A combination of molecular markers for diagnosis and treatment of sepsis-related diseases and use thereof

Abstract

This invention belongs to the field of biomedicine, specifically relating to a combination of molecular markers for the diagnosis and treatment of sepsis-related diseases and their uses. The combination of molecular markers includes... PLA2G4A , FLOT1 , NEU1 , MSH5 , AGER , CCHCR1 , APOM , CLIC1 , HLA-DRA , NOTCH4 , PPT2 , PDIK1L and ABHD16A One or more of these provide new directions for the treatment and management of sepsis.

Description

Technical Field

[0001] This invention belongs to the field of biomedicine, specifically relating to a combination of molecular markers for the diagnosis and treatment of sepsis-related diseases and their uses. Background Technology

[0002] Sepsis is a life-threatening organ dysfunction syndrome caused by infection-induced dysregulation of the host immune response. It is one of the leading causes of death in intensive care units worldwide, placing a heavy burden on healthcare systems. Despite ongoing clinical research, effective treatments for sepsis remain extremely scarce, with most clinical trials ending in failure. Existing diagnostic and treatment methods struggle to achieve early identification, risk stratification, and precise intervention. The core bottleneck lies in the unclear genetic susceptibility mechanisms of sepsis, the unknown causal targets, and the lack of a systematic technological framework from genomic discovery to clinical translation. In the treatment and management of sepsis, early prediction of patient prognosis is crucial for implementing appropriate treatment strategies. In recent years, with the development of biotechnology and information technology, increasing research has shown a close correlation between gene expression levels and disease occurrence and progression. Molecular biomarker combinations based on gene expression data have become important tools for disease prognosis prediction. By analyzing patient gene expression data, key genes related to disease prognosis can be identified, and predictive models can be built to accurately predict patient outcomes. On the other hand, the application of machine learning algorithms has also received widespread attention in the medical field. Composite machine learning algorithms can learn from large amounts of data and discover hidden patterns and regularities, thereby building predictive models and improving the accuracy and reliability of predictions.

[0003] However, existing sepsis-related research has the following technical limitations: (1) Genetic research often relies on single Mendelian randomization analysis, lacks multi-omics cross-validation, and has insufficient credibility of causal genes; (2) The lack of a complete technology chain covering "genetic discovery - functional analysis - clinical translation" makes it difficult to translate basic research results into practical applications; (3) Drug screening relies on traditional methods, which have low throughput, high false positive rate, and insufficient prediction of safety and drugability; (4) Clinical prediction models rely heavily on electronic medical record data, have few molecular markers, high testing costs, poor accessibility at the grassroots level, and poor predictive performance.

[0004] Therefore, constructing a precision diagnosis and treatment model for sepsis patients by combining multi-omics-identified causal gene expression data with composite machine learning algorithms has significant clinical application prospects. This molecular marker-based approach can provide clinicians with more accurate prognostic assessments, early identification, and risk stratification, helping them develop personalized treatment plans and improve patient outcomes and survival rates. Summary of the Invention

[0005] To achieve the above-mentioned technical objectives, the present disclosure proposes the following technical solutions: On one hand, this disclosure relates to a combination of molecular biomarkers for early diagnosis, prognostic prediction, prognostic assessment, and / or treatment selection of sepsis-related diseases, characterized in that it includes... PLA2G4A , FLOT1 , NEU1 , MSH5 , AGER , CCHCR1 , APOM , CLIC1 , HLA-DRA , NOTCH4 , PPT2 , PDIK1L and ABHD16A One or more of them.

[0006] On the other hand, this disclosure relates to the use of the aforementioned combination of molecular markers in the preparation of reagents or kits for the diagnosis and / or screening of sepsis-related diseases; characterized in that the reagents or kits are used for the early diagnosis, prognostic prediction, prognostic assessment and / or treatment selection of sepsis-related diseases.

[0007] In some preferred embodiments, the product is used to detect samples from subjects who are diagnosed with or suspected of having sepsis.

[0008] In some preferred embodiments, the sample is derived from the subject's body fluids, cells, tissues, metabolites, and / or excretions.

[0009] In some preferred embodiments, the sample is derived from the subject's bodily fluids, selected from at least one of blood, plasma, extracellular fluid, tissue fluid, urine, lymph, or cerebrospinal fluid.

[0010] In some preferred embodiments, the reagent or kit contains a substance capable of specifically detecting the aforementioned molecular marker or its active fragment; preferably, it is a primer, probe, antibody, biochip, or small molecule compound.

[0011] In some preferred embodiments, the reagent is directly or indirectly labeled with a detectable tag, which is selected from at least one of the following.

[0012] Radioactive isotopes, fluorescent groups, chemiluminescent components, enzymes, enzyme substrates, enzyme cofactors, enzyme inhibitors, dyes, metal ions or ligands.

[0013] On the other hand, this disclosure relates to a product for diagnosing and / or screening sepsis-related diseases, characterized in that the product comprises a substance that is specific to or capable of specifically recognizing the aforementioned molecular marker composition.

[0014] In some implementations, the substance includes reagents, kits, and / or detection devices.

[0015] In some preferred embodiments, the substance comprises or is selected from at least one of the following: sequencing reagents, immunological detection reagents (e.g., ELISA reagents), molecular biological detection reagents (e.g., qPCR reagents, ddPCR reagents), flow cytometry detection reagents, immunological detection kits (e.g., ELISA kits), molecular biological detection kits (e.g., qPCR kits, ddPCR kits), biochip reagents (e.g., protein chips), and biochip kits.

[0016] In some preferred embodiments, the detection device is adapted to the detection method, such as a detection device for digital PCR, enzyme-linked immunosorbent assay (ELISA), radioimmunoassay (RIA), real-time quantitative reverse transcription PCR, chemiluminescent immunoassay (CLIA), biochip detection, DNA blotting, RNA blotting in situ hybridization, immunofluorescence, immunohistochemistry, or RNA blotting or in situ hybridization.

[0017] In some preferred embodiments, the substance comprises or is selected from at least one of the following: primers, probes, antibodies, biochips, or small molecule compounds.

[0018] In some preferred embodiments, the substance has or is coupled with a detectable marker.

