In recent years, the application of artificial intelligence (AI) in the field of biology has witnessed remarkable advancements. Among these, the most notable achievements have emerged in the domain of protein structure prediction and design, with AlphaFold and related innovations earning the 2024 Nobel Prize in Chemistry. These breakthroughs have transformed our ability to understand protein folding and molecular interactions, marking a pivotal milestone in computational biology. Looking ahead, it is foreseeable that the accurate prediction of various physicochemical properties of proteins—beyond static structure—will become the next critical frontier in this rapidly evolving field. One of the most important protein properties is thermodynamic stability, which refers to a protein’s ability to maintain its native conformation under physiological or stress conditions. Accurate prediction of protein stability, especially upon single-point mutations, plays a vital role in numerous scientific and industrial domains. These include understanding the molecular basis of disease, rational drug design, development of therapeutic proteins, design of more robust industrial enzymes, and engineering of biosensors. Consequently, the ability to reliably forecast the stability changes caused by mutations has broad and transformative implications across biomedical and biotechnological applications. Historically, protein stability was assessed via experimental methods such as differential scanning calorimetry (DSC) and circular dichroism (CD), which, while precise, are time-consuming and resource-intensive. This prompted the development of computational approaches, including empirical energy functions and physics-based simulations. However, these traditional models often fall short in capturing the complex, high-dimensional nature of protein conformational landscapes and mutational effects. Recent advances in machine learning (ML) have significantly improved predictive performance in this area. Early ML models used handcrafted features derived from sequence and structure, whereas modern deep learning models leverage massive datasets and learn representations directly from data. Deep neural networks (DNNs), graph neural networks (GNNs), and attention-based architectures such as transformers have shown particular promise. GNNs, in particular, excel at modeling spatial and topological relationships in molecular structures, making them well-suited for protein modeling tasks. Furthermore, attention mechanisms enable models to dynamically weigh the contribution of specific residues or regions, capturing long-range interactions and allosteric effects. Nevertheless, several key challenges remain. These include the imbalance and scarcity of high-quality experimental datasets, particularly for rare or functionally significant mutations, which can lead to biased or overfitted models. Additionally, the inherently dynamic nature of proteins—their conformational flexibility and context-dependent behavior—is difficult to encode in static structural representations. Current models often rely on a single structure or average conformation, which may overlook important aspects of stability modulation. Efforts are ongoing to incorporate multi-conformational ensembles, molecular dynamics simulations, and physics-informed learning frameworks into predictive models. This paper presents a comprehensive review of the evolution of protein thermodynamic stability prediction techniques, with emphasis on the recent progress enabled by machine learning. It highlights representative datasets, modeling strategies, evaluation benchmarks, and the integration of structural and biochemical features. The aim is to provide researchers with a structured and up-to-date reference, guiding the development of more robust, generalizable, and interpretable models for predicting protein stability changes upon mutation. As the field moves forward, the synergy between data-driven AI methods and domain-specific biological knowledge will be key to unlocking deeper understanding and broader applications of protein engineering.

ObjectiveExosomes are microvesicles which could be secreted by all cell types with diameters between 30 and 150 nm. It was widely distributed in body fluids including blood, urine, and breast milk. Exosomes are considered as potential biomarkers and drug carriers by reason of containing nucleic acids, lipids, proteins and other bioactive molecules. Milk-derived exosomes have been widely used as drug delivery carriers to treat targeted diseases with a lower cost, higher biocompatibility and lower immunogenicity. Until now, there is no research about the milk-derived exosomes phosphorylation to reveal the difference of protein phosphorylation in different species of milk. To investigate the pathways and proteins with specific functions, phosphorylated proteomic analysis of milk-derived exosomes from different species is performed, and provide new ideas for exploring diversified treatments of disease. MethodsWhey and exosomes derived from bovine, porcine and caprine milk were performed for proteomics and phosphoproteomics analysis. The relationship between milk exosome proteins from different species and signaling pathways were analyzed using bioinformatics tools. ResultsA total of 4 191 global proteins, 1 640 phosphoproteins and 4 064 phosphosites were identified from 3 species of milk-derived exosomes, and the exosome proteins and phosphoproteins from different species were significantly higher than those of whey. Meanwhile, some special pathways were enriched like Fcγ-mediated phagocytosis from bovine exosomes, pathways related with neural and immune system from caprine exosomes, positive and negative regulation of multiple activities from porcine exosomes. ConclusionIn this study, the proteomic and phosphoproteomic analyses of exosomes and whey from bovine, porcine and caprine milk were carried out to reveal the difference of composition and related signaling pathways of milk exosome from different species. These results provided powerful support for the application of exosomes from different milk sources in the field of disease treatment.

