypertension， and APRI alone has no diagnostic value. The combined model LAF had the best diagnostic efficacy， which provides a certain reference for clinical promotion and application.

Objective:To explore the application value of radiomics models based on MRI of vertebrae and paravertebral muscles in identifying potential vertebral fragility fracture risk in osteoporosis and osteopenia.Methods:This cross-sectional study collected data from patients who underwent both dual-energy X-ray absorptiometry (DXA) and lumbar MRI at the Third Affiliated Hospital of Southern Medical University between January 2014 and December 2023,retrospectively. Based on DXA results, patients were categorized into osteoporosis group ( n=302) and osteopenia group ( n=264), with fracture and non-fracture patients matched at 1∶1 ratio by propensity score matching based on age, gender, and body mass index. The fourth lumbar vertebra was selected as the region of interest (ROI) for the vertebral body, and the bilateral psoas major, erector spinae, and multifidus muscles were selected as the ROIs for the paraspinal muscles. A total of 7 259 radiomics features were extracted from these ROIs. The dataset was divided into a training set and a test set in an 8∶2 ratio by simple random sampling (osteoporosis group 241 and 61 cases, osteopenia group 211 and 53 cases). The T-score was used to establish the clinical model. After feature normalization and dimensionality reduction, logistic regression was applied to build three radiomics models: vertebral model, paraspinal muscle model, and vertebral-paraspinal muscle model. The T-score was then combined with the radiomics model that achieved the highest area under the receiver operating characteristic curve (AUC) in the test set to construct a clinical-radiomics combined model. Model performance was evaluated using the AUC. The DeLong test was used to compare the diagnostic efficacy between models. Results:In the test set, the vertebral-paravertebral muscle model achieved the highest AUC among radiomics models and was selected for combination with the T-score. In identifying potential vertebral fragility fractures of osteoporosis group, the AUC (95% CI) of the clinical model, vertebral model, paraspinal muscle model, vertebral-paraspinal muscle model, and clinical-radiomics model were 0.523 (0.373-0.672), 0.869 (0.779-0.959), 0.608 (0.464-0.752), 0.876 (0.791-0.961), and 0.860 (0.769-0.952), respectively. For osteopenia group, the corresponding AUC(95% CI) were 0.625 (0.467-0.783), 0.696 (0.547-0.845), 0.706 (0.563-0.848), 0.816 (0.702-0.930), and 0.820 (0.710-0.930). The DeLong test showed that the vertebral model for identifying the potential vertebral fracture risk in osteoporosis group had better performance than the paraspinal muscle model ( Z=3.28, P=0.001). While for osteopenia group, there was no significant difference in diagnostic performance between the vertebral model and the paraspinal muscle model ( Z=0.09, P=0.932). The recognition efficacy of the clinical model and the vertebral-paraspinal muscle model was significantly different ( Z=3.69, 1.98; P<0.001, P=0.047), while there was no significant difference between the clinical-radiomics combined model and the vertebral-paraspinal muscle model ( Z=1.51, 0.12; P=0.131, 0.904). Conclusion:The MRI-based vertebral-paraspinal muscle radiomics model can effectively identify osteoporosis or osteopenia patients with potential fragility fracture risk. In osteopenia group, the efficacy of the MRI radiomics models based on the vertebra and paraspinal muscles in identifying potential vertebral fragility fracture risk is comparable.