A proteomic panel, kit and system for predicting secondary infection in patients with hepatitis b-related liver failure

Abstract

The present application relates to a kind of protein combination, kit and system for predicting secondary infection of hepatitis B related liver failure patient, including the use of protein combination in the preparation for predicting secondary infection of hepatitis B related liver failure patient, the combination includes lysozyme LYZ, calmodulin 1 CALM1, heparin cofactor II SERPIND1 and cutin DPT.It is directly reflected that the combination is biomarker "inflammation-coagulation imbalance" this core mechanism of liver disease secondary infection, with high specificity, and the combination is not interfered by liver disease background, more can truly reflect the potential reaction state of body to infection under liver disease background, to overcome the problem of "failure" and "false negative" of traditional index ";The combination can significantly change concentration when patient is admitted to hospital early, without clinical infection signs, to provide key time window for early preventive treatment.

Description

Technical Field

[0001] This invention belongs to the field of biomedical detection technology, specifically relating to a protein combination, reagent kit, and system for predicting secondary infections in patients with hepatitis B-related liver failure. Background Technology

[0002] Patients with hepatitis B virus-associated acute-on-chronic liver failure (HBV-ACLF) are at extremely high risk of hospital-acquired infections due to severely impaired liver function and systemic immune dysfunction. Secondary infections in liver disease patients have distinct characteristics, fundamentally different from infections in the general population or other hospitalized patients. Specifically: First, due to decreased liver capacity to synthesize immune-related proteins and disordered cellular immunity, patients often exhibit an "immune paralysis" state, leading to a lack of typical clinical manifestations and elevated inflammatory markers in the early stages of infection, making it highly susceptible to missed diagnosis. Second, liver disease patients, especially those with cirrhosis and portal hypertension, have impaired intestinal mucosal barrier function and a very high rate of intestinal flora translocation, making intra-abdominal infections and spontaneous bacterial peritonitis the most common and dangerous types of infection, a stark contrast to the infection spectrum of ordinary hospitalized patients. Furthermore, liver disease patients often have hyperbilirubinemia, which may interfere with the detection and interpretation of routine inflammatory markers. Currently, clinical early warning of secondary infections in liver disease patients mainly relies on non-specific indicators such as C-reactive protein, white blood cell count, and neutrophil percentage. However, in the context of liver disease, these indicators often react sluggishly or show abnormal baselines, failing to provide early and accurate predictions. For example, C-reactive protein, synthesized by the liver, may be insufficiently produced in patients with severe liver failure, resulting in no significant increase in its level even during severe infections, leading to "false negatives." Similarly, due to hypersplenism, white blood cell and platelet counts are inherently low, making it difficult to reflect the true state of infection. While some existing technologies have investigated the role of individual proteins in general infections or sepsis, none have focused on researching and applying them to secondary infections in the specific context of "liver disease." In summary, current technologies cannot solve the fundamental problems of atypical immune responses and inaccurate conventional indicators in patients with liver disease. Therefore, there is an urgent need in this field to find specific molecular markers that can overcome the interference of the liver disease background and directly reflect the infection risk specific to patients with liver disease, in order to develop truly applicable early and accurate warning tools for people with liver failure. Summary of the Invention

[0003] (I) Technical Problem to be Solved: Based on the shortcomings of existing technologies, the purpose of this invention is to address the problem that existing technologies cannot accurately predict the occurrence of secondary infections in patients with HBV-related liver failure at an early stage. This invention provides a protein combination, predictive reagent kit, and predictive system that are unaffected by background liver disease, have high specificity, and good sensitivity, enabling early identification and intervention in high-risk patients. The solution of this invention can, in the early stages of patient admission, assess the risk of secondary infection and even death within 28 days based on the specificity of their plasma samples.

[0004] (II) Technical Solution

[0005] In a first aspect, the present invention provides the use of a protein combination in the preparation of a product for predicting secondary infections in patients with hepatitis B-related liver failure, said protein combination comprising lysozyme LYZ, calmodulin 1 (CALM1), heparin cofactor II (SERPIND1), and dermatan DPT.

[0006] According to a preferred embodiment of the present invention, the protein combination further comprises thymosin β4X, vitamin D-binding protein, C-reactive protein, α2-macroglobulin, platelet factor 4 variant 1, and α-synuclein.

[0007] Secondly, the present invention provides a detection kit for predicting secondary infections in patients with hepatitis B-related liver failure, comprising reagents for specifically detecting lysozyme LYZ, calmodulin 1 (CALM1), heparin cofactor II (SERPIND1), and dermatan DPT.

[0008] According to a preferred embodiment of the present invention, the reagent includes an antibody or fragment thereof that specifically binds to lysozyme (LYZ), an antibody or fragment thereof that specifically binds to calmodulin 1 (CALM1), an antibody or fragment thereof that specifically binds to heparin cofactor II (SERPIND1), and an antibody or fragment thereof that specifically binds to dermatan (DPT).

[0009] According to a preferred embodiment of the present invention, the kit further comprises reagents for specifically detecting one or more proteins selected from thymosin β4X, vitamin D-binding protein, C-reactive protein, α2-macroglobulin, platelet factor 4 variant 1, and α-synuclein; the reagents include antibodies or fragments thereof capable of specifically binding to the corresponding proteins.

[0010] According to a preferred embodiment of the present invention, the kit further comprises reagents or materials for obtaining plasma from a patient.

[0011] According to a preferred embodiment of the present invention, the detection kit is suitable for detecting lysozyme LYZ, calmodulin 1 (CALM1), heparin cofactor II (SERPIND1), and dermatan DPT by liquid chromatography-mass spectrometry or immunoassay.

[0012] Alternatively, the detection kit is suitable for detecting lysozyme LYZ, calmodulin 1 (CALM1), heparin cofactor II (SERPIND1), dermatan DPT, thymosin β4X, vitamin D-binding protein, C-reactive protein, α2-macroglobulin, platelet factor 4 variant 1, and α-synuclein by liquid chromatography-mass spectrometry or immunoassay techniques; the immunoassay technique includes chemiluminescence.

[0013] Thirdly, the present invention provides a method for predicting the risk of secondary infection before anti-infective therapy in patients with hepatitis B-related acute-on-chronic liver failure, comprising the following steps:

[0014] S1. Sample index detection: Obtain the patient's plasma sample to be tested, and quantitatively detect the protein content of lysozyme, calmodulin 1, heparin cofactor II and dermatan in the plasma sample to be tested. At the same time, detect the values ​​of the patient's clinical biochemical index aspartate aminotransferase and total bilirubin to obtain six sets of test data.