[0019] In some preferred embodiments, the detectable markers include, but are not limited to, radioactive isotopes, fluorescent groups, chemiluminescent components, enzymes, enzyme substrates, enzyme cofactors, enzyme inhibitors, dyes, metal ions, or ligands.

[0020] On the other hand, this disclosure relates to the use of the aforementioned combination of molecular markers in establishing models for early diagnosis, prognostic prediction, prognostic assessment and / or treatment selection for sepsis-related diseases.

[0021] In some implementations, the model is based on immune typing and a combined machine learning algorithm.

[0022] In some preferred embodiments, the composite machine learning algorithm includes, but is not limited to, deep learning.

[0023] In some implementations, the prognostic model is stored in a computer-readable storage medium.

[0024] In some implementations, the computer storage medium stores computer programs.

[0025] In some implementations, the computer storage medium may operate on a computer device.

[0026] On the other hand, this disclosure relates to a system for a combination of molecular biomarkers for predicting, assessing, and / or selecting treatment regimens for sepsis-related diseases, characterized in that the system comprises: (1) A data processing module that receives and / or processes molecular-level data containing combinations of the aforementioned molecular markers in the object; (2) The analysis and calculation module predicts the prognosis of the object based on the data obtained in (1).

[0027] In some preferred embodiments, the system further includes a data acquisition module for acquiring a sample of the object, the sample containing data on the levels of the aforementioned molecular marker compositions; In some preferred embodiments, the analysis and calculation module further includes a module for building a prediction model based on the obtained data; In some preferred embodiments, the step of establishing the prediction model includes: (a) Determination of samples from the subject PLA2G4A , FLOT1 , NEU1 , MSH5 , AGER , CCHCR1 , APOM , CLIC1 , HLA-DRA , NOTCH4 , PPT2 , PDIK1L and / or ABHD16A The levels include the nucleic acid molecule level, preferably the gene level, the mRNA level, the mRNA cDNA level, and / or the protein molecule level; (b) Obtain predictive values ​​for sepsis-related diseases in the subjects; (c) Compare the predicted value with a preset threshold value; when the predicted value of the sample is greater than or equal to the threshold value, the patient is judged to have a high risk of sepsis-related disease; when the predicted value of the sample is less than the threshold value, the patient is judged to have a low risk of sepsis-related disease.

[0028] In some implementations, the system further includes a module for predicting lead compounds based on the obtained data.

[0029] In some preferred embodiments, the lead compound prediction is based on a deep learning algorithm.

[0030] On the other hand, this disclosure relates to a computer-readable storage medium containing a computer program that, when executed, runs any of the aforementioned systems to perform the functions of any of the aforementioned systems.

[0031] On the other hand, this disclosure relates to a computer device comprising a memory and a processor, the memory storing a computer program and the processor executing the computer program to perform the functions of any of the aforementioned systems.

[0032] On the other hand, this disclosure relates to the use of pharmaceutical compositions in the preparation of medicaments for the prevention and / or treatment of sepsis-related diseases, characterized in that the pharmaceutical composition comprises a therapeutically effective amount of an active ingredient capable of regulating the levels of the aforementioned molecular marker group.

[0033] In some preferred embodiments, the active ingredients capable of regulating the levels of the aforementioned molecular marker groups have a synergistic effect when used in combination.

[0034] In some preferred embodiments, the pharmaceutical composition comprises substances capable of regulating NEU1 , CLIC1 and FLOT1 Levels of medication.

[0035] In some preferred embodiments, the pharmaceutical composition comprises NEU1 Inhibitors CLIC1 Inhibitors and FLOT1 Inhibitors.

[0036] In some preferred embodiments, the NEU1 inhibitor, CLIC1 inhibitor, and FLOT1 inhibitor target, respectively. NEU1 Gene, CLIC1 Genes and FLOT1 Gene siRNA, shRNA, or antisense oligonucleotides (ASO).

[0037] In some preferred embodiments, the NEU1 Inhibitors CLIC1 Inhibitors and FLOT1 Inhibitors are targeted NEU1 protein, CLIC1 Protein and FLOT1 Antibodies to proteins or their antigen-binding fragments or small molecule compounds.

[0038] In some embodiments, the pharmaceutical composition further comprises a pharmaceutically acceptable carrier or excipient.

[0039] In some embodiments, the pharmaceutical composition has at least one of the following applications: (a) Inhibits the expression of sepsis-related inflammatory factors; (b) Reduce mortality rates in individuals with sepsis-related diseases; (c) Reduce multi-organ damage in patients with sepsis-related diseases; (d) Improve the survival status of people with sepsis-related diseases.

[0040] The technical solution disclosed herein achieves a systematic breakthrough across the entire technological chain, from genetic discovery to clinical translation, through a three-stage progressive design.

[0041] (1) Discovery phase: A large-scale genome-wide association study (GWAS) meta-analysis was conducted in a European population cohort, successfully identifying sepsis risk sites and their causal variations. By integrating multi-omics evidence from 11 independent gene prioritization pipelines, the accuracy and reliability of causal gene screening were significantly improved, ultimately prioritizing high-confidence sepsis causal genes, laying a solid molecular foundation for subsequent functional analysis and translational applications.

[0042] (2) Translational Application Stage: Breakthroughs in clinical translation were achieved in two major directions. ① Targeted Drug Discovery: Multidimensional evidence was integrated to screen potential druggable targets. Combined with a deep learning-driven virtual drug screening and evaluation system, the throughput and accuracy of lead compound discovery were significantly improved, and the false positive rate and late-stage development risks were reduced. ② Precision Management Model: Based on the expression characteristics of the three most likely causal genes, a general deep learning training process was constructed. The best model for sepsis risk prediction, early diagnosis and prognosis assessment was successfully selected, realizing the efficient clinical application of minimal molecular markers. Attached Figure Description

[0043] Figure 1 The process and results of prioritizing sepsis genes are shown.

[0044] Figure 2 The results of the screening for causal genes of sepsis are shown.

[0045] Figure 3 The results of the sepsis risk prediction task are shown.

[0046] Figure 4 The results of the early sepsis identification task are shown.

[0047] Figure 5 The results of the sepsis clinical deterioration detection task are shown.

[0048] Figure 6 The map of the recombinant lentivirus plasmid constructed in Example 4 is shown.

[0049] Figure 7The results of the validation of shRNA target knockout efficiency in RAW264.7 cells are shown.

