Objective To systematically analyze the literature characteristics of Journal of Organ Transplantation since its inception. Methods Using the China National Knowledge Infrastructure (CNKI) academic journal full-text database as the data source, all articles published in the Journal of Organ Transplantation from January 2010 to August 2025 were retrieved. After excluding non-academic papers, a total of 1 568 research papers were included. R language 4.3.0, Bibliometrix package 3.2.1, and Citespace software were used to analyze the number of publications, publishing institutions, authors, keywords and other aspects. Results The number of publications in Journal of Organ Transplantation increased from an average of 82 articles per year in the early years after its inception to 113 articles per year in recent years, a growth of 37.8%. The geographical distribution of publishing institutions covers 32 provinces, cities and autonomous regions nationwide, mainly concentrated in the South China, East China and North China regions, and has now basically covered the central and western regions in recent years. The author collaboration network includes 45 authors distributed across 7 major collaboration clusters, forming a stable multi-level national research system centered on key university-affiliated hospitals. The high-frequency keywords are dominated by "liver transplantation" (425 times) and "kidney transplantation" (396 times). The theme evolution shows a clear three-stage characteristic: initially focusing on clinical technology application, deepening to immune mechanism exploration in the middle stage, and recently (since 2022) focusing on cutting-edge research areas such as xenotransplantation. Conclusions Journal of Organ Transplantation has witnessed the rapid development of China's organ transplantation cause, fully reflecting the research status and trends in China's organ transplantation field, and has provided an important platform for the future development and international cooperation in China's organ transplantation field.

ObjectiveTo analyze the research status and development trends of non-pharmacological therapies for post-stroke spasticity (PSS) over the past two decades. MethodsRelevant literatures on non-pharmacological rehabilitation of PSS published from January, 2004 to June, 2024 were retrieved from Web of Science Core Collection. CiteSpace 6.3.R6 and VOSviewer 1.6.18 were used for visualization analysis. ResultsA total of 780 publications were included. The annual number of publications showed an overall upward trend. China, the USA, and Italy contributed the highest number of publications. The Hong Kong Polytechnic University and researcher Noureddin Nakhostin Ansari were identified as the most influential institution and author, respectively. High-frequency keywords and cluster labels included electric stimulation, transcranial magnetic stimulation, robot and acupuncture. ConclusionOver the past 20 years, researches on non-pharmacological therapies for PSS have remained active, with hotspots focusing on diverse interventions such as electrical stimulation, magnetic stimulation and robot-assisted therapy.

ObjectiveTo construct and validate a clinical prediction model for inflammatory remission outcomes in rheumatoid arthritis （RA） patients with liver and kidney deficiency syndrome treated with Bushen Zhiwang Decoction （补肾治尪汤， BZD） based on metabolomics. MethodsA prospective cohort study was conducted， enrol-ling 60 RA patients with liver and kidney deficiency syndrome. All patients were treated with BZD and conventional-dose oral conventional synthetic disease-modifying antirheumatic drugs （csDMARDs） for 12 months. Clinical data were collected， and the change in disease activity score in 28 joints （DAS28） after treatment compared with baseline （△DAS28） was used as the primary outcome and grouping criterion. Peripheral blood samples were collected before treatment to analyze plasma metabolites. Differential analysis and least absolute shrinkage and selection operator （LASSO） regression were used to preliminarily screen differential metabolites， followed by machine learning algorithms to further identify a core metabolite combination. Based on the expression levels of the core metabolite combination， a novel metabolite index， namely the metabolomics-based inflammatory remission score （Met-IRS）， was calculated using standar-dized metabolite values， and its clinical applicability was evaluated. A clinical prediction model was constructed by integrating clinical characteristics and Met-IRS， and the model performance was assessed. ResultsAmong the 60 patients， those with △DAS28 ≥ 0.27 were assigned to the high inflammatory remission group， while those with △DAS28 ＜ 0.27 were assigned to the low inflammatory remission