[0015] S2. Risk probability prediction: The four sets of protein content quantitative detection data and two sets of clinical biochemical index values ​​obtained in step S1 are used as input feature variables and input into the pre-trained machine learning prediction model. The machine learning prediction model outputs a continuous secondary infection risk probability value between 0 and 1.

[0016] S3. Risk Level Determination: When the risk probability value is ≥0.5, the patient is determined to be at high risk of secondary infection; when the risk probability value is <0.5, the patient is determined to be at low risk of secondary infection.

[0017] The machine learning prediction model is constructed based on the logistic regression algorithm. The training process of the model includes: taking the set of feature variables of the training set samples as input and the actual secondary infection outcome of the training set samples as output labels, optimizing the model parameters through the maximum likelihood estimation method until the model prediction accuracy meets the preset threshold; in the set of feature variables of the training set samples, each sample corresponds to a set of feature variable data, and each set of feature variable data consists of the quantitative detection data of protein content of LYZ, CALM1, SERPIND1, and DPT in the plasma of the sample, as well as the clinical biochemical index values ​​of AST and Tbil corresponding to the sample.

[0018] Fourthly, the present invention provides a system for predicting the risk of secondary infection in patients with hepatitis B-related liver failure (HBV-ACLF) prior to anti-infective therapy, comprising:

[0019] The data acquisition module is used to acquire the following data:

[0020] The data on the detection of the patient's clinical biochemical indicators aspartate aminotransferase (AST) and total bilirubin (Tbil), as well as the data on the levels of lysozyme (LYZ), calmodulin 1 (CALM1), heparin cofactor II (SERPIND1), and dermatan (DPT) in the plasma sample.

[0021] The risk prediction module, connected to the data acquisition module, has a built-in machine learning prediction model. It receives the content detection data of AST, Tbil, LYZ, CALM1, SERPIND1 and DPT, inputs them into the machine learning prediction model to obtain a risk probability value, and outputs a judgment result on the risk of secondary infection of the patient based on the comparison result of the risk probability value and the preset cutoff value. If the risk probability value is higher than the preset cutoff value, the judgment result is output that the patient is at high risk of secondary infection.

[0022] The machine learning prediction model is built based on the logistic regression algorithm. The training process of the model includes: taking the set of feature variables of the training set samples as input and the actual secondary infection outcome of the training set samples as output labels, optimizing the model parameters through the maximum likelihood estimation method until the model prediction accuracy meets the preset threshold; in the set of feature variables of the training set samples, each sample corresponds to a set of feature variable data, and each set of feature variable data consists of the quantitative detection data of protein content of LYZ, CALM1, SERPIND1, and DPT in the plasma of the sample, as well as the clinical biochemical index values ​​of AST and Tbil corresponding to the sample.

[0023] Preferably, the risk prediction module includes a readable carrier, which is software or a chip.

[0024] Preferably, the cutoff value is 0.5.

[0025] Preferably, the model is constructed and trained based on a prospective multi-center discovery cohort, and its effectiveness is verified in an independent validation cohort.

[0026] Preferably, during model construction, the logistic regression algorithm can also be replaced by other classification algorithms known in the art, such as support vector machines, random forests, or gradient boosting decision trees.

[0027] Preferably, the system further comprises:

[0028] The plasma sample acquisition module is used to acquire plasma samples from patients.

[0029] The protein quantification detection module is connected to the plasma sample acquisition module and is used to detect the contents of lysozyme LYZ, calmodulin 1 CALM1, heparin cofactor II SERPIND1 and dermatin DPT in the plasma sample using liquid chromatography-mass spectrometry or immunoassay, and transmit the quantitative detection data to the data acquisition module.

[0030] Preferably, the protein quantification detection module includes a detection device for performing liquid chromatography-mass spectrometry or immunoassay techniques, and a data analysis unit for processing the detection data to obtain protein quantification results; the immunoassay technique includes chemiluminescence immunoassay.

[0031] (III) Beneficial effects: This invention is based on the discovery that the protein combination of "lysozyme LYZ, calmodulin 1 CALM1, heparin cofactor II SERPIND1 and dermatan DPT" is strongly correlated with the occurrence of secondary infection in HBV-ACLF patients. By utilizing the changes in the content of the proteins involved in this combination in plasma samples, it is possible to accurately predict secondary infection in patients with hepatitis B-related liver failure and effectively avoid the interference of liver disease itself on traditional inflammatory markers.

[0032] Specifically, the protein combination consists of biomarkers screened through non-targeted proteomics, which are significantly enriched in the "inflammation" and "coagulation" pathways. The core mechanism of secondary infection in liver disease is precisely "inflammation-coagulation imbalance," and the biomarkers of this invention directly reflect this core pathophysiological process, exhibiting high specificity due to their targeting of the specific characteristics of liver disease. Furthermore, the protein combination is not a traditional acute-phase reactive protein (such as CRP) that is easily affected by liver function; therefore, changes in the expression of these proteins more accurately reflect the body's potential response to infection in the context of liver disease, thus overcoming the technical problems of "failure" and "false negatives" of traditional indicators. This protein combination shows significant changes in the early stages of patient admission, even before clinical signs of infection, providing the possibility for early intervention. It possesses high prospectivity and high sensitivity, providing a critical time window for early preventative treatment.

[0033] The protein combination provided by this invention, consisting of "lysozyme LYZ, calmodulin 1 CALM1, heparin cofactor IISERPIND1, and dermatan DPT", can independently constitute a minimal predictive biomarker combination, or it can be used in combination with different auxiliary clinical indicators (such as prothrombin time international normalized ratio, platelet count, etc.). The aforementioned protein combination is used as the core indicator, and prothrombin time international normalized ratio, platelet count, etc. are used as reference indicators.

[0034] Furthermore, this invention also constructs a prediction model based on the aforementioned protein combination, aspartate aminotransferase (AST), and total bilirubin (Tbil), which has the following technical advantages:

[0035] (1) Excellent model performance: the prediction accuracy exceeds that of existing technologies and the results are reliable.