Analyze the changes in phenolic components and gene expression profiles of Tetrastigma hemsleyanum leaves with supplemental blue light, and screen differentially expressed genes related to phenolic metabolism, providing a basis for improving the quality of T. hemsleyanum leaves. Using the leaves of T. hemsleyanum under supplemental blue light and visible light as research materials, the content of phenolic components such as neochlorogenic acid, chlorogenic acid, and 3-O-coumaroyl quinic acid was significantly increased by supplemental blue light. A total of 102 949 unigenes were obtained from the transcriptome, including 14 564 differentially expressed unigenes. They can be divided into 52 subclasses by gene ontology (GO) function, kyoto encyclopedia of genes and genomes (KEGG) pathway enrichment analysis showed that these differentially expressed unigenes were significantly enriched in phenolic metabolic pathways such as phenylpropane biosynthesis, flavonoid biosynthesis and isoflavone biosynthesis. A total of 63 differentially expressed unigenes were identified in the pathways related to the biosynthesis of phenolic cpmponents, including 19 key enzymes such as phenylalanine ammonia-lyase (PAL), cinnamyl-alcohol dehydrogenase (CAD), flavonol synthase (FLS), and so on. This study greatly enriched the genetic data information of T. hemsleyanum and laid a foundation for analyzing the mechanism of blue light promoting the biosynthesis of phenols and molecular breeding of T. hemsleyanum.

Objective To investigate the value of multimodal MRI radiomics in predicting muscle-invasive bladder cancer.Methods A total of 178 patients with pathology diagnosis of bladder cancer were retrospectively collected,including 31 cases of muscle invasive bladder cancer(MIBC)and 147 cases of non-muscle invasive bladder cancer(NMIBC).Patients were randomly divided into training group and testing group at a ratio of 7︰3.The range of bladder tumors in T2WI,diffusion weighted imaging(DWI)and apparent diffusion coefficient(ADC)images were segmented as volume of interest(VOI)by using ITK-SNAP software.Radiomics features were extracted through A.K software.The optimal radiomics features were obtained through radiomics algorithm and least absolute shrinkage and selection operator(LASSO)method.Finally,the logistic regression analysis method and random forest model method were used to construct prediction models.The performance of prediction models was evaluated by the receiver operating characteristic(ROC)curve.Results This study constructed four groups of models containing T2WI prediction model,DWI prediction model,ADC prediction model,and T2WI+DWI+ADC prediction model.The area under the curve(AUC)of T2WI,DWI,and ADC prediction models for identifying MIBC and NMIBC were separately 0.920,0.914,and 0.954 in the training group while those were respectively 0.881,0.773,and 0.871 in the testing group.There was no statistical significance between T2WI,DWI,and ADC prediction models.In training and testing groups,the AUC of T2WI+DWI+ADC prediction model were respectively 0.959 and 0.909,which were higher than the single sequence prediction model.The sensitivity and specificity of the training group were 0.905 and 0.853 and the sensitivity and specificity of the testing group were 0.778 and 0.795.Conclusion MRI radiomics prediction model can effectively differentiate MIBC and NMIBC.The T2WI+DWI+ADC prediction model shows better prediction efficiency.

The traditional commodity specifications of Chinese medicinal materials are mainly divided into different grades based on macroscopic characteristics. As the basis for high quality and good price, there is still a lack of systematic evaluation on whether they are consistent with the current standards and whether they can reflect the internal quality of medicinal material. Panax notoginseng is a commonly used, large consumption of Chinese medicinal material. At present, it is divided into 8 grades in the market based on "Tou" (the number of crude drug /500 g), but it is not related to the standard of total saponins of Panax notoginseng (the sum of three saponins) in Chinese Pharmacopoeia. In this study, ultra-performance liquid chromatography-quadrupole-time of flight mass spectrometry (UPLC-Q-TOF-MS/MS) coupled with mass spectrometry molecular network were used for the rapid identification of saponins of Panax notoginseng with different "Tou" and a total of 64 saponins were identified. Seventeen saponins related to "Tou" were screened by orthogonal partial least squares discriminant analysis (OPLS-DA). The content of five saponins R₁, Rb₁, Rg₁, Rd, and Re in Panax notoginseng with different "Tou" was determined by high performance liquid chromatography (HPLC). The results of correlation analysis showed that Rd and R₁ with the largest VIP values among the differential saponins, which significantly negatively correlated with "Tou" (P < 0.05). Based on the determination results of 36 batches of samples, using Rd/Total Panax notoginseng saponins (TPNS) ratio ( > 0.08) as the index, Panax notoginseng can be divided into two grades: 20-60 "Tou" (superior) and 80-200 "Tou" (qualified). Based on the concept of "macroscopic characteristics and chemical profiling", this study integrates the non-targeted analysis and quantitative determination methods to provide a new strategy for quality evaluation of Panax notoginseng.

