ObjectiveThe essence of syndrome manifestation recognition in traditional Chinese medicine (TCM) is to infer the body’s latent pathogenesis state from clinical observational information, rather than to perform simple label matching. However, previous studies have largely modeled this task as syndrome pattern classification within a fixed label space, which does not adequately reflect the cognition process of TCM syndrome differentiation centered on pathogenesis reasoning, and is also insufficient to capture the openness, semantic variability, and cross-disease reusability of syndrome manifestation expression. This study aimed to investigate whether introducing pathogenesis reasoning chain-of-thought (PR-CoT) supervision into large language models (LLMs) could improve the quality and cognitive consistency of syndrome manifestation recognition and support cross-disease transfer. MethodsSyndrome manifestation recognition was formulated as a conditional generation task under the framework of clinical observational information (X)→pathogenesis structure (Z)→syndrome pattern output (Y), where Z serves as an explicit intermediate structural variable linking the clinical evidence and syndrome judgment. Within this framework, a PR-CoT-supervised dataset for syndrome manifestation recognition was constructed based on medical case records of spleen-stomach disorders. After preprocessing, information extraction, manual proofreading, and data cleaning, the dataset comprised 4 800 training cases, 400 development cases, and 400 test cases. Each sample was annotated with a structured PR-CoT consisting of three progressive levels: clinical information summarization, comprehensive pathogenesis analysis, and syndrome pattern output. Supervised fine-tuning was conducted on open-source LLMs, with an end-to-end model serving as the baseline. Qwen3-32B was used as the primary experimental model, and Qwen3-14B as the scale comparison model. A progressive multidimensional evaluation framework was further established, comprising a structural parsing level, a semantic similarity level, and an expert blind review level. At the structural parsing level, syndrome pattern expressions were decomposed into structural elements and evaluated using Precision, Recall, F1 score, and Jaccard similarity. At the semantic similarity level, independent LLMs scored the theoretical proximity between predicted and reference syndrome patterns. At the expert blind review level, three TCM experts independently evaluated model outputs on two dimensions: syndrome differentiation consistency and terminology standardization of syndrome patterns. In addition, zero-shot cross-disease transfer evaluation was conducted on gynecological and heart-system disorder test sets. ResultsAt the structural parsing level, PR-CoT supervision did not lead to a stable improvement in the element-wise overlap of syndrome pattern structural components. Compared with the corresponding baselines, neither Qwen3-32B nor Qwen3-14B showed consistent advantages in structural matching metrics after the introduction of PR-CoT supervision. In contrast, at the semantic similarity level, PR-CoT supervision produced stable positive gains across different model scales and evaluation systems. The average semantic score of Qwen3-32B increased from 6.425 8 in the baseline model to 6.585 0 after PR-CoT supervision, and that of Qwen3-14B increased from 5.870 0 to 5.964 2. At the expert blind review level, the overall score of Qwen3-32B (PR-CoT) was 7.026 0±0.107 7, higher than 6.416 3±0.288 9 for its baseline. In zero-shot cross-disease testing, the PR-CoT model still showed advantages in semantic evaluation and expert evaluation on both gynecological and heart-system disorder test sets, indicating a certain degree of transferability. ConclusionThe benefits of PR-CoT supervision are mainly reflected in TCM semantic consistency and clinical plausibility, rather than in improved hard matching of structural elements. These findings support understanding syndrome manifestation recognition as a process of generating and expressing latent pathogenesis structures, rather than as a classification task within a traditional fixed label space. By introducing pathogenesis reasoning as an explicit intermediate structure into the modeling process and combining it with a progressive multidimensional evaluation framework, this study provides a methodological pathway for intelligent TCM syndrome differentiation that integrates theoretical alignment, interpretability, and multi-level evaluation.