[0036] The predictive ability for secondary infections is outstanding: The predictive model, based on the combination of four core proteins (LYZ, CALM1, SERPIND1, DPT) and clinical indicators such as AST and Tbil, performed excellently in the independent validation cohort. The AUROC (Area Under the Receiver Operating Characteristic) for predicting secondary infections reached 0.873, which means that the model has a very strong ability to distinguish whether secondary infection occurs. The predictive accuracy is significantly higher than that of conventional inflammatory indicators (such as C-reactive protein, procalcitonin, etc.) and existing authoritative clinical scoring systems (such as MELD score, CLIF-C ACLF score, etc.).

[0037] The model demonstrates extremely high accuracy in predicting short-term mortality risk: it can not only predict secondary infections but also achieve a high-precision assessment of patients' 28-day mortality risk, with an AUROC of 0.957, approaching the ideal predictive performance (1.0). This performance advantage can help clinicians identify high-risk patients in advance, adjust treatment strategies accordingly, and reduce the probability of poor prognosis.

[0038] (2) It has the timeliness of application: it enables early risk assessment and seizes the treatment window.

[0039] The model can complete risk assessment within 48 hours of patient admission, before obvious clinical signs of infection (such as fever, elevated white blood cell count, or symptoms at the site of infection) appear. Compared to the traditional approach of waiting for "intervention after clinical symptoms appear," this model provides clinicians with a crucial "early warning window." By identifying high-risk patients for secondary infections in advance, it supports doctors in initiating preventative anti-infection treatment as early as possible, reducing the probability of infection at its source, or controlling the disease in its early stages to prevent progression to severe illness and significantly improve patient prognosis.

[0040] (3) It has a rigorous scientific basis: it is based on a clear molecular mechanism and has strong interpretability.

[0041] The model is not built upon a "data-driven black-box algorithm," but rather based on the core molecular mechanism of secondary infection in HBV-ACLF patients: "inflammation-coagulation imbalance."

[0042] The selection of core biomarkers (such as heparin cofactor II involved in coagulation regulation and lysozyme associated with inflammatory responses) is directly related to the pathophysiological process of "inflammatory activation and coagulation dysfunction," ensuring that the biomarker combination has clear biological significance. Compared with predictive models that lack mechanistic support and rely solely on statistical association, this model can clearly explain "why a specific biomarker combination can predict risk," which not only enhances clinicians' confidence in the model results but also provides a scientific direction for subsequent biomarker optimization or disease mechanism exploration.

[0043] (4) Strong clinical translation: few variables and clear detection, suitable for large-scale application.

[0044] The variable composition is simple and controllable: the model input variables only include "a maximum of 10 proteins (4 core + 6 optional) + 2 routine clinical biochemical indicators (AST, Tbil)", with a small number of variables and clear objectives, eliminating the need for complex multi-dimensional data collection and reducing the operational threshold for clinical application.

[0045] Standardized reagent kit development: With core detection targets being clearly defined proteins and routine biochemical indicators, standardized detection kits can be developed based on various platforms such as enzyme-linked immunosorbent assay (ELISA), chemiluminescence immunoassay, protein chip technology, or liquid chromatography-mass spectrometry (LC-MS), avoiding result discrepancies caused by inconsistent detection methods. This characteristic allows for easy adaptation to multi-center clinical scenarios, supporting large-scale deployment and application, and addressing the pain point of some complex models achieving good laboratory results but facing difficulties in clinical implementation.

[0046] (5) Multi-dimensional functional dimensions: taking into account both risk prediction and supporting clinical decision-making.

[0047] The predictive model of this invention breaks through the limitations of "single predictive function" and can simultaneously meet two core clinical needs: ① prediction of secondary infection risk: helping doctors identify high-risk groups for infection and guide preventive treatment; ② prediction of 28-day mortality risk: assisting doctors in judging the short-term prognosis of patients and developing individualized intensive care and treatment plans.

[0048] The predictive model developed based on the protein combination described in this invention can provide more comprehensive risk information for clinical decision-making, reduce doctors' reliance on multiple assessment tools, and improve diagnostic and treatment efficiency. It has important clinical practice value, especially for diseases such as HBV-ACLF, which are complex and have large differences in prognosis. Attached Figure Description

[0049] Figure 1 This is the overall technical roadmap of the present invention.

[0050] Figure 2 To discover the results of cohort non-targeted proteomics analysis.

[0051] Figure 3 To validate the cohort-targeted proteomics validation results.

[0052] Figure 4 This refers to the process of constructing and selecting features for the prediction model.

[0053] Figure 5 This represents the model's predictive performance in the discovery queue.

[0054] Figure 6 To validate the model's performance and biomarkers in an independent validation cohort. Detailed Implementation

[0055] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to specific embodiments. It should be understood that the specific embodiments described herein are for illustrative purposes only and are not intended to limit the invention. Unless otherwise specified, the experimental methods in the following embodiments are conventional methods. Unless otherwise specified, the raw materials and reagents used in the following embodiments are commercially available products.

[0056] like Figure 1 The diagram shown is a general technical roadmap of the present invention. Figure 1 The technical route shown is divided into six modules, illustrating the complete process of this invention from research design to clinical application. Specific details are as follows: Module 1: Patient recruitment and sample collection process for the discovery and validation cohorts; Module 2: Non-targeted proteomics analysis and targeted validation process for plasma samples; Module 3: Bioinformatics analysis (including differentially expressed protein screening, functional enrichment analysis, and protein-protein interaction network construction); Module 4: Construction, training, and optimization process of the machine learning model; Module 5: Internal and external independent validation of model performance; Module 6: Application and translation path of the final clinical prediction model and reagent kit.

[0057] Step 1: Design of the cohort study (discovery cohort and validation cohort).

[0058] Discovery cohorts: Two types of patients were recruited from Beijing Ditan Hospital, Shanghai Jiao Tong University Affiliated First People's Hospital, and Huashan Hospital: HBV-ACLF patients with secondary infection (SI patients, n=38) and those without secondary infection (NSI patients, n=76) for preliminary screening of biomarkers.

[0059] Validation cohorts: Patients were recruited from the same three hospitals, including those with secondary infections (n=20) and those without infections (n=40), to validate the results of the discovery cohort. Sample types included blood tests, plasma proteins, and follow-up data. Sample and clinical information were collected through a "data collection" process.