group， with 30 cases in each group. Compared to the low inflammatory remission group， the high inflammatory remission group showed a higher frequency of methotrexate use and a lower positive rate of rheumatoid factor （RF） （P＜0.05）. Seven core metabolites were identified as the optimal combination， including mangiferic acid， fatty acid-hydroxy fatty acid ester 40∶6， fatty acid-hydroxy fatty acid ester 18∶0， fatty acid-hydroxy fatty acid ester 36∶1， glucosylceramide， lysophosphatidylcholine 22∶5， and pregnanetriol ketone. The calculated Met-IRS comprehensively reflected the characteristics of differential metabolites and demonstrated clinical applicability. Met-IRS was significantly higher in the high inflammatory remission group than in the low inflammatory remission group， and was positively correlated with high inflammatory remission outcomes （P＜0.05）. Based on the variables Met-IRS， methotrexate use， leflunomide use， and RF positivity， a clinical prediction model for inflammatory remission in RA treatment （Cj-RTRM） was constructed. Model performance evaluation demonstrated that the model had good clinical predictive ability， with an area under the receiver operating characteristic curve （AUC） of 0.880， sensitivity 0.967， specificity 0.700 and Youden's index 0.667. ConclusionThe clinical prediction model Cj-RTRM constructed based on the metabolomics-based inflammatory remission score Met-IRS can effectively predict clinical inflammatory remission outcomes in RA patients treated with BZD and accurately identify the advantageous population for this treatment. This model provides guiding evidence for dynamic inflammation monitoring， targeted management， and identification of populations with advantages in traditional Chinese medicine.

The initiation of adaptive immune responses relies on the precise recognition and interpretation of antigenic information. In this process, the specific binding of T cell receptors (TCRs) to peptide-major histocompatibility complex (pMHC) molecules represents one of the key molecular events in the initiation of adaptive immune responses. Accordingly, the structural features of TCR-pMHC complexes provide a fundamental basis for dissecting antigen recognition mechanisms and support rational vaccine design, therapeutic target discovery in TCR-based immunotherapy, and TCR identification and optimization. However, experimental determination of TCR-pMHC structures remains costly, time-consuming, and limited in coverage, making computational approaches essential for rapidly obtaining reliable structural information. Computational methods for predicting the structures of TCR-pMHC complexes have advanced rapidly in recent years, driven by progress in deep learning-based modeling frameworks and the increasing availability of structural and sequence resources. Despite these developments, most existing tools do not adequately distinguish the key structural and biophysical differences between MHC class I (MHC-I) and MHC class II (MHC-II) complexes during model construction. As a consequence, their predictive performance differs substantially between class I and class II complexes. In general, structural predictions for class I complexes outperform those for class II complexes. This discrepancy may be related to several fundamental differences between the two systems, including the architecture of the peptide-binding groove, the distribution of peptide lengths, and the properties of peptide flanking residues (PFRs). Compared with MHC-I molecules, MHC-II molecules usually bind longer antigenic peptides, which typically range from 13 to 25 amino acids in length. PFRs at both termini of these peptides participate in regulating the overall conformation of TCR-pMHC class II complexes and exert a pronounced effect on the geometric and physicochemical characteristics of the TCR-pMHC binding interface. Furthermore, within the TCR recognition interface, the complementarity-determining regions (CDRs) consist of segments that differ markedly in conformational behavior. They commonly include regions that are relatively rigid and structurally stable, together with highly flexible segments exhibiting substantial conformational plasticity. These rigidity-flexibility features constitute an essential structural basis enabling TCRs to recognize diverse peptide-MHC ligands and to accommodate conformational heterogeneity at the interface. However, many current modeling tools, in an effort to enforce global conformational stability or reduce structural noise, tend to over-constrain intrinsically flexible regions. Such oversimplification may lead to inappropriate rigidification of flexible CDR loops, resulting in local structural distortions, compromised interface geometry, or even complete modeling failure for specific complexes. Against this