ObjectiveThe essence of syndrome manifestation recognition in traditional Chinese medicine (TCM) is to infer the body’s latent pathogenesis state from clinical observational information, rather than to perform simple label matching. However, previous studies have largely modeled this task as syndrome pattern classification within a fixed label space, which does not adequately reflect the cognition process of TCM syndrome differentiation centered on pathogenesis reasoning, and is also insufficient to capture the openness, semantic variability, and cross-disease reusability of syndrome manifestation expression. This study aimed to investigate whether introducing pathogenesis reasoning chain-of-thought (PR-CoT) supervision into large language models (LLMs) could improve the quality and cognitive consistency of syndrome manifestation recognition and support cross-disease transfer. MethodsSyndrome manifestation recognition was formulated as a conditional generation task under the framework of clinical observational information (X)→pathogenesis structure (Z)→syndrome pattern output (Y), where Z serves as an explicit intermediate structural variable linking the clinical evidence and syndrome judgment. Within this framework, a PR-CoT-supervised dataset for syndrome manifestation recognition was constructed based on medical case records of spleen-stomach disorders. After preprocessing, information extraction, manual proofreading, and data cleaning, the dataset comprised 4 800 training cases, 400 development cases, and 400 test cases. Each sample was annotated with a structured PR-CoT consisting of three progressive levels: clinical information summarization, comprehensive pathogenesis analysis, and syndrome pattern output. Supervised fine-tuning was conducted on open-source LLMs, with an end-to-end model serving as the baseline. Qwen3-32B was used as the primary experimental model, and Qwen3-14B as the scale comparison model. A progressive multidimensional evaluation framework was further established, comprising a structural parsing level, a semantic similarity level, and an expert blind review level. At the structural parsing level, syndrome pattern expressions were decomposed into structural elements and evaluated using Precision, Recall, F1 score, and Jaccard similarity. At the semantic similarity level, independent LLMs scored the theoretical proximity between predicted and reference syndrome patterns. At the expert blind review level, three TCM experts independently evaluated model outputs on two dimensions: syndrome differentiation consistency and terminology standardization of syndrome patterns. In addition, zero-shot cross-disease transfer evaluation was conducted on gynecological and heart-system disorder test sets. ResultsAt the structural parsing level, PR-CoT supervision did not lead to a stable improvement in the element-wise overlap of syndrome pattern structural components. Compared with the corresponding baselines, neither Qwen3-32B nor Qwen3-14B showed consistent advantages in structural matching metrics after the introduction of PR-CoT supervision. In contrast, at the semantic similarity level, PR-CoT supervision produced stable positive gains across different model scales and evaluation systems. The average semantic score of Qwen3-32B increased from 6.425 8 in the baseline model to 6.585 0 after PR-CoT supervision, and that of Qwen3-14B increased from 5.870 0 to 5.964 2. At the expert blind review level, the overall score of Qwen3-32B (PR-CoT) was 7.026 0±0.107 7, higher than 6.416 3±0.288 9 for its baseline. In zero-shot cross-disease testing, the PR-CoT model still showed advantages in semantic evaluation and expert evaluation on both gynecological and heart-system disorder test sets, indicating a certain degree of transferability. ConclusionThe benefits of PR-CoT supervision are mainly reflected in TCM semantic consistency and clinical plausibility, rather than in improved hard matching of structural elements. These findings support understanding syndrome manifestation recognition as a process of generating and expressing latent pathogenesis structures, rather than as a classification task within a traditional fixed label space. By introducing pathogenesis reasoning as an explicit intermediate structure into the modeling process and combining it with a progressive multidimensional evaluation framework, this study provides a methodological pathway for intelligent TCM syndrome differentiation that integrates theoretical alignment, interpretability, and multi-level evaluation.