Objective To investigate the efficacy and safety of camrelizumab monoclonal antibody combined with molecular-targeted therapy in elderly patients with unresectable or advanced hepatocellular carcinoma(HCC).Methods A retrospective analysis was performed for the patients with unresectable/advanced HCC who attended six hospitals from January 1,2019 to March 31,2021,and all patients received camrelizumab monoclonal antibody treatment,among whom 84.8%also received targeted therapy.According to the age of the patients,they were divided into elderly group(≥65 years)and non-elderly group(＜65 years).The two groups were assessed in terms of overall survival(OS),progression-free survival(PFS),objective response rate(ORR),disease control rate(DCR),and immune-related adverse events(irAE).The chi-square test or the Fisher's exact test was used for comparison of categorical data between groups;the independent samples t-test was used for comparison of normally distributed continuous data,and the Mann-Whitney U test was used for comparison of non-normally distributed continuous data between two groups.The Kaplan-Meier method was used for survival analysis,and the log-rank test was used for comparison of survival curves.Univariate and multivariate Cox proportional hazards regression analyses were used to determine the independent influencing factors for PFS and DCR at 6 months.Results A total of 99 HCC patients were enrolled,with 27 in the elderly group and 72 in the non-elderly group.The elderly group had an OS rate of 67.8%,an ORR of 44.4%,and a DCR of 74.1%at 12 months and a median PFS of 6.4(95%confidence interval[CI]:3.0-12.4)months,with no significant differences compared with the non-elderly group(all P＞0.05).The median OS was unavailable for the elderly group,while the non-elderly group had an OS of 18.9(95%CI:13.0-24.8)months;there was no significant difference between the two groups(P=0.485).The univariate and multivariate Cox regression analyses showed that major vascular invasion(MVI)was an independent risk factor for PFS(hazard ratio[HR]=2.603,95%CI:1.136-5.964,P=0.024)and DCR(HR=3.963,95%CI:1.671-9.397,P=0.002)at 6 months,while age,sex,etiology of HBV infection,presence of extrahepatic metastasis,Child-Pugh class B,and alpha-fetoprotein＞400 ng/mL were not associated with PFS or DCR at 6 months.For the elderly group,the incidence rates of any irAE and grade 3/4 irAE were 51.9%and 25.9%,respectively,with no significant differences compared with the non-elderly group(P＞0.05),and skin disease was the most common irAE in both groups(39.4%).Conclusion Camrelizumab monoclonal antibody combined with molecular-targeted therapy has similar efficacy and safety in patients with unresectable/advanced HCC aged≥65 years and those aged＜65 years.MVI is associated with suboptimal response to immunotherapy and poor prognosis.

Objective To explore the comparative study of constructing prediction models for muscle invasive of bladder cancer based on different machine learning algorithms combined with MRI radiomic.Methods A total of 187 bladder cancer patients who underwent MRI examination and were confirmed by pathology were retrospectively selected.Patients were randomly divided into a training set and a test set in a 7∶3 ratio.The patients were divided into muscle invasive bladder cancer(MIBC)group and non-muscle invasive bladder cancer(NMIBC)group according to the surgical pathology results.Tumor volume of interest(VOI)was outlined on the images of T2 WI,diffusion weighted imaging(DWI),and apparent diffusion coefficient(ADC),and the radiomic features were extracted by A.K software,and dimensionality reduction was performed using the maximum relevance minimum redundancy(mRMR)algorithm combined with least absolute shrinkage and selection operator(LASSO).Six machine learning algorithms,including K-nearest neighbor(KNN),decision tree(DT),support vector machine(SVM),logistic regression(LR),random forest(RF),and explainable boosting machine(EBM)were used to construct the radiomic model and calculate the corresponding area under the curve(AUC),accuracy,sensitivity,and specificity,respectively.Results Six machine learning algorithms,including KNN,DT,SVM,LR,RF,and EBM were used to construct the radiomic model,and the AUC values for predicting MIBC in the training set were 0.863,0.838,0.853,0.866,0.977,0.997,and in the test set were 0.748,0.833,0.860,0.868,0.870,0.900.Among them,the MRI radiomic model constructed based on EBM had the highest predictive efficacy for MIBC,with AUC values,accuracy,sensitivity and specificity of 0.997,0.977,0.957 and 0.981 in the training set,and 0.900,0.877,0.800,and 0.894 in the test set,respectively.Conclusion Multiple machine learning algorithms combined with MRI radiomic to construct models have good predictive efficacy for MIBC,and the model constructed based on EBM shows the highest predictive value.