background, the review approaches the field from the perspective of computational differences between MHC-I and MHC-II complexes. We first systematically organize and summarize available resources related to TCRs and pMHCs, including structural datasets, sequence databases, prediction tools, and benchmarking studies. We then focus on five representative tools capable of predicting both class I and class II complexes—AlphaFold2, AlphaFold3, TCRmodel2, tFold-TCR, and TCR-pHLA_ModellerS. After excluding structures present in the training sets of these tools, we constructed a benchmark dataset comprising 25 class I and 10 class II TCR-pMHC complexes in the bound state and conducted a systematic evaluation using this dataset. We first employ widely used general evaluation metrics, including All-Atom Root Mean Square Deviation (All-Atom RMSD), Backbone RMSD, Template Modeling score (TM-score), and DockQ, to assess the global conformational accuracy and interface modeling quality of class I and class II complexes. For class II complexes, we propose for the first time a peptide flanking residue deviation index, including the PFRs-Deviation Index (PFRs-DI), N-PFR-Deviation Index (N-PFR-DI), and C-PFR-Deviation Index (C-PFR-DI), to quantitatively characterize conformational deviations in PFRs. In addition, we propose the CDR conformational consistency index (CCC) designed to qualitatively evaluate the ability of prediction tools to capture TCR CDR conformational flexibility. These metrics collectively assess a tool’s ability to model both overall conformation and critical functional regions, thereby addressing the limitations of existing evaluation criteria that overemphasize global structure while inadequately capturing modeling quality in key functional areas. This establishes a unified analytical framework for MHC-I and MHC-II complexes to guide data resource selection, modeling strategy formulation, and evaluation system development. The framework further advances computational modeling and provides crucial support for multi-scale analysis of TCR-pMHC recognition mechanisms and their biological functions.

The initiation of adaptive immune responses relies on the precise recognition and interpretation of antigenic information. In this process, the specific binding of T cell receptors (TCRs) to peptide-major histocompatibility complex (pMHC) molecules represents one of the key molecular events in the initiation of adaptive immune responses. Accordingly, the structural features of TCR-pMHC complexes provide a fundamental basis for dissecting antigen recognition mechanisms and support rational vaccine design, therapeutic target discovery in TCR-based immunotherapy, and TCR identification and optimization. However, experimental determination of TCR-pMHC structures remains costly, time-consuming, and limited in coverage, making computational approaches essential for rapidly obtaining reliable structural information. Computational methods for predicting the structures of TCR-pMHC complexes have advanced rapidly in recent years, driven by progress in deep learning-based modeling frameworks and the increasing availability of structural and sequence resources. Despite these developments, most existing tools do not adequately distinguish the key structural and biophysical differences between MHC class I (MHC-I) and MHC class II (MHC-II) complexes during model construction. As a consequence, their predictive performance differs substantially between class I and class II complexes. In general, structural predictions for class I complexes outperform those for class II complexes. This discrepancy may be related to several fundamental differences between the two systems, including the architecture of the peptide-binding groove, the distribution of peptide lengths, and the properties of peptide flanking residues (PFRs). Compared with MHC-I molecules, MHC-II molecules usually bind longer antigenic peptides, which typically range from 13 to 25 amino acids in length. PFRs at both termini of these peptides participate in regulating the overall conformation of TCR-pMHC class II complexes and exert a pronounced effect on the geometric and physicochemical characteristics of the TCR-pMHC binding interface. Furthermore, within the TCR recognition interface, the complementarity-determining regions (CDRs) consist of segments that differ markedly in conformational behavior. They commonly include regions that are relatively rigid and structurally stable, together with highly flexible segments exhibiting substantial conformational plasticity. These rigidity-flexibility features constitute an essential structural basis enabling TCRs to recognize diverse peptide-MHC ligands and to accommodate conformational heterogeneity at the interface. However, many current modeling tools, in