ObjectiveProtein-protein interactions (PPIs) are fundamental to the execution of biological functions within living cells. However, traditional biochemical methods, such as co-immunoprecipitation (Co-IP), often fail to capture transient, weak, or membrane-associated interactions due to the stringent detergent requirements for cell lysis. Proximity labeling (PL) has emerged in recent years as a transformative technology for mapping the proteomes of specific subcellular compartments and identifying dynamic interactomes in situ. Golgi protein 73 (GP73, also known as GOLPH2), a resident type II Golgi transmembrane protein, is a well-recognized clinical biomarker for liver diseases, including hepatocellular carcinoma (HCC). Despite its clinical significance, the comprehensive physiological and pathological functions of GP73 remain partially understood. This study aims to establish an APEX2-mediated proximity labeling system specifically targeting GP73 to map its interactome in a living cellular environment, thereby providing new insights into its molecular roles and regulatory mechanisms. MethodsTo achieve spatial specificity, we first constructed a stable cell line expressing a fusion protein consisting of GP73 and the engineered soybean peroxidase APEX2. The localization of the GP73-APEX2 fusion protein was validated to ensure it correctly targeted the Golgi apparatus. The proximity labeling reaction was initiated by incubating the cells with biotin-phenol (BP) for 30 min, followed by a brief (1 min) treatment with1 mmol/L hydrogen peroxide (H₂O₂). This catalytic reaction converts BP into highly reactive, short-lived biotin-phenoxyl radicals that covalently attach to endogenous proteins within a small labeling radius of the GP73-APEX2 enzyme. Subsequently, the cells were quenched, and biotinylated proteins were enriched using high-affinity streptavidin-coated magnetic beads. The captured “neighbor” proteins were subjected to on-bead digestion and analyzed via liquid chromatography-tandem mass spectrometry (LC-MS/MS) for high-throughput identification. Rigorous bioinformatics analysis, including Gene Ontology (GO) enrichment, Kyoto Encyclopedia of Genes and Genomes (KEGG) pathway analysis, and protein-protein interaction network mapping, was performed to interpret the biological significance of the identified candidates. ResultsOur results demonstrate the successful establishment of a robust and sensitive APEX2-based proximity labeling system for GP73. We identified a total of 95 high-confidence interacting proteins that were significantly enriched in the GP73 proximity proteome compared to control groups. Bioinformatics analysis revealed that these interactors were predominantly associated with biological processes such as vesicular transport, protein localization, and, most notably, molecular functions related to “ribosome binding” and “translation regulation”. This suggested an unexpected role for the Golgi-resident GP73 in the cellular translation machinery. To validate these findings, we performed targeted biochemical assays which confirmed a direct interaction between GP73 and the subunits of the eukaryotic translation initiation factor 3 (eIF3) complex, specifically EIF3G and EIF3I. Furthermore, functional validation using the surface sensing of translation (SUnSET) assay—a non-radioactive method to monitor protein synthesis—revealed that the overexpression of GP73 significantly promoted global protein translation levels in the cell, whereas its depletion or inhibition resulted in reduced translation efficiency. ConclusionThis study successfully utilized APEX2-mediated proximity labeling to provide the first systematic map of GP73 interactome in living cells. Our findings uncover a novel, unconventional function of GP73 as a regulator of cellular protein translation, likely mediated through its interaction with the eIF3 complex. This discovery significantly broadens our understanding of the biological roles of GP73 beyond its traditional function in the Golgi apparatus and suggests that it may act as a bridge between Golgi-related trafficking and the protein synthesis machinery. Furthermore, the technical framework established in this study provides a valuable template for investigating other complex organelle-associated protein networks and resolving transient macromolecular interactions in various physiological and pathological contexts.

ObjectiveCerebral ischemic injury triggers a complex pathological cascade characterized by excessive reactive oxygen species (ROS) accumulation, persistent oxidative stress, and sustained neuroinflammation in the injured brain microenvironment. These events collectively drive mitochondrial dysfunction, microglial overactivation, pro-inflammatory cytokine release, and progressive neuronal apoptosis, ultimately leading to severe and irreversible neurological deficits. However, conventional therapeutic strategies face critical limitations, including poor blood-brain barrier penetration, insufficient local drug concentration, uncontrolled drug release, and off-target systemic side effects. To address this pathological process, we rationally designed and fabricated an injectable ROS-responsive hydrogel loaded with polydopamine nanoparticles (PDA NPs) for spatiotemporally controlled antioxidation, anti-inflammation, and neuroprotection in the ischemic injury microenvironment. The present study aimed to systematically characterize the physicochemical properties, ROS-responsive drug release behavior, biocompatibility, and neuroprotective efficacy of this composite hydrogel