[0050] Figure 8 The results of the validation of shRNA target knockout efficiency in RAW264.7 cells are shown.

[0051] Figure 9 The effect of shRNA targets on LPS-induced RAW264.7 cell viability was demonstrated.

[0052] Figure 10 The effect of shRNA targets on LPS-induced secretion levels of inflammatory factors in RAW264.7 was demonstrated.

[0053] Figure 11 The effect of shRNA targets on LPS-induced expression of RAW264.7 inflammatory cytokine mRNA was demonstrated.

[0054] Figure 12 The results of validation of the mouse target gene knockout effect are shown.

[0055] Figure 13 The results of validation of the mouse target gene knockout effect are shown.

[0056] Figure 14 The effects of shRNA targets on the secretion of serum inflammatory factors in septic mice were demonstrated.

[0057] Figure 15 The effect of shRNA targets on the mRNA levels of inflammatory factors in the spleen tissue of septic mice was demonstrated.

[0058] Figure 16 The effect of shRNA targets on the 28-day survival rate of septic mice was shown.

[0059] Figure 17 The effects of shRNA targets on hair condition in septic mice were shown.

[0060] Figure 18 The effect of shRNA targets on climbing ability in septic mice was shown.

[0061] Figure 19 The effect of shRNA targets on the activity level of septic mice was shown.

[0062] Figure 20 The effect of shRNA targets on body temperature in septic mice was shown.

[0063] Figure 21 The effect of shRNA targets on body weight in septic mice was shown.

[0064] Figure 22HE results of tissue sections from some organs of a mouse are shown.

[0065] Figure 23 HE results of tissue sections from some organs of a mouse are shown.

[0066] Figure 24 The results of semi-quantitative scoring of tissue sections from various organs of mice after HE staining are shown. Detailed Implementation

[0067] I. Definition In this disclosure, unless otherwise stated, scientific and technical terms used herein have the meanings commonly understood by those skilled in the art. Furthermore, the terms and laboratory procedures related to protein and nucleic acid chemistry, molecular biology, cell and tissue culture, microbiology, and immunology used herein are all widely used terms and routine procedures in their respective fields. To better understand this disclosure, definitions and explanations of relevant terms are provided below.

[0068] As used herein and unless otherwise stated, the terms “comprising,” “including,” “having,” and “containing,” including their grammatical equivalents, should generally be understood as open-ended and non-restrictive, e.g., not excluding other unlisted elements or steps.

[0069] As used herein, the terms “optional,” “any,” “arbitrary,” or “any one” mean that the event or circumstance described below may, but does not have to, occur, including the circumstances under which the event or circumstance may or may not occur.

[0070] As used in this article, “and / or” should be understood as meaning any one of the options or any combination of two or more of the options.

[0071] As used in this article, the term "sepsis-related diseases" refers to a class of diseases characterized by life-threatening organ dysfunction caused by a dysregulated host response to infection, including, but not limited to, sepsis, septic shock, and severe sepsis. For a specific definition and clinical diagnostic criteria for sepsis, please refer to the relevant descriptions in recognized professional guidelines, authoritative consensus statements, and standard textbooks in this field.

[0072] As used herein, the terms "molecular marker" or "biomarker" refer to biochemical indicators that can mark changes or potential changes in the structure or function of systems, organs, tissues, cells, and subcellular structures, and have a very wide range of applications. Biomarkers can be used for disease diagnosis, disease staging, or to evaluate the safety and efficacy of new drugs or therapies in target populations.

[0073] As used herein, the term "reference value" refers to a standardized value used for comparison with the level (content or activity) of a biomarker in a sample. Reference values ​​can be calculated based on the average level of the biomarker determined in samples isolated from multiple healthy individuals or individuals with sepsis. Typically, reference values ​​are given as a range. Therefore, the terms "reference value" and "range of reference values" as used herein are defined as having the same meaning. Reference values ​​can be negative reference values ​​established from healthy individuals or those with ordinary infection, positive reference values ​​established from patients with sepsis, or a series of positive reference values ​​established from patients with sepsis at different stages of progression. A measured value equivalent to a positive reference value typically indicates the corresponding degree of sepsis progression, immune status, or prognosis.

[0074] As used herein, the term “reduction” can refer to a reduction of 5%, 10%, 20%, 25%, 30%, 40%, 50%, 60%, 70%, 80%, 85%, 90%, 95%, 96%, 97%, 98%, 99% or more in the level of a detected biomarker (e.g., expression or activity) compared to a reference value.

[0075] As used herein, the terms "reagent" and "detection reagent" are used interchangeably and refer to substances that specifically target a target molecule and can be used to directly or indirectly detect the presence and / or content of a target gene, mRNA, cDNA, and protein. Those skilled in the art can prepare reagents specifically targeting target molecules using conventional methods or obtain them commercially available. For example, detection reagents available in this disclosure include, but are not limited to, antibodies (preferably monoclonal antibodies) with detection specificity for HLA-DQA molecules, probes, gene chips, PCR primers, gRNA, etc. Furthermore, for ease of detection, the detection reagents of this disclosure may also carry detectable labels, including but not limited to, radioisotopes, fluorophores, chemiluminescent components, enzymes, enzyme substrates, enzyme cofactors, enzyme inhibitors, dyes, metal ions, ligands (e.g., biotin or haptens), etc. The detection reagents of this disclosure may exist in solution, immobilized on a carrier (e.g., substrate, adsorbent), or in other conventional ways in the art, provided that such mode of existence is suitable for the detection of target molecules in biological samples. For example, when the detection reagent of this disclosure is a nucleotide probe, it can exist in the form of a biochip (or "microarray").

[0076] As used herein, the terms "product" and "detection product" are used interchangeably, referring to substances, combinations of reagents, or devices used in conjunction with or in conjunction with specific target molecules. Depending on the requirements of the detection method used, appropriate target molecule detection substances can be selected and formulated into products suitable for the detection method, such as kits. Those skilled in the art can adjust and modify the detection method and the reagents contained in the product according to actual conditions and needs. The products disclosed herein may also include other reagents clinically used for early diagnosis, risk assessment, immune status assessment, prognostic prediction (or assessment, judgment), and / or treatment selection of sepsis-related diseases in subjects, to assist or verify the results obtained by detecting the target molecule. Those skilled in the art can make routine selections according to specific needs.