an effort to enforce global conformational stability or reduce structural noise, tend to over-constrain intrinsically flexible regions. Such oversimplification may lead to inappropriate rigidification of flexible CDR loops, resulting in local structural distortions, compromised interface geometry, or even complete modeling failure for specific complexes. Against this background, the review approaches the field from the perspective of computational differences between MHC-I and MHC-II complexes. We first systematically organize and summarize available resources related to TCRs and pMHCs, including structural datasets, sequence databases, prediction tools, and benchmarking studies. We then focus on five representative tools capable of predicting both class I and class II complexes—AlphaFold2, AlphaFold3, TCRmodel2, tFold-TCR, and TCR-pHLA_ModellerS. After excluding structures present in the training sets of these tools, we constructed a benchmark dataset comprising 25 class I and 10 class II TCR-pMHC complexes in the bound state and conducted a systematic evaluation using this dataset. We first employ widely used general evaluation metrics, including All-Atom Root Mean Square Deviation (All-Atom RMSD), Backbone RMSD, Template Modeling score (TM-score), and DockQ, to assess the global conformational accuracy and interface modeling quality of class I and class II complexes. For class II complexes, we propose for the first time a peptide flanking residue deviation index, including the PFRs-Deviation Index (PFRs-DI), N-PFR-Deviation Index (N-PFR-DI), and C-PFR-Deviation Index (C-PFR-DI), to quantitatively characterize conformational deviations in PFRs. In addition, we propose the CDR conformational consistency index (CCC) designed to qualitatively evaluate the ability of prediction tools to capture TCR CDR conformational flexibility. These metrics collectively assess a tool’s ability to model both overall conformation and critical functional regions, thereby addressing the limitations of existing evaluation criteria that overemphasize global structure while inadequately capturing modeling quality in key functional areas. This establishes a unified analytical framework for MHC-I and MHC-II complexes to guide data resource selection, modeling strategy formulation, and evaluation system development. The framework further advances computational modeling and provides crucial support for multi-scale analysis of TCR-pMHC recognition mechanisms and their biological functions.

Laboratory animal science is an emerging interdisciplinary field supporting life science and medical research, and a key component of talent cultivation and technological innovation in medical colleges. Currently, this field faces challenges such as a significant imbalance between talent supply and demand and a weak systematic training system. To respond to national strategies and societal needs, Shandong First Medical University established the School of Laboratory Animal through industry-education integration during the 14th Five-Year Plan period. The school is based on The Medical Laboratory Technology major, cultivates application-oriented, interdisciplinary professionals in laboratory animal science, and has constructed a three-in-one training system integrating "course learning–scientific research–industry practice". Specific measures for talent cultivation include: optimizing the general education, professional courses, and intensive practical modules in the talent training program, and establishing the "Yellow River Class" integrating industry and education; implementing an undergraduate mentor system, and conducting scientific research training based on The Model Animal Research and Development Engineering Laboratory; collaborating with bases across the entire industrial chain of laboratory animal production, research, application, and quality control, appointing industry mentors, and strengthening practical teaching. Practice has shown that this system has been remarkably effective: a total of 320 undergraduate students have been enrolled since 2020; the employment rate of undergraduate graduates for two consecutive cohorts has reached 100%, and the postgraduate enrollment rate has exceeded 50%; undergraduate students have won numerous national and provincial awards in academic competitions, and have obtained multiple patents and published papers. In the future, the school will further integrate the advantages of medicine, agriculture, and science, optimize the depth and breadth of courses, strengthen the construction of faculty and teaching materials, and improve the undergraduate-master's integrated training mechanism. This article can provide a reference for the training of professionals in laboratory animal science and related biomedical fields in medical colleges.