[0060] Step 2: Proteomics Analysis (Non-targeted + Targeted Validation)

[0061] First, untargeted plasma proteomics analysis was performed on plasma samples to screen for differentially expressed proteins. Then, targeted plasma proteomics validation was conducted to clarify the expression levels of core biomarkers, providing a molecular basis for subsequent model construction.

[0062] Step 3: Bioinformatics Analysis

[0063] Bioinformatics techniques such as differentially expressed protein screening, functional enrichment analysis, and protein-protein interaction network construction are used to explore the biological functions and associations of proteins, and further clarify the core protein combinations used for prediction.

[0064] Step 4: Machine Learning Model Building and Training

[0065] Model building (Machine learning model - Building models): Regression learning models are used for training and optimization. Through the "Model Training & Evaluation" process, the predictive performance of the model is continuously improved (the model's discriminative ability is reflected by the AUROC curve).

[0066] Model objective: To construct a predictive model for the primary outcome of "28-day liver-related mortality" by combining clinical data with follow-up time (≥28 days).

[0067] Step 5: Model Validation

[0068] First, internal validation (internal evaluation of the training set) is carried out through the "Model Validation" stage. Then, external independent validation ("Test Set (X)" is conducted. The model's ability to distinguish between high-risk and low-risk patients is evaluated through "Validation Metrics" such as AUROC, F1 score, accuracy, and precision, as well as the decision curve. Finally, the "Validated Model" is obtained.

[0069] Step Six: Clinical Translational Application

[0070] After model validation, a clinical prediction model and testing kit are developed and promoted to clinical practice along the "application translation path" to achieve accurate prediction of the risk of secondary infection and short-term death in HBV-ACLF patients and assist clinical decision-making.

[0071] The overall technical approach of the present invention will be described in detail below with reference to Examples 1-2.

[0072] Example 1: This example illustrates the discovery and screening process of biomarkers, with the following steps.

[0073] 1. Study Subjects and Sample Collection: With informed consent and meeting the inclusion criteria, patients with hepatitis B virus-related acute-on-chronic liver failure were prospectively recruited at three clinical centers: Beijing Ditan Hospital affiliated to Capital Medical University, the First Affiliated Hospital of Zhejiang University School of Medicine, and Huashan Hospital affiliated to Fudan University.

[0074] The cohort included 114 patients, of whom 38 had secondary infections and 76 had no infections (matched 1:2).

[0075] Validation cohort: A total of 60 patients were included, with 20 in the secondary infection group and 40 in the non-infection group. All patients underwent peripheral venous blood collection (10 mL) using EDTA anticoagulant tubes within 48 hours of admission. Within 30 minutes of collection, the blood was centrifuged at 4°C and 1500×g for 15 minutes. The supernatant plasma was carefully aspirated, aliquoted into pyrogen-free cryovials, and immediately stored at -80°C to avoid repeated freeze-thaw cycles.

[0076] 2. Plasma sample pretreatment: Centrifuge the collected EDTA-anticoagulated whole blood at 1200×g for 10 minutes at room temperature, carefully aspirate the supernatant plasma, and store it at -80°C. For proteomics pretreatment, take 5 μL of thawed plasma and add 50 μL of lysis buffer (containing 8M urea and 100mM ammonium bicarbonate) for denaturation.

[0077] 3. Protein digestion (FASP method): Take 100 μg of protein solution, add 200 μL of UA buffer (8 M urea, 150 mM Tris-HCl, pH 8.0), and centrifuge at 14,000 × g for 15 minutes in a 30 kD ultrafiltration centrifuge tube (Millipore, USA). Discard the filtrate. Add 100 μL of UA buffer and centrifuge again. Then add 100 μL of 50 mM iodoacetamide (dissolved in UA buffer) and incubate at room temperature in the dark for 20 minutes to alkylate and reduce cysteine ​​residues. After centrifugation, wash twice with 100 μL of UA buffer, and then twice with 100 μL of 50 mM ammonium bicarbonate buffer. Finally, add 1 μg of sequencing-grade trypsin (Promega, USA) (dissolved in 40 μL of 50 mM ammonium bicarbonate buffer) and incubate overnight (16 hours) at 37°C. The following day, peptides were collected by centrifugation at 14,000×g for 15 minutes and then desalted and purified using a C18 StageTip. Peptide concentrations were determined at 280 nm using a NanoDrop One micro-volume spectrophotometer (Thermo Fisher Scientific, USA).

[0078] 4. Liquid Chromatography-Mass Spectrometry (LC-MS) Analysis: An Orbitrap Astral mass spectrometer coupled with a Vanquish Neo ultra-high performance liquid chromatography system (Thermo Fisher Scientific, USA) was used for analysis. 1 μg of peptide sample was loaded onto a 50 cm Low-Load µPACT™ Neo HPLC column (Thermo Scientific).

[0079] Mobile phase A: 0.1% formic acid aqueous solution; Mobile phase B: 0.1% formic acid acetonitrile solution. The flow rate was set to 1.25 μL / min, and a linear gradient of 8 min was used for separation: 0–0.1 min, 4–6% B; 0.1–1.1 min, 6–12% B; 1.1–4.3 min, 12–25% B; 4.3–6.1 min, 25–45% B; 6.1–6.5 min, 45–99% B; 6.5–8 min, 99% B.

[0080] Mass spectrometry employs a data-independent acquisition mode. Level 1 full scan range: 380-980 m / z, resolution 240,000. DIA MS / MS scan range: 150-2000 m / z, isolation window 2 m / z. Collision energy is set to 25%.

[0081] 5. Bioinformatics Analysis and Biomarker Screening: Raw mass spectrometry data were analyzed using Spectronaut software (version 18, Biognosys AG), and the UniProt human proteome database was searched. Identified proteins were normalized and quantified. For example... Figure 2 As shown, this is the result of non-targeted proteomics analysis of the discovery cohort.

[0082] First, t-tests (P < 0.05) and fold change (FC > 1.5 or FC < 0.67) were used to screen for differentially expressed proteins between the infected and non-infected groups in the cohort, resulting in 974 proteins. Figure 2 Figure (B) shows a clustering heatmap based on these differentially expressed proteins, visually demonstrating the significant differences in protein expression profiles between the infected and non-infected groups. Figure 2 (A) shows the statistical bar chart of differentially expressed proteins, which shows that among 974 differentially expressed proteins, 159 were upregulated and 815 were downregulated.

