Parkinson's disease （PD） is a chronic progressive neurodegenerative disorder primarily characterized by motor dysfunction. The main pathological features include the loss of dopaminergic neurons in the substantia nigra， abnormal aggregation of alpha-Synuclein （α-Syn）， and the formation of Lewy bodies. However， the exact mechanisms remain unclear. In recent years， the PD incidence has gradually increased， while current treatment methods are limited to symptom alleviation， incapable of halting disease progression， and prone to adverse effects， thus making it urgent to search for medicines effective for PD. Modern research indicates that the Kelch-like ECH-associated protein 1 （Keap1）/nuclear factor E₂ related factor 2 （Nrf2）/antioxidant response element （ARE） signaling pathway is closely related to oxidative stress， neuroinflammation， apoptosis， ferroptosis， and mitochondrial dysfunction， playing a crucial role in the pathophysiological development of PD. A large number of studies have further confirmed that traditional Chinese medicine （TCM） can regulate diseases through a holistic view of Syndrome differentiation and microscopic molecular pathways. With unique advantages， such as multiple targets， multiple pathways， and fewer adverse reactions， TCM provides a new strategy for PD treatment. This article elucidates the mechanism of the Keap1/Nrf2/ARE signaling pathway in the occurrence and development of PD， while summarizing the latest research on PD intervention by TCM monomers， active ingredients， and compounds， as well as acupuncture via the precise targeted regulation of the Keap1/Nrf2/ARE pathway， aiming to provide a reference for clinical medicine development to prevent and treat PD.

Rare diseases pose significant diagnostic and therapeutic challenges,carrying a high disease burden,their management critically reflects a nation's public health resilience.Currently,China faces key challenges such as scarce treatments,fragmented services,and low drug accessibility in rare disease care,which urgently require systemic solutions.Digital-intelligent technology as a key breakthrough are expected to resolve the challenges in this field.Although its application in the field of rare diseases is gradually expanding,there is a lack of systematic compilation of studies to elucidate how to precisely enhance the precision,synergy and sustainability of diagnosis and treatment.The key challenges in rare disease care concentrate in four areas:inefficiency in prenatal screening,uneven distribution of medical resources,low efficiency in social organization collaboration,and ineffective information dissemination.The"4C"strategy,based on digital-intelligent technology,can address these issues:①coordination,boost prenatal screening awareness and capacity via digital-intelligent platforms to strengthen prevention;②cooperation,deepen collaboration within specialist networks,empowering institutions to enhance diagnostic capacity;③co-creation,empower support organizations to optimize resources,efficiency;④cognition,minimize information dissipation through efficient platforms,improving patient and family quality of life.This establishes an integrated digital-intelligent rare disease model encompassing"screening-diagnosis-treatment-care".

Objective We apply a class imbalance treatment method that can solve the between-class imbalance problem and the within-class imbalance problem of the minority class and the majority class at the same time.And combining it with RF classifier to achieve early recurrence prediction in DLBLC patients,which provided a reference for the treatment of DLBLC patients.Methods Firstly,we apply a class imbalance processing method based on Gaussian mixture model bi-directional clustering resampling to process the data.And compared with ROS,SMOTE,Borderline-1 SMOTE,Borderline-2 SMOTE,GMM oversampling,GMM undersampling,SMOTE+RUS,SMOTE+GMM and GMM+RUS.Afterwards,in order to verify the performance of RF,we use logistic regression and decision tree models as controls.Finally,the evaluation of the model is carried out in terms of discrimination and calibration.Results The RF model with GMM-GMM resampling achieved relatively optimal classification performance(accuracy=0.79,AUC=0.87,sensitivity=0.71,specificity=0.87,G-means=0.79,MSE=0.21).Conclusion GMM-GMM is superior to other traditional resampling methods,and combining it with the RF model for the prediction of early recurrence in DLBCL patients has achieved relatively good classification results,which can well realize the prediction of early recurrence in DLBCL patients.

Objective To construct a 2-year relapse risk prediction model for 498 patients diagnosed with diffuse large B-cell lymphoma(DLBCL)who achieved complete response(CR)following treatment at the hematology department of a cancer hospital in Shanxi Province between 2011 and 2020,providing a reference for clinical management.Methods The least absolute shrinkage and selection operator(LASSO)feature selection algorithm,combined with clinical expertise,was first used to identify 21 significant variables influencing the 2-year relapse rate in DLBCL patients with CR.To address data imbalance,synthetic minority oversampling technique(SMOTE)and synthetic minority oversampling technique and edited nearest neighbor(SMOTE-ENN)were applied.Relapse predictions were conducted using seven classifiers on both the original and balanced datasets.The deep forest(DF)algorithm was then employed to build the relapse risk prediction model.Model performance was evaluated using accuracy,precision,sensiti vity/recall,specificity,F1-score,and G-means,while calibration was assessed using the Brier score.Results The deep forest algorithm,when combined with the SMOTE-ENN method for data imbalance,achieved the best performance(accuracy=0.932,precision=0.949,recall=0.944,specificity=0.910,F1-score=0.946,G-means=0.926,Brier score=0.068).Conclusion This study successfully combines the SMOTE-ENN technique with the deep forest classifier to predict 2-year relapse risk in DLBCL patients who achieved CR.The model demonstrates excellent performance and meets expectations.