[0077] As used herein, the term “level” or “content” refers to the amount (e.g., measured in grams, moles, or counts such as ions or fluorescence counts) or concentration (e.g., absolute or relative concentration) of the target gene, mRNA, cDNA, and / or protein described herein. The term “level” or “content” also includes scaled amounts or values, normalized amounts or values, or scaled and normalized amounts or values. Preferably, the level determined herein is an expression level.

[0078] As used herein, the terms "prognostic prediction" or "prognostic assessment" refer to the prediction of a subject's disease course and / or outcome, or the assessment or monitoring of the effectiveness of therapeutic agents. In some instances, the subject has sepsis-related disease. In some instances, the subject also has other diseases besides sepsis-related disease. In some instances, the subject has previously received treatment for sepsis-related disease.

[0079] As used in this article, the term "inhibitor" refers to a gene that can reduce, neutralize, block, or inhibit a target gene (such as...). NEU1 , CLIC1 , FLOT1Inhibitors are any substances that inhibit the expression of a target gene or the biological activity of a target protein. They may act through a variety of mechanisms, including but not limited to: inhibiting transcription or translation of the target gene, promoting the degradation of the target gene, binding to the target protein and blocking its interaction with ligands or downstream effector molecules, reducing the stability of the target protein, or promoting the degradation of the target protein. Types of inhibitors include, but are not limited to: small molecule compounds, nucleic acid molecules (e.g., interfering RNA (RNAi), short hairpin RNA (shRNA), antisense oligonucleotides, aptamers, ribozymes), polypeptides or proteins (e.g., antibodies or their antigen-binding fragments, receptor traps, soluble receptors), gene editing systems (e.g., CRISPR-Cas systems, zinc finger nucleases, transcription activator-like effector nucleases), and any combination thereof. "Inhibiting the expression of a target gene" means reducing the amount of the target gene product at the gene transcription or translation level, including but not limited to reducing the stability of the target mRNA, promoting the degradation of the target mRNA, preventing the translation of the target mRNA, or altering the transcription rate of the target gene. "Inhibition of the biological activity of the target protein" means that the catalytic activity, binding activity, signal transduction function or other biological function of the target protein is reduced by at least 10%, 20%, 30%, 40%, 50%, 60%, 70%, 80%, 90% or 100% compared with the protein activity in the absence of an inhibitor.

[0080] As used herein, an "antibody fragment" or "antigen-binding fragment" refers to any portion of a full-length antibody that is less than full-length but contains at least a portion of the antibody's variable region (e.g., one or more CDRs and / or one or more antigen-binding sites) that binds to an antigen, and thus retains binding specificity and at least a portion of the full-length antibody's specific binding capacity. Therefore, an antigen-binding fragment refers to an antibody fragment containing an antigen-binding portion that binds to the same antigen as an antibody fragment derived from the antibody fragment. Antibody fragments include antibody derivatives produced by enzymatic treatment of a full-length antibody, as well as synthetically produced derivatives, such as recombinant derivatives. Non-limiting examples of antibodies include: monoclonal antibodies, polyclonal antibodies, multispecific antibodies (including bispecific antibodies), single-chain antibodies, chimeric antibodies, single-domain antibodies / nanobodies, and humanized antibodies. In some embodiments, an antibody may contain the Fc region of a human antibody. The term antibody also includes derivatives thereof, such as bispecific antibodies, single-chain antibodies, dimeric molecules, linear antibodies, and multispecific antibodies formed from antibody fragments. Non-limiting examples of antibody fragments include fragments such as Fab, Fab', F(ab')2, and Fv.

[0081] As used herein, the term "treatment" includes suppressing, slowing, stopping, or reversing the progression or severity of existing symptoms or a patient's condition. Therefore, treatment includes prevention, treatment, and / or cure. Furthermore, "treatment method," "treatment," "relief," or "improvement" may be used interchangeably herein. Prevention refers to preventing potential disease and / or preventing the worsening of symptoms or the development of disease. "Improvement" refers to at least one indicator of the reduction, slowing, cessation, or reversal of symptoms or the severity of a disease. The severity of an indicator can be determined by subjective or objective measures known to those skilled in the art. As used herein, "efficacy" refers to the effect resulting from treatment in an individual, which alters, generally improves, or enhances the symptoms of a disease or disease condition, or cures a disease or disease condition. As used herein, "therapeutic effective amount" or "therapeutic effective dose" refers to the amount of a substance, compound, material, or composition comprising a compound that, when applied to a subject, is at least sufficient to produce a therapeutic effect. II. Detailed Description of Embodiments 1. Biomarkers involved in this disclosure In some embodiments of this disclosure, the biomarker includes PLA2G4A , FLOT1 , NEU1 , MSH5 , AGER , CCHCR1 , APOM , CLIC1 , HLA-DRA , NOTCH4 , PPT2 , PDIK1L and ABHD16A One or more of them; among which, The PLA2G4A (phospholipase A2 group IVA) Gene ID: 5321; Uniprot ID: P47712; The FLOT1 (flotillin 1)Gene ID: 10211; Uniprot ID: O75955; The NEU1 (Neuraminidase 1) Gene ID: 4758; Uniprot ID: Q99519; The MSH5 (mutS homolog 5) Gene ID: 4439; Uniprot ID: O43196; The AGER (advanced glycosylation end-product specific receptor) GeneID: 177; Uniprot ID: Q15109; The CCHCR1 (coiled-coil alpha-helical rod protein 1) Gene ID: 54535; Uniprot ID: Q8TD31; The APOM (apolipoprotein M) Gene ID: 55937; Uniprot ID: O95445; The CLIC1 (chloride intracellular channel 1) Gene ID: 1192; Uniprot ID: O00299; The HLA-DRA (major histocompatibility complex, class II, DR alpha) GeneID: 3122; Uniprot ID: P01903; The NOTCH4 (notch receptor 4) Gene ID: 4855; Uniprot ID: Q99466; The PPT2 (palmitoyl-protein thioesterase 2) Gene ID: 9374; Uniprot ID: Q9UMR5; The PDIK1L (PDLIM1 interacting kinase 1 like)Gene ID: 149420; UniprotID: Q8N165; The ABHD16A (abhydrolase domain containing 16A) Gene ID: 7920; UniprotID: O95870.