[0083] Further enrichment analysis of the differentially expressed proteins using R software (version 4.4.0) using KEGG and WikiPathways revealed significant enrichment in inflammation- and coagulation-related pathways such as "complement and coagulation cascades" and "platelet activation." Figure 2 As shown in (C), the pie chart classifying differentially expressed proteins by function shows that 27.78% of the proteins are related to inflammation / immunity, 19.84% to metabolic processes, 9.52% to coagulation, 7.94% to the cytoskeleton / connectivity, 6.35% to apoptosis / stress response / signal transduction, 5.56% to cell secretion and transport, and 6.35% to cell transcription and translation control.

[0084] Finally, the minimum redundancy maximum correlation algorithm was used to screen the top 20 key proteins from the differentially expressed proteins. For example... Figure 2 Figure (D) shows the Spearman correlation (Spearman correlation coefficient) analysis of 20 differentially expressed proteins and clinical indicators (CRP, NE%, PT, INR), displaying the top 20 related proteins and their correlation strength and direction.

[0085] Subsequently, these 20 proteins were sequentially incorporated into a logistic regression model, and the predictive performance of different protein combinations was evaluated using 10 replicates of five-fold cross-validation (AUROC as the metric). The combination of 10 proteins (CALM1, LYZ, TMSB4X, GC, CRP, SERPIND1, DPT, A2M, PF4V1, SNCA) was determined to have the best predictive power. The expression data of these 10 proteins, along with clinical variables, were included in a univariate logistic regression analysis. Variables with P < 0.05 (including ASC, ALT, AST, TBIL, INR, WBC, NE%, and the 10 proteins) were further included in a multivariate logistic regression model. LYZ, CALM1, SERPIND1, DPT, AST, and Tbil were ultimately identified as independent factors predicting secondary infections, and their respective regression coefficients were obtained and used to construct the final predictive model.

[0086] like Figure 3 The image shows the validation results for cohort-targeted proteomics. Figure 3 (A) is a statistical graph of differentially expressed proteins. 31 differentially expressed proteins were successfully validated in the validation cohort, of which 5 proteins were upregulated and 26 proteins were downregulated. Figure 3 (B) is a clustering heatmap of differentially expressed proteins: the heatmap presents the expression levels of 31 differentially expressed proteins in different samples. The intensity of the color in the heatmap represents the level of protein expression, with red indicating high expression and blue indicating low expression. The graph clearly shows the differences in expression patterns between the infected and uninfected groups. Figure 3 (C) is a pie chart of protein function classification, showing the functional distribution of these differentially expressed proteins. Among them, 24.14% are related to inflammation / immunity, 27.59% are related to coagulation pathways, 6.90% are related to the cytoskeleton, 6.90% are related to transport and binding, and the rest are involved in metabolic, oxidative stress and other functional categories.

[0087] Figure 3 Figure (D) shows the Spearman correlation coefficient analysis of 20 differentially expressed proteins with four clinical indicators (CRP, NE%, PT, and INR), demonstrating the correlation between the top 20 related proteins and clinical indicators. Among them, LYZ showed a significant positive correlation with inflammatory markers (such as CRP and NE%), suggesting its role in the inflammatory response; CALM1, SERPIND1, and DPT showed significant negative correlations with coagulation markers (such as PT and INR), indicating that these proteins are closely related to the coagulation pathway, further supporting the molecular mechanism of "inflammation-coagulation imbalance".

[0088] Example 2: This example utilizes LYZ, CALM1, SERPIND1, DPT, AST, and Tbil as independent factors for predicting secondary infections, constructs a predictive model, and validates its performance. The steps include the following.

[0089] (1) Model building

[0090] Based on the data from the cohort (n=114) identified in Example 1, a logistic regression prediction model was constructed. The model input features were the Z-score-normalized expression levels of LYZ, CALM1, SERPIND1, and DPT proteins, as well as the raw clinical test values ​​of AST and Tbil.

[0091] The model formula is: P(SI) = 1 / (1 + e^(-z))

[0092] in:

[0093]

[0094] In the above formula, b0 is the intercept, and b1 to b6 are the regression coefficients corresponding to each feature. These coefficients are obtained from the training data using the maximum likelihood estimation method and are the core parameters of the model. The model output P(SI) is the probability of a patient developing secondary infection, ranging from 0 to 1. A decision threshold (cutoff value) is set at 0.5, meaning that when P(SI) ≥ 0.5, it is considered a high risk. 0.5 is the Youden index.

[0095] like Figure 4 The diagram illustrates the process of constructing the prediction model and selecting features according to the present invention.

[0096] in, Figure 4 (A) is the partial least squares discriminant analysis (PLS-DA) score plot. Dimensionality reduction analysis was performed on infected and non-infected patients based on protein expression levels. Different colored points in the figure represent the two groups of patients, showing clear population separation in the dimensionality reduction space. This indicates that differences in protein expression can effectively distinguish between "infected" and "non-infected" states, providing a basis for subsequent screening of biomarkers.

[0097] Figure 4 (B) compares the AUROC performance of five machine learning algorithms: it shows the AUROC (Area Under the Receiver Operating Characteristic) performance of logistic regression, support vector machine, random forest, XGBoost, and LASSO in predicting secondary infections. The closer the AUROC is to 1, the stronger the model's discriminative ability. The comparison shows that the AUROC of random forest, XGBoost, and LASSO is 0.99, while the AUROC of logistic regression and support vector machine is 1. Figure 4(C) is a bar chart showing the overall performance of the five algorithms: comparing their performance across multiple dimensions, including accuracy, balanced precision, precision, recall, and F1 score. The "bar height" for each dimension in the bar chart reflects the overall capability of the algorithm. The aforementioned comparison results demonstrate that all five machine learning algorithms are well-suited for this scenario.

[0098] Figure 4 The curve (D) represents the change in the number of features versus the model's AUROC: the horizontal axis represents the number of protein features input into the model, and the vertical axis represents the model's AUROC value. The curve shows that the model's AUROC reaches its peak when 10 proteins are included, indicating that the combination of these 10 proteins (CALM1, LYZ, TMSB4X, GC, CRP, SERPIND1, DPT, A2M, PF4V1, SNCA) is a feature set with superior predictive performance. However, using a combination of 4 proteins (LYZ, CALM1, SERPIND1, and DPT) can also independently constitute a minimal predictive biomarker, still exhibiting good predictive performance. To simplify the model, the machine learning prediction model of this invention is constructed using a combination of 4 proteins and 2 clinical biochemical indicators.