system in vitro. MethodsPDA NPs were fabricated via oxidative self-polymerization. The ROS-responsive hydrogel was cross-linked using N1-(4-boronobenzyl)-N3-(4-boronophenyl)-N1,N1,N3,N3-tetramethylpropane-1, 3-diaminium (TSPBA) and polyvinyl alcohol (PVA). Morphology, particle size, Zeta potential, and structure of PDA NPs were characterized by dynamic light scattering (DLS), Zeta potential analysis, scanning electron microscopy (SEM), and transmission electron microscopy (TEM). Microstructure, rheological properties, shear-thinning behavior, and ROS-triggered release profiles of the hydrogel were examined by SEM and rheometry. Biocompatibility was evaluated using HT22 mouse hippocampal neurons with CCK-8 and live/dead staining. An oxygen-glucose deprivation/reoxygenation (OGD/R) model was established to simulate ischemic injury in vitro. ROS levels and neuronal apoptosis were detected by DHE staining and TUNEL assay. Microglial polarization and pro-inflammatory cytokine expression were analyzed using immunofluorescence and RT-qPCR in BV-2 microglia. Transwell co-culture was used to verify the indirect neuroprotection mediated by modulated microglia. ResultsCharacterization results confirmed that the as-prepared PDA NPs were monodispersed spherical nanoparticles with uniform diameter and negative surface potential, demonstrating favorable dispersibility and robust ROS-scavenging activity. The TSPBA-PVA hydrogel exhibited a highly porous interconnected network, suitable mechanical strength, and obvious shear-thinning behavior, supporting its application as an injectable implant. More importantly, the hydrogel displayed typical ROS-responsive degradation and on-demand PDA NP release in a ROS-concentration-dependent manner. In vitro cellular experiments demonstrated that the PDA NP-loaded hydrogel possessed excellent biocompatibility with HT22 cells. In the OGD/R model, the hydrogel significantly reduced intracellular ROS accumulation and markedly suppressed neuronal apoptosis. Furthermore, the composite hydrogel effectively redirected BV-2 microglia from the pro-inflammatory M1 toward the anti-inflammatory M2 phenotypes, downregulated the expression of pro-inflammatory cytokines including TNF-α, IL-1β, and IL-6, and reduced inflammatory damage. Transwell co-culture assays further validated that M2-polarized microglia mediated by the hydrogel significantly enhanced the survival of OGD/R-injured HT22 neurons and attenuated apoptosis. ConclusionIn this study, we successfully developed a novel injectable ROS-responsive hydrogel loaded with PDA NPs for synergistic antioxidative and anti-inflammatory neuroprotection. This intelligent hydrogel system enables ROS-triggered on-demand release of PDA NPs, efficiently scavenges excessive ROS, inhibits oxidative stress injury, modulates microglial polarization, and suppresses neuroinflammation, thereby exerting robust neuroprotective effects in vitro. This biomaterial platform provides a promising strategy for the targeted and controlled delivery of bioactive nanomaterials in the central nervous system diseases and establishes a solid experimental foundation for the development of in situ injectable therapies for ischemic brain injury.

Objective To develop a deep learning model based on fundus retinal images to improve the detection rate of mild cognitive impairment(MCI)in patients with coronary heart disease,achieve early intervention and improve prognosis.Methods The study was a single-center cross-sectional study that retrospectively included patients diagnosed with coronary heart disease(CHD)by coronary angiography(≥50％ stenosis of at least one coronary vessel)from Beijing Anzhen Hospital between November 2021 and December 2022.The whole data set was randomly divided into the training set and the testing set according to the ratio of 8∶2 for model development.After that,the patient data of the same center from January 2023 to April 2023 were included in the time verification method to verify the model.The diagnostic criteria for MCI were MMSE＜27 or MoCA＜26.Four kinds of convolutional neural network(CNN)architectures were used to train fundus images,and a comprehensive vision model of MCI detection was established through model integration.The area under the curve(AUC),sensitivity and specificity of the receiver operating curve(ROC)were used to evaluate the performance of the AI model.Results We collected 5 880 eligible fundus images from 3 368 CHD patients.Based on the results of the MMSE scale,the algorithm was labeled,including 2 898 males and 527 MCI patients.The AUC of the deep learning model in the test group is 0.733(95％CI 0.688-0.778),and the sensitivity of the algorithm in the test group is 0.577(95％CI 0.528-0.625)by using the operating point with the maximum sum of sensitivity and specificity.With a specificity of 0.758(95％CI 0.714-0.802),corresponding to a validated AUC of 0.710(95％CI 0.601-0.818).Based on the results of the MoCA scale,the algorithm labels 2 437 males and 1 626 MCI patients.The AUC of the deep learning model in the test group was 0.702(95％CI 0.671-0.733).The operating point with the maximum sum of sensitivity and specificity was selected,and the sensitivity of the algorithm was 0.749(95％CI 0.719-0.778)and the specificity was 0.561(95％CI 0.527-0.595),corresponding to the AUC value of the verification group was 0.674(95％CI 0.622-0.726).Conclusions The deep learning algorithm model based on fundus images has good diagnostic performance,and may be used as a new non-invasive,convenient and rapid screening method for MCI in CHD population.