[0082] 2. Detection methods for biomarkers In some embodiments of this disclosure, the sample used for biomarker detection is a blood sample or a blood-derived sample, preferably plasma or serum.

[0083] In some embodiments of this disclosure, samples from other sources may also be used, but samples derived from blood are preferred.

[0084] In some embodiments of this disclosure, the level of a biomarker or fragment thereof is determined by measuring gene expression, preferably at the gene level, mRNA level, or mRNA-cDNA level, to determine the level of the biomarker or fragment thereof in a sample.

[0085] In some embodiments of this disclosure, measuring gene expression includes performing a quantitative nucleic acid amplification reaction, such as polymerase chain reaction, preferably real-time PCR or digital droplet PCR, using cDNA corresponding to the mRNA encoding the biomarker for detection. Based on genome annotation information and identified mRNA molecules encoding various forms of the biomarker, those skilled in the art can develop suitable primers and / or probes.

[0086] In some embodiments of this disclosure, the level of a biomarker or a fragment thereof is determined by measuring protein levels.

[0087] In some embodiments of this disclosure, methods selected from the group consisting of: mass spectrometry (MS), luminescent immunoassay (LIA), radioimmunoassay (RIA), chemiluminescence and fluorescence immunoassay, enzyme immunoassay (EIA), enzyme-linked immunosorbent assay (ELISA), luminescent microbead arrays, magnetic bead arrays, protein microarray analysis, rapid detection methods (e.g., immunochromatographic strip tests), and automated systems / analyzers are used to determine the level of a biomarker or fragment thereof.

[0088] In some embodiments of this disclosure, the method for detecting biomarkers is an immunoassay, and the assay is performed in a homogeneous or heterogeneous phase, wherein a sandwich complex formed by the antibody / antibody and the biomarker to be detected remains suspended in the liquid phase. In this case, when two antibodies are used, it is preferable that both antibodies are labeled with components of the detection system, and if both antibodies are integrated into a sandwich complex, a signal or trigger signal will be generated.

[0089] In some embodiments of this disclosure, the method further includes comparing the level of the measured biomarker or a fragment thereof with a reference level, a threshold, a cutoff value, a preset cutoff value, and / or a population mean of biomarker levels corresponding to patients with a systemic inflammatory condition without infection.

[0090] Reference levels corresponding to biomarker levels in patients with systemic inflammatory conditions without infection may vary depending on the specific assay method used. Those skilled in the art can determine such reference levels for any given analytical / diagnostic method. For example, reference levels will depend on the sensitivity and accuracy of the analytical technique used, and will depend on the type of method used; for instance, measuring mRNA encoding a biomarker in a sample will obviously differ from measuring protein levels.

[0091] In some embodiments of this disclosure, the reference level may be a cutoff value or a mean fold change, wherein an infection is indicated when the biomarker level is measured above the cutoff value or a specific fold change. A range of possible cutoff values ​​or fold changes are applicable, taking into account individual patient level variations and differences in analytical techniques.

[0092] In some embodiments of this disclosure, when using, for example, blood samples or blood-derived samples for detection (preferably in blood samples or blood-derived samples), the cutoff value for biomarkers in serum or plasma samples can be a value in the range of 0.001 to 500 ng / mL, or 0.01 to 500 ng / mL. In a preferred embodiment, the cutoff value for the biomarker can be in the range of 0.005 to 400, 0.01 to 300, 0.02 to 350, 0.05 to 300, 0.05 to 200, 0.1 to 100, 0.1 to 50, 0.1 to 20, or 0.2 to 2 ng / mL, or preferably 0.032 to 0.3 ng / mL. Any value within these ranges can be considered a suitable cutoff value. For example, concentrations of 0.005, 0.01, 0.03, 0.04, 0.05, 0.1, 0.2, 0.3, 0.4, 0.5, 0.6, 0.7, 0.8, 0.9, 1.0, 1.1, 1.2, 1.3, 1.4, 1.5, 1.6, 1.7, 1.8, 1.9, 2.0, 3, 4, 5, 6, 7, 8, 9, 10, 15, 20, 25, 30, 35, 40, 45, 50, 55, 60, 65, 70, 75, 80, 85, 90, 95, 100, 120, 140, 160, 180, 200, 250, 300, 350, 400, 450, or 500 ng / mL can be used.

[0093] Such cutoff values ​​are a non-limiting implementation of a biomarker reference level that corresponds to the biomarker level in patients with symptoms of a systemic inflammatory condition but without infection.

[0094] Alternatively, in some implementations, the cutoff value can be used in the following manner / steps: (a) Determination of samples from the subject PLA2G4A , FLOT1 , NEU1 , MSH5 , AGER , CCHCR1 , APOM , CLIC1 , HLA-DRA , NOTCH4 , PPT2 , PDIK1L and / or ABHD16A The levels include the nucleic acid molecule level, preferably the gene level, the mRNA level, the mRNA cDNA level, and / or the protein molecule level; (b) Obtain predictive values ​​for sepsis-related diseases in the subjects; (c) Compare the predicted value with a preset threshold value; when the predicted value of the sample is greater than or equal to the threshold value, the patient is judged to have a high risk of sepsis-related disease; when the predicted value of the sample is less than the threshold value, the patient is judged to have a low risk of sepsis-related disease.

[0095] In practical implementation, the threshold value can be appropriately adjusted by those skilled in the art based on actual testing needs, including but not limited to the sensitivity and specificity requirements of the detection method used, the pathological characteristics of the target population, and the specific needs of the clinical application scenario. For example, the threshold value can be set to 0.5: if the predicted value of the sample is ≥0.5, the patient has a higher risk of sepsis-related disease; if the predicted value of the sample is <0.5, the patient has a lower risk of sepsis-related disease. It should be noted that in practical applications, adjustments can be made according to different diagnostic criteria and requirements, and these adjustments all fall within the scope of protection of this invention.

[0096] The biomarker detection method according to this disclosure can be partially implemented by a computer. For example, the step of comparing the detected biomarker level with a reference level can be performed in a computer system. In the computer system, the measured biomarker level can be combined with other biomarker levels and / or parameters of the subject to calculate a score indicating diagnosis, prognosis, risk assessment, and / or risk stratification. The computer system stores the values ​​on a computer-readable medium and calculates the score based on predefined and / or pre-stored reference levels or reference values. The generated score is displayed and / or printed to the user.