Objective To evaluate the diagnostic efficacy of different pathogen detection methods in active tuberculosis (ATB) of kidney transplant recipients (KTRs) and analyze their diagnostic value in different types of clinical specimens. Methods A retrospective observational study was conducted, including adult KTRs with suspected ATB from the Eighth Medical Center of Chinese People's Liberation Army General Hospital from January 2020 to January 2025. Baseline data of the recipients were collected and analyzed, and the diagnostic criteria were based on a composite reference standard (including clinical diagnosis and bacteriological confirmation). The single-detection efficacy of acid-fast staining smear, mycobacterium tuberculosis culture, Xpert test for the detection of mycobacterium tuberculosis nucleic acid and resistance to rifampicin (Xpert MTB/RIF), as well as tuberculosis or non-tuberculosis mycobacterial DNA detection was compared. The differences in sensitivity among different methods were analyzed. Results A total of 95 KTRs with suspected ATB and 136 specimens were included. And 52 cases were diagnosed as ATB, among which 48 cases were confirmed by bacteriological means (92%). The sensitivity of Xpert MTB/RIF and DNA detection (0.55 and 0.52, respectively) was significantly higher than that of acid-fast staining smear (0.30) and mycobacterium tuberculosis culture (0.35) (P < 0.01). There was a statistically significant difference in the bacteriological diagnosis rate among different specimen types (P = 0.035). The detection sensitivity of each method was the highest in pus, puncture tissue and bronchoalveolar lavage fluid, followed by sputum and pleural and abdominal effusions, and the lowest in cerebrospinal fluid. Conclusions The sensitivity of nucleic acid amplification technology for immunosuppressed KTRs is slightly reduced, but it should still be recommended as the preferred initial screening tool. The comprehensive strategy of "multi-site sampling (prior to obtaining high bacterial load specimens) combined with multiple method detection" may improve the diagnostic efficacy.

ObjectiveTo study the effect and mechanism of Shentong Zhuyutang in treating intervertebral disc degeneration （IDD） in rats. MethodsIn the cell experiment， male rats were administrated with normal saline or low-， medium-， and high-dose （3.38， 6.75，13.5 g·kg^-1， respectively） Shentong Zhuyutang by gavage， respectively， and serum samples were collected after 7 days of continuous administration. Another 10 male rats were selected for the isolation of nucleus pulposus cells. The cell model of IDD was established by treatment with interleukin （IL）-1β. The modeled cells were then treated with Shentong Zhuyutang-containing serum and the ferroptosis inhibitor ferrostatin-1 （Fer-1）， respectively， to investigate the effects of Shentong Zhuyutang-containing serum on the proliferation and ferroptosis of nucleus pulposus cells. To study the role of silent information regulator 1 （SIRT1）/nuclear factor erythroid 2-related factor 2 （Nrf2） in the regulation of ferroptosis in nucleus pulposus cells by Shentong Zhuyutang-containing serum， this study treated the cells with the SIRT1 inhibitor Ex 527 and the Nrf2 inhibitor ML385， respectively， in addition to the treatment with IL-1β and high-dose Shentong Zhuyutang-containing serum. The cell-counting kit-8 （CCK-8） assay and EdU staining were employed to measure the cell viability and proliferation， respectively. The Fe²⁺， glutathione （GSH）， and malondiadehyde （MDA） levels were measured by colorimetric assay. Western blot was employed to determine the protein levels of glutathione peroxidase 4 （GPX4）， acyl-CoA synthetase long-chain family 4 （ACSL4）， Collagen Ⅱ， Aggrecan， SIRT1， and Nrf2. Immunofluorescence was used detect SIRT1 expression. In the animal experiment， male rats were treated with anulus puncture for the modeling of IDD. Rats were randomly assigned into sham operation， model， Shentong Zhuyutang-containing serum （13.5 g·kg^-1）， and positive control （nimesulide dispersible tablets， 0.18 mg·kg^-1） groups. Rats in the drug intervention groups were administrated with corresponding agents at 1 mL·kg^-1， and those in the sham operation and model groups were administrated with equal volumes of normal saline， once daily for 28 consecutive days. At the end of the last administration， the histopathological changes in the intervertebral discs of rats were observed by hematoxylin-eosin staining and scored by the Masuda method. Western blot was employed to determine the protein levels of SIRT1， Nrf2， GPX4， and Collagen Ⅱ in the nucleus pulposus tissue. ResultsCompared with the control group， the IL-1β group of nucleus pulposus cells showed elevated levels of Fe²⁺， MDA， and ACSL4 （P<0.05）， decreased