Infectious diseases continue to pose a threat to global public health. Successive global shocks caused by emerging and re-emerging pathogens have continuously challenged existing surveillance systems, highlighting the urgent need to build efficient and precise pathogen surveillance networks. Pathogen genomic databases have been developed rapidly in recent two decades, significantly improving the molecular identification, evolutionary analysis, and transmission tracking of pathogens, and changing disease surveillance strategies and patterns. This paper summarizes the developmental history and current state of pathogen genomic databases, and discusses their applications in public health, including pathogen variation surveillance, emerging or suspected pathogen identification, and epidemiological tracing. Furthermore, this paper systematically analyzes the limitations and key challenges faced by current global health prevention and control system, and suggests the focus of the development of online pathogen databases to address existing shortcomings, ultimately improve global infectious disease surveillance and early warning

Infectious diseases continue to pose a threat to global public health. Successive global shocks caused by emerging and re-emerging pathogens have continuously challenged existing surveillance systems, highlighting the urgent need to build efficient and precise pathogen surveillance networks. Pathogen genomic databases have been developed rapidly in recent two decades, significantly improving the molecular identification, evolutionary analysis, and transmission tracking of pathogens, and changing disease surveillance strategies and patterns. This paper summarizes the developmental history and current state of pathogen genomic databases, and discusses their applications in public health, including pathogen variation surveillance, emerging or suspected pathogen identification, and epidemiological tracing. Furthermore, this paper systematically analyzes the limitations and key challenges faced by current global health prevention and control system, and suggests the focus of the development of online pathogen databases to address existing shortcomings, ultimately improve global infectious disease surveillance and early warning

Objective In this study,clinical data of Alzheimer's disease(AD)patients,structural magnetic resonance imaging(sMRI)data,and positron emission tomography(PET)data were used to construct an auxiliary diagnostic model with good classification effects,so as to formulate a personalized treatment plan at the early stage of the patients,which is of great significance for the prevention and treatment of AD.Methods In this study,a total of 401 study subjects containing complete sMRI images and PET images were selected from the ADNI-1(Alzheimer's disease neuroimaging initiative-1,ADNI-1)database.We used statistical parameters mapping(SPM)and voxel-based morphometric(VBM)analysis of MATLAB to perform pre-processing operations such as spatial normalization and skull stripping on sMRI images and PET images of the study subjects.With the help of the brain atlas was used to segment the brain tissue structure.After that,the segmented gray matter was extracted from the corresponding brain regions based on anatomical automatic labeling,and the feature values of all brain regions were obtained.Then the extracted brain region feature values are then subjected to fisher score,support vector machine-recursive feature elimination(SVM-RFE)and least absolute shrinkage and selection operator(LASSO),a hybrid filtered-wrapped-embedded feature selection method with three different principles,to realize the dimensionality reduction of high-dimensional image data.Finally,the PAC-Bayesian strategy boosting based multi-view learning(PB-MVBoost)model is constructed based on multi-view decision fusion for clinical,sMRI and PET data.And it is compared with the traditional machine learning models support vector machine(SVM),decision tree(DT),K-nearest neighbor(KNN),random forests(RF),adaptive boosting(AdaBoost),and extreme gradient boosting(XGBoost)which are constructed after concatenating views.It is compared with multi-view multi-kernel learning models(AverageMKL,EasyMKL)and multi-view confusion matrix boosting,which is also the same multi-view decision fusion.Results Among all the multi-view fusion models of AD-MCI,the PB-MVBoost model based on decision fusion has the best performance(accuracy=0.98,F1-score=0.97,precision=0.98,recall=0.96,MSE=0.07).Among all the multi-view fusion models of MCI-NC,the model performance of PB-MVBoost based on decision fusion was the best(accuracy=0.99,F1-score=0.98,precision=0.99,recall=0.98,MSE=0.05).Conclusion In the classification of AD-MCI and MCI-NC,the distinction and calibration degree of PB-MVBoost model were optimized,indicating that the auxiliary diagnosis model of Alzheimer's disease constructed by PB-MVBoost classifier based on decision fusion performed the best,which could improve the recognition of patients with mild cognitive impairment and then assist clinical diagnosis.