[0097] In some embodiments of this disclosure, a software system, preferably employing machine learning algorithms, can be used to identify patients at risk of sepsis using electronic health record data. Machine learning methods can be trained using electronic health record data from patients. Processing at the biomarker level can be integrated into appropriate software to compare against existing datasets to aid in diagnosis or infection prediction.

[0098] In some embodiments of this disclosure, the method further includes treating the subject based on the results of the method: for infectious diseases, the treatment includes administering an appropriate anti-infective agent; or for non-infectious systemic inflammatory conditions, no anti-infective agent is administered.

[0099] In some embodiments of this disclosure, one or more other biomarkers may be used in conjunction with the biomarker detection method according to this disclosure to potentially improve or confirm the diagnostic conclusions of the method. Disharmony or altered levels of other biomarkers relative to appropriate controls provide further indications of the presence of infection.

[0100] Example Example 1: Identification and Prioritization of Causal Genes in Sepsis 1. GWAS meta-analysis and genetic association mapping 1.1 GWAS Meta-analysis This study included pooled statistics from four sepsis GWAS studies from the UK Biobank, the US Million Veterans Project, and FinnGen, comprising 46,029 cases and 1,315,442 controls of European descent. Quality control of the pooled GWAS statistics was performed using GWASInspector with default parameters. Fixed-effects inverse variance weighted meta-analysis was performed in METAL software (corrected for sample overlap, with a Z-statistic threshold of 1). P ≤ 1.0 × 10⁻⁶ was used as the cutoff value. -5 The threshold for genome-wide nominal significance is P ≤ 5.0 × 10⁻⁶. -8 This is the standard genome-wide significance threshold.

[0101] 1.2 Functional localization and annotation of genetic associations Functional localization and annotation were performed using the FUMA platform. The threshold for cascading disequilibrium (LD) was used. r 2 <0.1% lead variants were identified from conventional genome-wide significant variants. Risk site identification parameters included: ① 1000G phase III European population reference panel; ② LD non-GWAS tag variants included in the reference panel; ③ minimum allele frequency (MAF) ≥ 0; ④ maximum LD block merge distance of 250kb.

[0102] 1.3 Identification of Independent Variations Initially, variants within 500 kb upstream and downstream of the lead variant were included. Stepwise model selection (cojo-slct) was performed in GCTA software using default parameters (MAF = 0.01, significance threshold = 5.0 × 10⁻⁶). -8 Identify independent associated variants.

[0103] 1.4 Identification of Causal Variation Variations within ±500 kb of the independent leader variant were included, and paired LDs were calculated using PLINK software and a 1000 G European LD reference panel. r 2 ), construct the LD matrix. Use FINEMAP software (shotgun method) to identify causal variations, expressed as logarithmic values. 10(bayesfactor)>2 is used as the criterion for determining causal variants. FM-summary (a simplified Bayesian fine-grained localization method with a flat prior and steepest descent approximation) is further employed to identify a 99% reliable set of causal variants (cumulative posterior probability ≥75%) in each region (independent leading variant ±500kb, assuming at least one causal variant).

[0104] 2. Multi-omics level gene prioritization 2.1 Genome-level gene priority ranking Two complementary methods were used to conduct variant-gene level association analysis: ① fastBAT (LD-pruned gene level test in GCTA software); ② MAGMA (two-stage parametric framework for gene and gene set analysis). After multiple test correction... P <0.05 was considered statistically significant. A 1000G European LD reference panel was used to control for LD. Furthermore, a nearest-neighbor gene strategy was employed for causal gene prioritization. Using the GRCh37 reference genome and the vautils R package, the nearest-end protein-coding genes within a 50 kb window for each sepsis causal variant were identified.

[0105] 2.2 Transcriptome-level gene priority ranking TWAS Analysis: A transcriptome-wide association study (TWAS) was conducted using FUSION software and a 1000G European LD reference panel to summarize the meta-analysis statistics of sepsis GWAS. Whole blood gene expression weights were obtained from GTEx v8 and the Young Finns study, peripheral blood gene expression weights were obtained from the Dutch twin registry, and cross-tissue gene expression weights were generated using GTEx v8 sparse canonical correlation analysis (sCCA). FUSION conditional analysis was used to identify conditionally independent signals with an FDR < 0.05 (Benjamin-Hohberg correction); significant conditionally independent signals were considered causal signals for sepsis.

[0106] Single-organization and cross-organization TWAS: Single-organization TWAS was conducted using S-PrediXcan (integrating QTL information and GWAS summary statistics); S-MultiXcan (utilizing cross-organization QTL shared features and merging single-organization S-PrediXcan results based on a multivariate regression meta-analysis framework) was used to improve gene detection efficiency in cross-organization analysis. Gene expression was predicted using the MASHR model (integrating fine-mapped variants from all tissues and cross-organization patterns in GTEx v8). Complete harmonization and genotyping were performed before S-PrediXcan and S-MultiXcan analyses.

[0107] Cross-organizational TWAS supplement: UTMOST (a multi-task learning method for jointly estimating gene expression in human tissues and conducting cross-organizational TWAS) was used, and the covariance matrix was pre-calculated using single-organization and joint tests for 44 organs and tissues in GTEx v8. After multiple test correction, FDR < 0.05 (Benjamin-Hohberg correction) was considered statistically significant for the gene-sepsis association.

[0108] 2.3 Fine localization of statistical measures in transcriptome association studies Fine mapping of TWAS probabilistics was performed on the GWAS meta-analysis statistics using FOCUS (Fine Mapping of Causal Gene Sets) in MA-FOCUS software. The eQTL weighted database was constructed using the GTEx v8 multi-tissue expression panel from the FUSION database, and the LD reference panel was the 1000G European LD reference panel. A posterior inclusion probability (PIP) > 90% was used as the criterion for determining causal genes in sepsis.

[0109] 2.4 Epigenome-level gene priority ranking DNA methylation levels: Analyzed using SMR (integrated xQTL data and GWAS summary statistics). Peripheral blood cis-methylation quantitative trait loci ( cis -mQTLs) are from published studies, and blood cis-sQTLs are from GTEx v8. Only included P <1.0×10 -5 SMR analysis was performed on the QTLs. The HEIDI test was used to assess whether the association of the target trait was due to linkage or pleiotropy. The criteria were: FDR_SMR < 0.05 (Benjamin-Hohberg correction) and P_HEIDI > 0.05, indicating significant association and no heterogeneity. Gene annotation was performed on each CpG probe using annotation data from the Illumina Infinium HumanMethylation450 platform.