[0099] Figure 4 (E) is a lollipop plot showing the expression changes of 10 key proteins: The fold change (Log2FC) and statistical significance (P-value) of the final 10 selected key proteins are displayed in "lollipop" format. The length of each "lollipop" represents the degree of expression difference, visually presenting the expression trends and significance of these proteins between the infected and uninfected groups. As shown in the figure, TMSB4X, DPT, CALM1, SERPIND1, CRP, and LYZ showed significant differences between the infected and uninfected groups. Figure 4 (F) is a box plot showing the intergroup expression of 10 key proteins: The box plot illustrates the expression level distribution of the 10 key proteins in the infected group (red) and the non-infected group (blue). CRP and LYZ were positively correlated with positive results (infected group), while the remaining proteins CALM1, TMSB4X, GC, SERPIND1, DPT, A2M, PF4V1, and SNCA were negatively correlated with positive results (***P<0.001). This indicates that the intergroup differences were statistically significant, further validating that these proteins can serve as core biomarkers for distinguishing between infection and non-infection.

[0100] 2. Internal validation of model performance

[0101] like Figure 5 The image shows the model's prediction performance in the discovery queue. Figure 5(A) shows the receiver operating characteristic (ROC) curves for predicting secondary infections, comparing the performance of different models in predicting secondary infections. Model 1 is a model constructed based on four protein detection data (LYZ, CALM1, SERPIND1, and DPT) plus two clinical indicators (aspartate aminotransferase AST and total bilirubin Tbil). Model 2 is a model constructed based on a combination of ten proteins (CALM1, LYZ, TMSB4X, GC, CRP, SERPIND1, DPT, A2M, PF4V1, and SNCA). As shown, Model 1 has an AUROC of 0.980, significantly better than the ROC curves corresponding to traditional inflammatory indicators (such as CRP and NE%), indicating that Model 1 has a very strong ability to distinguish secondary infections. Figure 5 (B) and (C) represent the balanced accuracy (B) and F1 score (C) for secondary infection prediction. Model 1 outperforms traditional metrics (CRP, NE%, WBC) and Model 2, validating the advantages of Model 1 in secondary infection prediction from two dimensions: "overall accuracy" and "precision-recall balance".

[0102] Figure 5 (D) represents the receiver operating characteristic (ROC) curve for predicting 28-day mortality risk, demonstrating the performance of different models in predicting 28-day mortality risk. Model 1 has an AUROC of 0.883, which is superior to traditional clinical scoring systems (such as MELD, CLIF-C AD, etc.), reflecting its excellent ability in predicting short-term mortality risk. Figure 5 Figures (E) and (F) show the balance accuracy and F1 score of 28-day mortality risk prediction. In Figure E (balance accuracy) and Figure F (F1 score), Model 1 significantly outperforms authoritative clinical scoring systems such as MELD and CLIF-C AD, further validating its comprehensive performance in mortality risk prediction.

[0103] Figure 5 (G) represents the validation of the infectivity prediction results (confusion matrix). The horizontal axis "Actual" represents the actual category (0 = non-infected, 1 = infected); the vertical axis "Predicted" represents the model's predicted category (0 = non-infected, 1 = infected). There were 3 false positives (actually non-infected, but the model misclassified them as infected), 32 true positives (correctly identified by the model), 73 true negatives (correctly identified by the model), and 6 false negatives (actually infected, but the model misclassified them as non-infected). Figure 5(H) represents the 28-day mortality prediction results (confusion matrix): 1 false positive (no actual death, model misjudged as death risk), 9 true positives (correctly identified by the model), 73 true negatives (correctly identified by the model), and 26 false negatives (actual deaths, model misjudged as no death risk).

[0104] Figure 5 (I) shows the importance ranking of the predictor variables in Model 1, illustrating the importance of the six predictor variables (LYZ, CALM1, SERPIND1, DPT, AST, and Tbil). Among them, the core proteins LYZ, CALM1, SERPIND1, and DPT are more important than clinical indicators such as AST and Tbil, clarifying the core role of these four protein biomarkers in the model. Therefore, the machine learning prediction model of this invention is constructed using a combination of the aforementioned four core proteins and the two clinical biochemical indicators.

[0105] Depend on Figure 5 As can be seen, the predictive model of this application has verified its excellent performance in the discovery cohort from three dimensions: "prediction of secondary infection", "prediction of 28-day mortality risk" and "variable importance", which clarifies its advantages over traditional indicators and scoring systems and the contribution of core predictive variables.

[0106] In summary, within the discovery cohort, the model predicted a secondary infection AUROC of 0.980 (95% CI: 0.961–0.999). Its predictive accuracy significantly outperformed traditional indicators CRP (AUROC: 0.645), WBC (AUROC: 0.674), and NE% (AUROC: 0.642), with all DeLong test p-values ​​less than 0.001.

[0107] 3. External Independent Validation: To assess the model's generalization ability, absolute quantification of four proteins—LYZ, CALM1, SERPIND1, and DPT—was performed using targeted proteomics methods (such as parallel response monitoring) in an independent validation cohort (n=60), and AST and Tbil values ​​were collected from patients. The quantitative data were then input into the trained model for calculation.

[0108] The calculation results show that the model has an AUROC of 0.873 (95% CI: 0.771-0.974) for predicting secondary infection in the independent validation cohort and an AUROC as high as 0.957 (95% CI: 0.897-1.000) for predicting the risk of death at 28 days, which fully demonstrates the robustness and clinical applicability of the model.

[0109] like Figure 6From three aspects—model performance (ROC, balanced accuracy, F1 score), prediction result validation (confusion matrix), and biomarker mechanism (expression differences)—the system demonstrates the advantages of Model 1 in predicting "secondary infection" and "28-day mortality risk" in the validation cohort.

[0110] Figure 6 (A) shows the ROC curves of different models predicting secondary infections in the validation queue. The AUROC (area under the curve) of Model 1 is 0.873, indicating that it has a good predictive ability in distinguishing between "secondary infection" and "non-infection". The closer the curve is to the upper left corner, the stronger the model's ability to distinguish. Figure 6 (D) shows the ROC curves of different models predicting 28-day mortality risk in the validation cohort. Among them, Model 1 has an AUROC as high as 0.957, indicating that it has excellent discriminative ability in predicting "28-day mortality risk", approaching perfect classification.