Objective In this study,clinical data of Alzheimer's disease(AD)patients,structural magnetic resonance imaging(sMRI)data,and positron emission tomography(PET)data were used to construct an auxiliary diagnostic model with good classification effects,so as to formulate a personalized treatment plan at the early stage of the patients,which is of great significance for the prevention and treatment of AD.Methods In this study,a total of 401 study subjects containing complete sMRI images and PET images were selected from the ADNI-1(Alzheimer's disease neuroimaging initiative-1,ADNI-1)database.We used statistical parameters mapping(SPM)and voxel-based morphometric(VBM)analysis of MATLAB to perform pre-processing operations such as spatial normalization and skull stripping on sMRI images and PET images of the study subjects.With the help of the brain atlas was used to segment the brain tissue structure.After that,the segmented gray matter was extracted from the corresponding brain regions based on anatomical automatic labeling,and the feature values of all brain regions were obtained.Then the extracted brain region feature values are then subjected to fisher score,support vector machine-recursive feature elimination(SVM-RFE)and least absolute shrinkage and selection operator(LASSO),a hybrid filtered-wrapped-embedded feature selection method with three different principles,to realize the dimensionality reduction of high-dimensional image data.Finally,the PAC-Bayesian strategy boosting based multi-view learning(PB-MVBoost)model is constructed based on multi-view decision fusion for clinical,sMRI and PET data.And it is compared with the traditional machine learning models support vector machine(SVM),decision tree(DT),K-nearest neighbor(KNN),random forests(RF),adaptive boosting(AdaBoost),and extreme gradient boosting(XGBoost)which are constructed after concatenating views.It is compared with multi-view multi-kernel learning models(AverageMKL,EasyMKL)and multi-view confusion matrix boosting,which is also the same multi-view decision fusion.Results Among all the multi-view fusion models of AD-MCI,the PB-MVBoost model based on decision fusion has the best performance(accuracy=0.98,F1-score=0.97,precision=0.98,recall=0.96,MSE=0.07).Among all the multi-view fusion models of MCI-NC,the model performance of PB-MVBoost based on decision fusion was the best(accuracy=0.99,F1-score=0.98,precision=0.99,recall=0.98,MSE=0.05).Conclusion In the classification of AD-MCI and MCI-NC,the distinction and calibration degree of PB-MVBoost model were optimized,indicating that the auxiliary diagnosis model of Alzheimer's disease constructed by PB-MVBoost classifier based on decision fusion performed the best,which could improve the recognition of patients with mild cognitive impairment and then assist clinical diagnosis.

Objective We apply a class imbalance treatment method that can solve the between-class imbalance problem and the within-class imbalance problem of the minority class and the majority class at the same time.And combining it with RF classifier to achieve early recurrence prediction in DLBLC patients,which provided a reference for the treatment of DLBLC patients.Methods Firstly,we apply a class imbalance processing method based on Gaussian mixture model bi-directional clustering resampling to process the data.And compared with ROS,SMOTE,Borderline-1 SMOTE,Borderline-2 SMOTE,GMM oversampling,GMM undersampling,SMOTE+RUS,SMOTE+GMM and GMM+RUS.Afterwards,in order to verify the performance of RF,we use logistic regression and decision tree models as controls.Finally,the evaluation of the model is carried out in terms of discrimination and calibration.Results The RF model with GMM-GMM resampling achieved relatively optimal classification performance(accuracy=0.79,AUC=0.87,sensitivity=0.71,specificity=0.87,G-means=0.79,MSE=0.21).Conclusion GMM-GMM is superior to other traditional resampling methods,and combining it with the RF model for the prediction of early recurrence in DLBCL patients has achieved relatively good classification results,which can well realize the prediction of early recurrence in DLBCL patients.

Objective To construct a 2-year relapse risk prediction model for 498 patients diagnosed with diffuse large B-cell lymphoma(DLBCL)who achieved complete response(CR)following treatment at the hematology department of a cancer hospital in Shanxi Province between 2011 and 2020,providing a reference for clinical management.Methods The least absolute shrinkage and selection operator(LASSO)feature selection algorithm,combined with clinical expertise,was first used to identify 21 significant variables influencing the 2-year relapse rate in DLBCL patients with CR.To address data imbalance,synthetic minority oversampling technique(SMOTE)and synthetic minority oversampling technique and edited nearest neighbor(SMOTE-ENN)were applied.Relapse predictions were conducted using seven classifiers on both the original and balanced datasets.The deep forest(DF)algorithm was then employed to build the relapse risk prediction model.Model performance was evaluated using accuracy,precision,sensiti vity/recall,specificity,F1-score,and G-means,while calibration was assessed using the Brier score.Results The deep forest algorithm,when combined with the SMOTE-ENN method for data imbalance,achieved the best performance(accuracy=0.932,precision=0.949,recall=0.944,specificity=0.910,F1-score=0.946,G-means=0.926,Brier score=0.068).Conclusion This study successfully combines the SMOTE-ENN technique with the deep forest classifier to predict 2-year relapse risk in DLBCL patients who achieved CR.The model demonstrates excellent performance and meets expectations.