[0110] RNA splicing level: As a supplement to cis-sQTLs SMR, S-PrediXcan and S-MultiXcan were used to predict RNA splicing level based on finely mapped sQTLs using the GTEx v8 MASHR model to identify gene-sepsis association. After multiple validation correction, FDR < 0.05 (Benjamin-Hohberg correction) was considered statistically significant.

[0111] 2.5 Gene Prioritization at the Proteomic Level Included are three pooled statistics from blood proteomics GWAS: UK Biobank Pharmaceutical Proteomics Project (UKB-PPP, 54,219 participants, 2,941 plasma proteins). cis-pQTLs), deCODE study (35,559 Icelandic participants, 4,907 plasma proteins) cis -pQTLs), ARIC study (7,217 Caucasian adults aged 45-64, 4,657 plasma proteins) cis -pQTLs). Using PLINK software for... cis -pQTLs are used for clustering (window 5,000kb, maximum LDR). 2 =0.01, 1000G European LD reference panel), LD-related variants were removed, and included P <5.0×10 -8 And MAF > 0.01 cis -pQTLs is used as an instrumental variable.

[0112] The two-sample Mendelian randomization (TSMR) method was used to... cis -pQTLs represent exposure, and GWAS meta-analysis statistics represent outcomes in a proteome-wide association study (PWAS): for ≥2... cis Plasma proteins of -pQTLs were analyzed using the inverse variance weighted (IVW) method, with only 1... cis -pQTLs were analyzed using the Wald ratio method. After multiple test correction, FDR < 0.05 (Benjamin-Hohberg correction) was considered significant. As a supplement, the FUSION combined with INTERVAL study (3,301 participants of European descent, 3,622 plasma proteins) was used. cis -pQTLs were used to perform PWAS analysis to identify conditionally independent signals with an FDR < 0.05 (Benjamin-Hohberg correction). Conditionally independent and significant plasma proteins were considered as causal proteins of sepsis.

[0113] 2.6 Similarity-based gene prioritization (PoPS) PoPS is a similarity-based gene prioritization method that integrates gene association statistics, cell type-specific gene expression, pathway and protein-protein interactions (PPIs) within a ridge regression framework to predict polygenic genetic associations. MAGMA is used to calculate gene-level association statistics, and a generalized least squares (GLS) model is applied to the MAGMA results for marginal feature selection, retaining features with p < 0.05 for PoPS calculation. A multi-gene enrichment joint model is fitted to fit all gene features, and the PoPS score for each gene is calculated. A PoPS z > 1 is used as the criterion for identifying potential causal genes.

[0114] 3. Identification and validation of druggable targets Five criteria were used to screen for druggable targets from the most likely causal genes: ① expression of immune organ proteins at intermediate or high levels; ② gene expression causally regulated by sepsis and significantly differentially expressed in sepsis; ③ association with sepsis-related phenotypes in gene knockout (KO) mice; ④ existence of at least one intervention targeting the gene; and ⑤ drug development / drugability level (TDL) higher than Tdark.

[0115] revTWMR analysis: RevTWMR (Mendelian randomization with multiple instrumental variables) was used, with the sepsis GWAS meta-analysis statistic as the exposure and significant trans-eQTLs (FDR < 0.05) from the eQTLGen consortium (31,684 whole blood samples) as instrumental variables. r was selected from the 10K pre-screened trait-associated trans-eQTLs. 2 <0.01 and P<5.0×10 -8 The trans-eQTLs were identified. After multiple test correction, an FDR < 0.05 (Benjamin-Hohberg correction) indicated that gene expression was causally regulated by sepsis. The Cochrane Q test was used to assess heterogeneity, with P > 0.05 indicating no pleiotropic effect / heterogeneity.

[0116] Differential expression analysis: Nine high-throughput batch RNA sequencing datasets from the Gene Expression Comprehensive Database (GEO) were included, involving 664 participants (547 sepsis patients and 117 healthy controls). Batch effects were corrected using ComBat (empirical Bayesian framework) in the sva R package. No missing values ​​were found in the gene expression matrix, eliminating the need for imputation. The Tuki test was used to detect extreme outliers (lower bound = Q1 - 1.5 × (Q3 - Q1), upper bound = Q3 + 1.5 × (Q3 - Q1)), and outliers were replaced with the median of the gene. Expression intensity was normalized using the normalizeBetweenArrays function in the limmaR package, with log2 transformation performed if any of the following conditions were met: ① 99th percentile > 100; ② Q1 in the range of 0-1, Q3 in the range of 1-2; ③ the difference between the maximum and minimum values ​​> 50 and Q1 > 0. The Wilcoxon test was used to analyze differential expression between sepsis patients and healthy controls, with P < 0.05 indicating differentially expressed genes.

[0117] TDL Level Query: Use the Pharos tool to query the TDL level of three candidate druggable targets. A TDL higher than Tdark (Tbio, Tchem, Tclin) indicates that the target has the potential to become a druggable target.

[0118] Mouse phenotype association: Evidence of association between three candidate druggable targets and sepsis-related phenotypes was collected using the Mouse Genome Information (MGI) database, and systematic abnormalities associated with all query gene knockout models were analyzed. Phenotypes related to sepsis complications, coagulation abnormalities, immune system dysfunction, and metabolic dysfunction were considered sepsis-related phenotypes.

[0119] Additional evidence of druggability: Additional druggability evidence was collected through DrugnomeAI regarding immune organ-specific expression, cellular localization, and target manageability.

[0120] 4. Results of Sepsis Target Screening At the genomic level, 8 significant gene-sepsis associations were identified using MAGMA, and 51 significant gene-sepsis associations were identified using fastBAT (Benjamini-Hochberg correction, FDR < 0.05). PRRC2A , RNF5 and AGER Both gene assays showed a significant association with sepsis. Furthermore, 31 unique genes most closely associated with the pathogenic variant were identified. In summary, 76 unique gene-sepsis associations were identified at the genomic level.