[0111] Balanced accuracy and F1 score can be used to evaluate the classification performance of a model. Balanced accuracy: Eliminates the effects of class imbalance and measures the model's average accuracy on both positive and negative samples. F1 score: Combines precision and recall, reflecting the model's ability to identify the "positive class". Figure 6 (B) and (C) represent the balance accuracy and F1 score for secondary infection prediction, respectively. The results show that Model 1 is significantly better than Model 2, CRP (C-reactive protein), NE% (neutrophil percentage), and WBC (white blood cell count) in both balance accuracy and F1 score, indicating that it performs better in the "secondary infection" classification task. Figure 6 (E) and (F) represent the balanced accuracy and F1 score for predicting 28-day mortality risk, respectively. This shows that Model 1 significantly outperforms Model 2, CLIF-C AD (chronic-acute liver failure-chronic decompensation), and MELD (end-stage liver disease model) in both balanced accuracy and F1 score, indicating that it performs better in the "28-day mortality risk" classification task.

[0112] The confusion matrix demonstrates the model's predictive accuracy through the cross-count of the "actual class - predicted class". Figure 6 G represents the confusion matrix of secondary infection prediction results. Among them, 36 cases were actually "not infected" and predicted as "not infected" (true negative, accurately identified by the model); 4 cases were actually "not infected" but predicted as "infected" (false positive, misjudged by the model); 12 cases were actually "infected" but predicted as "not infected" (false negative); and 8 cases were actually "infected" and predicted as "infected" (true positive, correctly identified by the model). Figure 6H represents the confusion matrix of the 28-day mortality risk prediction results. Among these, 44 cases actually "survived" and were predicted to "survive" (true negatives, correctly identified by the model); 10 cases actually "survived" but were predicted to "die" (false positives); 6 cases actually "died" but were predicted to "survive" (false negatives); and 0 cases actually "died" and were predicted to "die" (true positives, correctly identified by the model). These values ​​can be used to further calculate metrics such as precision and recall.

[0113] Figure 6 (I) illustrates the differences in expression levels of four core biomarkers between the "infected group" and the "non-infected group." Specifically, LYZ (lysozyme) was significantly upregulated in the infected group; CALM1 (calmodulin 1), SERPIND1 (serine protease inhibitor D1), and DPT (defensin β1) were significantly downregulated in the infected group. These differences validate their potential value as biomarkers for "secondary infection."

[0114] Example 3: This example provides a detection kit and its application for predicting secondary infections in patients with hepatitis B-related liver failure.

[0115] 1. Kit Components

[0116] The kit is a chemiluminescent immunoassay kit suitable for the quantitative detection of four target proteins in human plasma: lysozyme (LYZ), calmodulin 1 (CALM1), heparin cofactor II (SERPIND1), and dermatan (DPT). The kit contains a complete detection system.

[0117] (1) Solid-phase capture system: Four groups of superparamagnetic beads (3 μm in diameter) coated with specific mouse monoclonal antibodies, respectively. Among them, the anti-human LYZ antibody (clone A1B2) recognizes the C-terminal domain of the LYZ protein epitope, the anti-human CALM1 antibody (clone C3D4) targets the calcium-binding domain, the anti-human SERPIND1 antibody (clone E5F6) binds to the heparin binding site, and the anti-human DPT antibody (clone G7H8) recognizes the C-terminal fragment of dermatin. The four groups of magnetic beads were mixed in equal proportions to form a working suspension with a concentration of 0.5 mg / mL.

[0118] (2) Antibody detection system: Biotin-labeled mouse monoclonal antibody mixtures corresponding to four proteins, each antibody working concentration is 1 μg / mL, labeled with sulfonated NHS-biotin, and the biotin-labeled ratio of each antibody molecule is 3:1;

[0119] (3) Standards and controls: Six-point calibration curves (concentration gradients of 1, 5, 25, 100, 500, and 1000 ng / mL) were prepared using recombinant human LYZ, CALM1, SERPIND1, and DPT proteins (R&D Systems), and two levels of quality control were established (low-value control 15 ng / mL and high-value control 750 ng / mL). All standards and quality control used PBS buffer (pH 7.4) containing 1% BSA as the matrix.

[0120] (4) Signal detection system: horseradish peroxidase-labeled streptavidin (working concentration 0.5 μg / mL) and chemiluminescent substrate solution (solution A is 0.1 mol / L citrate buffer containing 0.5 mmol / L H2O2, solution B is 0.1 mol / L Tris-HCl buffer containing 1.25 mmol / L luminol and 1.8 mmol / L p-iodophenol);

[0121] (5) Auxiliary reagents: The sample dilution solution is PBS buffer (pH 7.4) containing 0.5% BSA and 0.05% Tween-20, and the 20-fold concentrated washing solution is Tris-HCl buffer (pH 8.0) containing 0.5 mol / L NaCl and 0.5% Tween-20.

[0122] 2. Detection steps (taking a fully automated chemiluminescence immunoassay analyzer as an example)

[0123] The specific detection procedure for this kit is as follows:

[0124] (1) Sample pretreatment: Take 10 μL of EDTA anticoagulated plasma and add 90 μL of special sample diluent, vortex mix for 30 seconds, and let stand at room temperature for 10 minutes;

[0125] (2) Initial incubation: Take 50 μL of diluted sample and 50 μL of mixed magnetic bead working suspension and add it to a polypropylene reaction cup. Place it in the 37℃ incubation tank of the fully automated chemiluminescence immunoassay analyzer and incubate with shaking at 600 rpm for 30 minutes.

[0126] (3) Magnetic separation and washing: Transfer the reaction vessel to the magnetic separation station, let it stand for 2 minutes, discard the supernatant, add 300 μL of 1× washing solution, vortex mix for 10 seconds, and repeat this washing step 3 times.