cell viability， lowered GSH level， and down-regulated protein levels of GPX4， Collagen Ⅱ， and Aggrecan （P<0.05）. Shentong Zhuyutang-containing serum and Fer-1 reversed the effects of IL-1β on the viability and ferroptosis of nucleus pulposus cells and up-regulated the protein levels of Collagen Ⅱ and Aggrecan in nucleus pulposus cells （P<0.05）. Compared with the control group， the IL-1β group showcased down-regulated expression of Sirt1 and Nrf2 in nucleus pulposus cells （P<0.05）. Compared with the IL-1β group， the high-dose Shentong Zhuyutang-containing serum+IL-1β group showed up-regulated expression of SIRT1 and Nrf2 in nucleus pulposus cells （P<0.05）. Compared with the high-dose Shentong Zhuyutang-containing serum+IL-1β group， the ML385 group showed down-regulated protein levels of Nrf2 and GPX4， lowered GSH level， and elevated Fe²⁺ and MDA levels （P<0.05）. In addition， the Ex 527 group showed down-regulated protein levels of SIRT1， Nrf2， and GPX4 （P<0.05）. The results of the animal experiment showed that compared with the sham operation group， the model group had severe degeneration of the intervertebral disc tissue with increased pathological score， up-regulated protein level of ACSL4 （P<0.05）， and down-regulated protein levels of SIRT1， Nrf2， GPX4， and Collagen Ⅱ （P<0.05）. Compared with the model group， the Shentong Zhuyutang group showed alleviated IDD with declined pathological score， down-regulated protein level of ACSL4 （P<0.05）， and up-regulated protein levels of SIRT1， Nrf2， GPX4， and Collagen Ⅱ （P<0.05）. ConclusionShentong Zhuyutang may activate the SIRT1/Nrf2 signaling pathway to inhibit the ferroptosis of nucleus pulposus cells， thereby delaying the process of IDD in rats.

Alzheimer’s disease (AD) is a central neurodegenerative disease characterized by progressive cognitive decline and memory impairment in clinical. Currently, there are no effective treatments for AD. In recent years, a variety of therapeutic approaches from different perspectives have been explored to treat AD. Although the drug therapies targeted at the clearance of amyloid β-protein (Aβ) had made a breakthrough in clinical trials, there were associated with adverse events. Neuroinflammation plays a crucial role in the onset and progression of AD. Continuous neuroinflammatory was considered to be the third major pathological feature of AD, which could promote the formation of extracellular amyloid plaques and intracellular neurofibrillary tangles. At the same time, these toxic substances could accelerate the development of neuroinflammation, form a vicious cycle, and exacerbate disease progression. Reducing neuroinflammation could break the feedback loop pattern between neuroinflammation, Aβ plaque deposition and Tau tangles, which might be an effective therapeutic strategy for treating AD. Traditional Chinese herbs such as Polygonum multiflorum and Curcuma were utilized in the treatment of AD due to their ability to mitigate neuroinflammation. Non-steroidal anti-inflammatory drugs such as ibuprofen and indomethacin had been shown to reduce the level of inflammasomes in the body, and taking these drugs was associated with a low incidence of AD. Biosynthetic nanomaterials loaded with oxytocin were demonstrated to have the capability to anti-inflammatory and penetrate the blood-brain barrier effectively, and they played an anti-inflammatory role via sustained-releasing oxytocin in the brain. Transplantation of mesenchymal stem cells could reduce neuroinflammation and inhibit the activation of microglia. The secretion of mesenchymal stem cells could not only improve neuroinflammation, but also exert a multi-target comprehensive therapeutic effect, making it potentially more suitable for the treatment of AD. Enhancing the level of TREM2 in microglial cells using gene editing technologies, or application of TREM2 antibodies such as Ab-T1, hT2AB could improve microglial cell function and reduce the level of neuroinflammation, which might be a potential treatment for AD. Probiotic therapy, fecal flora transplantation, antibiotic therapy, and dietary intervention could reshape the composition of the gut microbiota and alleviate neuroinflammation through the gut-brain axis. However, the drugs of sodium oligomannose remain controversial. Both exercise intervention and electromagnetic intervention had the potential to attenuate neuroinflammation, thereby delaying AD process. This article focuses on the role of drug therapy, gene therapy, stem cell therapy, gut microbiota therapy, exercise intervention, and brain stimulation in improving neuroinflammation in recent years, aiming to provide a novel insight for the treatment of AD by intervening neuroinflammation in the future.

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.