Objective To investigate the risk factors for bronchopulmonary dysplasia(BPD)in twin preterm infants with a gestational age of<34 weeks,and to provide a basis for early identification of BPD in twin preterm infants in clinical practice.Methods A retrospective analysis was performed for the twin preterm infants with a gestational age of<34 weeks who were admitted to 22 hospitals nationwide from January 2018 to December 2020.According to their conditions,they were divided into group A(both twins had BPD),group B(only one twin had BPD),and group C(neither twin had BPD).The risk factors for BPD in twin preterm infants were analyzed.Further analysis was conducted on group B to investigate the postnatal risk factors for BPD within twins.Results A total of 904 pairs of twins with a gestational age of<34 weeks were included in this study.The multivariate logistic regression analysis showed that compared with group C,birth weight discordance of>25%between the twins was an independent risk factor for BPD in one of the twins(OR=3.370,95%CI:1.500-7.568,P<0.05),and high gestational age at birth was a protective factor against BPD(P<0.05).The conditional logistic regression analysis of group B showed that small-for-gestational-age(SGA)birth was an independent risk factor for BPD in individual twins(OR=5.017,95%CI:1.040-24.190,P<0.05).Conclusions The development of BPD in twin preterm infants is associated with gestational age,birth weight discordance between the twins,and SGA birth.

Objective:To analyze the clinical characteristics and prognosis of patients with CD5+diffuse large B-cell lymphoma(DLBCL).Methods:The clinical data of 161 newly treated DLBCL patients in Gansu Provincial Hospital from January 2013 to January 2020 were retrospectively analyzed.According to CD5 expression,the patients were divided into CD5+group and CD5-group.The clinical characteristics and prognosis of the two groups were statistically analyzed.Results:The median age of patients in CD5+group was 62 years,which was higher than 56 years in CD5-group(P=0.048).The proportion of women in CD5+group was 62.96％,which was significantly higher than 41.79％in CD5-group(P=0.043).The proportion of patients with IPI score＞2 in CD5+group was 62.96％,which was higher than 40.30％in CD5-group(P=0.031).Survival analysis showed that the median overall survival and progression-free survival time of patients in CD5+group were 27(3-77)and 31(3-76)months,respectively,which were both shorter than 30(5-84)and 32.5(4-83)months in CD5-group(P=0.047,P=0.026).Univariate analysis showed that advanced age,positive CD5 expression,triple or double hit at initial diagnosis,high IPI score and no use of rituximab during chemotherapy were risk factors for the prognosis of DLBCL patients.Further Cox multivariate regression analysis showed that these factors were also independent risk factors except for advanced age.Conclusion:CD5+DLBCL patients have a worse prognosis than CD5-DLBCL patients.Such patients are more common in females,with advanced age and high IPI score,which is a special subtype of DLBCL.

Background: Intrahepatic cholangiocarcinoma (ICC) is a highly desmoplastic tumor with poor prognosis even after curative resection. We investigated the associations between the composition of the ICC stroma and immune cell infiltration and aimed to develop a stromal-immune signature to predict prognosis in surgically treated ICC. Patients and methods: We recruited 359 ICC patients and performed immunohistochemistry to detect α-smooth muscle actin (α-SMA), CD3, CD4, CD8, Foxp3, CD68, and CD66b. Aniline was used to stain collagen deposition. Survival analyses were performed to detect prognostic values of these markers. Recursive partitioning for a discrete-time survival tree was applied to define a stromal-immune signature with distinct prognostic value. We delineated an integrated stromal-immune signature based on immune cell subpopulations and stromal composition to distinguish subgroups with different recurrence-free survival (RFS) and overall survival (OS) time. Results: We defined four major patterns of ICC stroma composition according to the distributions of α-SMA and collagen: dormant (α-SMAlow/collagenhigh), fibrogenic (α-SMAhigh/collagenhigh), inert (α-SMAlow/collagenlow), and fibrolytic (α-SMAhigh/collagenlow). The stroma types were characterized by distinct patterns of infiltration by immune cells. We divided patients into six classes. Class I, characterized by high CD8 expression and dormant stroma, displayed the longest RFS and OS, whereas Class VI, characterized by low CD8 expression and high CD66b expression, displayed the shortest RFS and OS. The integrated stromal-immune signature was consolidated in a validation cohort. Conclusion We developed and validated a stromal-immune signature to predict prognosis in surgically treated ICC. These findings provide new insights into the stromal-immune response to ICC.