[0127] (4) Second incubation: Add 50 μL of mixed biotinylated detection antibody working solution and incubate at 37°C with shaking at 600 rpm for 20 minutes;

[0128] (5) Wash again: Repeat the magnetic separation and washing process of step (3) 3 times;

[0129] (6) Enzyme conjugate incubation: Add 50 μL of streptavidin-horseradish peroxidase working solution, incubate at 37 °C in the dark with shaking at 600 rpm for 15 minutes;

[0130] (7) Final washing: Repeat the magnetic separation and washing process in step (3) 4 times to ensure complete removal of unbound enzyme markers;

[0131] (8) Chemiluminescence detection: Immediately add 100 μL of premixed chemiluminescence substrate (solution A and solution B are mixed at a volume ratio of 1:1), and measure the relative luminescence unit value within 3 seconds after a 3-second delay;

[0132] (9) Result calculation: The instrument software automatically adopts a four-parameter logistic curve fitting algorithm, synchronously calculates the concentrations of the four target proteins in the sample according to the standard curves of each protein, and automatically performs quality control verification, and outputs a test report containing the quantitative results (ng / mL) of LYZ, CALM1, SERPIND1, and DPT. The entire detection process takes about 90 minutes.

[0133] 3. Risk assessment: Input the obtained four protein concentration values and the AST (U / L) and Tbil (μmol / L) values of this patient into the risk calculation software supporting the kit. The software incorporates the trained logistic regression model and coefficients in Example 2, can automatically calculate the probability of secondary infection risk P(SI) of this patient, and generate a report of "high risk" (P(SI) ≥ 0.5) or "low risk" (P(SI) < 0.5) to assist clinicians in making decisions.

[0134] Finally, it should be noted that: The above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit them; Although the present invention has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that: They can still modify the technical solutions described in the foregoing embodiments, or perform equivalent replacements on some or all of the technical features; And these modifications or replacements, or in the case where the technical features in the above embodiments do not conflict with each other, can be combined in the manner described in the embodiments, and these modifications, replacements or combinations do not make the essence of the corresponding technical solutions deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. The use of a protein combination in the preparation of a product for predicting secondary infections in patients with hepatitis B-related liver failure, characterized in that, The protein combination comprises lysozyme LYZ, calmodulin 1 CALM1, heparin cofactor II SERPIND1, and dermatin DPT; the product includes reagents capable of specifically and quantitatively detecting the protein content of lysozyme LYZ, calmodulin 1 CALM1, heparin cofactor II SERPIND1, and dermatin DPT in plasma samples.

2. The use according to claim 1, characterized in that, The protein combination also includes thymosin β4X, vitamin D-binding protein, C-reactive protein, α2-macroglobulin, platelet factor 4 variant 1, and α-synuclein; the product also includes reagents that can specifically quantify the content of one or more of the following proteins in a plasma sample: thymosin β4X, vitamin D-binding protein, C-reactive protein, α2-macroglobulin, platelet factor 4 variant 1, and α-synuclein.

3. A diagnostic kit for predicting secondary infection in patients with hepatitis B-related liver failure (HBV-ACLF), characterized in that, It contains reagents for the specific detection of lysozyme LYZ, calmodulin 1 CALM1, heparin cofactor II SERPIND1, and dermatin DPT; the reagents can quantitatively detect the protein content of lysozyme LYZ, calmodulin 1 CALM1, heparin cofactor II SERPIND1, and dermatin DPT in plasma samples.

4. The detection kit according to claim 3, characterized in that, It also contains reagents for the specific quantitative detection of one or more of the following proteins: thymosin β4X, vitamin D-binding protein, C-reactive protein, α2-macroglobulin, platelet factor 4 variant 1, and α-synuclein.

5. The detection kit according to claim 3, characterized in that, The reagents include antibodies or fragments thereof that specifically bind to lysozyme LYZ, antibodies or fragments thereof that specifically bind to calmodulin 1 (CALM1), antibodies or fragments thereof that specifically bind to heparin cofactor II (SERPIND1), and antibodies or fragments thereof that specifically bind to dermatan DPT.

6. The detection kit according to claim 3, characterized in that, The kit also contains reagents or materials for obtaining plasma from a patient.

7. The detection kit according to claim 3, characterized in that, The detection kit is suitable for detecting lysozyme LYZ, calmodulin 1 (CALM1), heparin cofactor IISERPIND1, and dermatin DPT using liquid chromatography-mass spectrometry or immunoassay techniques; the immunoassay technique includes chemiluminescence.

8. A system for predicting the risk of secondary infection in patients with hepatitis B-related liver failure (HBV-ACLF) prior to anti-infective therapy, characterized in that, include: The data acquisition module is used to acquire the following data: The data on the detection of clinical biochemical indicators aspartate aminotransferase (AST) and total bilirubin (Tbil), as well as the data on the content of lysozyme (LYZ), calmodulin 1 (CALM1), heparin cofactor II (SERPIND1), and dermatan (DPT) in plasma samples. The risk prediction module, connected to the data acquisition module, has a built-in machine learning prediction model. It receives the content detection data of AST, Tbil, LYZ, CALM1, SERPIND1, and DPT, inputs this data into the machine learning prediction model, and obtains a secondary infection risk probability value output by the machine learning prediction model, ranging from 0 to 1. Based on the comparison between the secondary infection risk probability value and a preset cutoff value, it outputs a judgment result indicating the patient's risk of secondary infection. If the secondary infection risk probability value is higher than or equal to the preset cutoff value, it outputs a judgment result indicating the patient is at high risk of secondary infection. The machine learning prediction model is built based on the logistic regression algorithm. The training process of the model includes: taking the set of feature variables of the training set samples as input and the actual secondary infection outcome of the training set samples as output labels, optimizing the model parameters through the maximum likelihood estimation method until the model prediction accuracy meets the preset threshold; in the set of feature variables of the training set samples, each sample corresponds to a set of feature variable data, and each set of feature variable data consists of quantitative detection data of protein content of LYZ, CALM1, SERPIND1, and DPT in plasma samples, as well as the values ​​of clinical biochemical indicators such as AST and Tbil corresponding to the plasma sample.

9. The system according to claim 8, characterized in that, The system also includes: The plasma sample acquisition module is used to acquire plasma samples from patients. A protein quantification detection module, connected to the plasma sample acquisition module, is used to detect the content of LYZ, CALM1, SERPIND1, and DPT in the plasma sample using liquid chromatography-mass spectrometry or immunoassay, and transmits the quantitative detection data to the data acquisition module; the protein quantification detection module includes detection equipment for performing liquid chromatography-mass spectrometry or immunoassay, and a data analysis unit for processing the detection data to obtain protein quantification results; The immunoassay technique includes chemiluminescence immunoassay.

10. The system according to claim 8, characterized in that, The preset cutoff value is 0.5.

Images