Objective:To investigate the value of gadoxetate disodium-enhanced MRI combined with T 1 mapping in preoperative prediction of cytokeratin 19 (CK19) and glypican-3 (GPC3) positive dual-phenotype hepatocellular carcinoma (DPHCC). Methods:This case-control study included retrospectively enrolled patients with pathologically confirmed HCC from Central People′s Hospital of Zhanjiang (training set, n=85; December 2020 to July 2022) and the Second Affiliated Hospital, School of Medicine, South China University of Technology (test set, n=35; April 2023 to April 2024). Patients were categorized into CK19 and GPC3 positive DPHCC group (training set=19, test set=11) and non-DPHCC group (training set=66, test set=24) based on postoperative immunohistochemical staining. All patients received preoperative MRI scans, including gadoxetate disodium-enhanced imaging and T 1 mapping. Clinical data were collected, qualitative MRI features were evaluated, and quantitative parameters were measured, including signal intensity, T 1 values, apparent diffusion coefficient (ADC), tumor-to-liver ADC ratio (rADC), tumor-to-liver signal intensity ratio, and T 1 relaxation time reduction rate (ΔT 1%). Statistical comparisons between groups were performed using independent t-tests, Mann-Whitney U tests, or χ2 tests. Multivariate logistic regression identified independent predictors of CK19 and GPC3 positive DPHCC, and a combined model was constructed. Predictive performance was evaluated using area under the receiver operating characteristic curve (AUC), with DeLong test comparing model performance. Results:Significant intergroup differences were observed in alpha-fetoprotein (AFP), total bilirubin, direct bilirubin, DWI target sign, rADC, hepatobiliary-phase T 1 (T 1HBP), and ΔT 1% ( P<0.05). Multivariate analysis identified AFP>20 ng/ml ( OR=5.835, 95% CI 1.019-33.397, P=0.048), DWI target sign ( OR=13.408, 95% CI 2.216-81.131, P=0.005), and ΔT1%≤31% ( OR=14.429, 95% CI 2.166-96.125, P=0.006) as independent predictors of DPHCC. The AUC values of the aforementioned three independent predictors and the combined model for predicting DPHCC were 0.641 (95% CI 0.530-0.742), 0.679 (95% CI 0.569-0.777), 0.740 (95% CI 0.634-0.829), and 0.886 (95% CI 0.799-0.945) in the training set, and 0.568 (95% CI 0.390-0.743), 0.669 (95% CI 0.490-0.818), 0.689 (95% CI 0.511-0.843), and 0.824 (95% CI 0.658-0.931) in the test set, respectively. The DeLong test results showed that in the training set, the diagnostic performance of the combined model was superior to those of the three individual features ( Z=3.68, P<0.001; Z=3.15, P=0.002; Z=3.15, P=0.002). In the test cohort, the combined model demonstrated better diagnostic performance than AFP>20 ng/ml and ΔT 1%≤31% ( Z=2.15, P=0.032; Z=2.12, P=0.034), while no statistically significant difference was observed compared with the DWI target sign ( Z=1.77, P=0.076). Conclusion:The integrated model incorporating clinical data, gadoxetate disodium-enhanced MRI, and T 1 mapping parameters effectively predicts CK19 and GPC3 positive DPHCC.

Protein-protein interactions play an extremely important role in the biochemical functions of cells,and in-depth analysis of protein interactions is the key to understanding cellular life activities.In this study,we systematically mined the interacting proteins of Golgi protein 73(GP73)using classical immunoprecipitation combined with mass spectrometry,and sought to further analyze the molecular func-tion of GP73.Hepatocellular carcinoma cell line HepG2 was selected,and a stable cell line overexpress-ing GP73-3Flag was constructed using lentiviral infection technology.A total of 78 high-confidence GP73 interacting proteins were identified by immunoprecipitation coupled with mass spectrometry.Bioinformat-ics analyses suggested that GP73 interacted with nearly 40 cytosolic proteins and participated in the bio-logical processes of RNA transport,splicing,and translation.Further immunofluorescence and cytosolic protein isolation experiments confirmed the cytosolic localization of GP73 in a variety of tumor cells.Based on the 78 interacting proteins,we further screened protein interaction networks related to mRNA splicing and verified the existence of interactions between GP73 and seven proteins,including HNRN-PH3,SMN1,RBM14,andNCBP1,by co-immunoprecipitation experiments.In addition,minigene spli-cing assay results indicated that GP73 inhibited the splicing efficiency of pre-mRNA by cells.This study contributes to the expansion of knowledge regarding the function of GP73 and aids in elucidating its criti-cal role in cell biology and its potential association with diseases.

The coil adverse events in the U.S.Food and Drug Administration Manufacturer and User Facility Device Experience(MAUDE)database from January 2021 to June 2024 were analyzed retrospectively.The risks of coils during the clinical application and their causes were explored with hospital survey and expert demonstration in Shandong Province.Some improving measures were put forward for the safe use of coils,including implementing the main responsibility of the registrant,enhancing the professional skills of the using institutions and strengthening the supervision of the supervisory authorities.[Chinese Medical Equipment Journal,2025,46(6):83-87]

Objective To develop a deep learning model based on fundus retinal images to improve the detection rate of mild cognitive impairment(MCI)in patients with coronary heart disease,achieve early intervention and improve prognosis.Methods The study was a single-center cross-sectional study that retrospectively included patients diagnosed with coronary heart disease(CHD)by coronary angiography(≥50％ stenosis of at least one coronary vessel)from Beijing Anzhen Hospital between November 2021 and December 2022.The whole data set was randomly divided into the training set and the testing set according to the ratio of 8∶2 for model development.After that,the patient data of the same center from January 2023 to April 2023 were included in the time verification method to verify the model.The diagnostic criteria for MCI were MMSE＜27 or MoCA＜26.Four kinds of convolutional neural network(CNN)architectures were used to train fundus images,and a comprehensive vision model of MCI detection was established through model integration.The area under the curve(AUC),sensitivity and specificity of the receiver operating curve(ROC)were used to evaluate the performance of the AI model.Results We collected 5 880 eligible fundus images from 3 368 CHD patients.Based on the results of the MMSE scale,the algorithm was labeled,including 2 898 males and 527 MCI patients.The AUC of the deep learning model in the test group is 0.733(95％CI 0.688-0.778),and the sensitivity of the algorithm in the test group is 0.577(95％CI 0.528-0.625)by using the operating point with the maximum sum of sensitivity and specificity.With a specificity of 0.758(95％CI 0.714-0.802),corresponding to a validated AUC of 0.710(95％CI 0.601-0.818).Based on the results of the MoCA scale,the algorithm labels 2 437 males and 1 626 MCI patients.The AUC of the deep learning model in the test group was 0.702(95％CI 0.671-0.733).The operating point with the maximum sum of sensitivity and specificity was selected,and the sensitivity of the algorithm was 0.749(95％CI 0.719-0.778)and the specificity was 0.561(95％CI 0.527-0.595),corresponding to the AUC value of the verification group was 0.674(95％CI 0.622-0.726).Conclusions The deep learning algorithm model based on fundus images has good diagnostic performance,and may be used as a new non-invasive,convenient and rapid screening method for MCI in CHD population.

The coil adverse events in the U.S.Food and Drug Administration Manufacturer and User Facility Device Experience(MAUDE)database from January 2021 to June 2024 were analyzed retrospectively.The risks of coils during the clinical application and their causes were explored with hospital survey and expert demonstration in Shandong Province.Some improving measures were put forward for the safe use of coils,including implementing the main responsibility of the registrant,enhancing the professional skills of the using institutions and strengthening the supervision of the supervisory authorities.[Chinese Medical Equipment Journal,2025,46(6):83-87]