[0121] At the transcriptome level, single-tissue TWAS (FUSION) identified 17 conditionally independent genes (FDR_TWAS<0.05, P_joint<0.05), and S-PrediXcan identified 12 gene-sepsis associations (representing 26 unique genes, FDR<0.05). Regarding cross-tissue TWAS, FUSION based on the GTEx v8 cross-tissue (sCCA) expression panel identified 21 gene-sepsis associations, and S-MultiXcan based on the GTEx v8 MASHR model identified 2 gene-sepsis associations (…). C4A FDR=0.0065 CLIC1 FDR=0.0096). A total of 38 genes were identified as having a causal association with sepsis at the transcriptome level. Fine mapping of the entire transcriptome using FOCUS identified 86 tissue-specific gene-sepsis associations (representing 44 unique most probable pathogenic genes, PIP≥90%). FLOT1 , C4A , PDIK1L , ABHD16A and HLA-DRA It is also listed as a priority pathogenic gene by TWAS.

[0122] At the proteomic level, the TSMR method identified 11 unique plasma proteins causally associated with sepsis (FDR < 0.05). The FUSION study, combined with PWAS from the INTERVAL study, identified... CDSN , HYAL1 and AGER Genes with causal association (FDR < 0.05 and P_joint < 0.05). At the proteomic level, 14 plasma proteins were identified as sepsis-causing genes, among which... CLIC1 It shows a significant causal relationship.

[0123] At the epigenomic level, DNA methylation EWAS (SMR & HEIDI) identified 86 significant CpG probe-sepsis associations (FDR_SMR < 0.05, P_HEIDI > 0.01), representing 31 unique genes, among which... CLIC1 (PWAS, TWAS, GWAS preferred) and NEU1 (TWAS and GWAS are preferred) are pathogenic genes. RNA splicing-level integration of trans-tissue cis-alternative splicing QTLs identified 28 intronic regions located at 23 unique gene loci significantly associated with sepsis risk, and validated these findings. NEU1 and FLOT1 The causal relationship.

[0124] PoPS analysis identified 77 unique sepsis-causing genes (PoPS z-score > 1), among which... PLA2G4A (Encoding cytoplasmic phospholipase A2, cPLA2) has the function of promoting the adaptive immune response of bacteria to sense infection.

[0125] By integrating gene prioritization results from GWAS, TWAS, TWAS fine mapping, PWAS, EWAS_mQTL (DNA methylation), EWAS_sQTL (RNA splicing), and PoPS, and through cross-validation using at least three independent procedures, 13 potential sepsis causal genes were identified, including... PLA2G4A , FLOT1 , NEU1 , MSH5 , AGER , CCHCR1 , APOM , CLIC1 , HLA-DRA , NOTCH4 , PPT2 , PDIK1L and ABHD16A, in NEU1 , CLIC1 and FLOT1 The most likely causal gene for sepsis ( Figure 1-2 ).

[0126] Example 2: Virtual Screening of Targeted Drugs and Identification of Lead Compounds Targeting 3 druggable targets ( NEU1 , CLIC1 , FLOT1 Large-scale virtual drug screening based on deep learning was conducted. Three compound libraries from the ZINC20 database were used: ① a global drug library (5,895 approved drugs, including those approved by the FDA and other major jurisdictions); ② a clinical trial library (9,793 investigational compounds under clinical evaluation); and ③ a natural product library (247,696 secondary metabolites). Protein sequences for the three druggable targets were obtained from AlphaFold.

[0127] CPI Prediction: First, the compound-protein interaction (CPI) of each candidate compound was predicted using TransformerCPI2.0 (a sequence-to-drug concept-based virtual drug screening model), and compounds with a prediction score > 0.9 were selected. Then, GraphBAN (a graph-based CPI prediction framework) was used for validation, employing GraphBAN models trained on three CPI datasets: BindingDB, BioSNAP, and KIBA. Selection Criteria: Compounds with a TransformerCPI2.0 prediction CPI score > 0.9 and at least two GraphBAN models predicting CPI scores > 0.5 were considered candidate lead compounds.

[0128] Off-target and toxicity prediction: An off-target spectrum prediction pipeline based on a multi-task graph neural network was employed. This pipeline constructed off-target prediction models for seven ligand groups (corresponding to GPCRs, ion channels, enzymes, kinases, transporters, nuclear receptors, and seven other target families) to predict the off-target spectrum of each candidate lead compound. The off-target spectrum was used as a molecular characterization for toxicity prediction classification and ADR enrichment analysis. Toxicity prediction employed a LightGBM classifier trained on a drug toxicity dataset. ADR enrichment analysis used a hypergeometric distribution to estimate the probability of predicted off-target effects being included in the off-target sets of four studies. ADRs were classified as non-toxic and without significant intolerable ADRs (coma, acute respiratory distress syndrome; corrected). P Compounds with a concentration <0.05 were considered non-toxic. As a supplement, the toxicity of each candidate lead compound was predicted using the NYAN (Graphical Variational Encoder model), trained on the Tox2110K chemical library (containing 8,599 unique chemicals). Compounds predicted as non-toxic by both off-target spectroscopy and NYAN were considered putative non-toxic lead compounds.

[0129] ADMET Screening: Deep-PK (a graph neural network-based drug discovery screening platform trained on a small molecule dataset with 73 endpoints) was used to predict the ADMET characteristics of prospectively non-toxic lead compounds. Judgment criteria: predicted toxicity probability <0.5 (focusing on cardiotoxicity, liver injury, and toxicity-related pathways / receptors / genes), and inhibition probability <0.5 for more than 3 major CYP450 subtypes. Subsequently, ADMET-AI (based on the Chemprop-RDKit deep learning framework) was used to calculate the ADMET predicted percentile for each compound compared to 2,579 approved drugs in DrugBank v5.1.10. Final screening criteria: ① Toxicity: acute toxicity (LD50), cardiotoxicity, and liver injury percentile <80%; ② Distribution: plasma protein binding percentile <80%; ③ Excretion: drug clearance percentile >50%. Compounds meeting all the above criteria were identified as lead compounds for further development as precision targeted therapies for sepsis.

[0130] Based on the above criteria, potential drug targets were screened from the most likely pathogenic genes. RevTWMR analysis showed a positive correlation between sepsis and gene expression of the three candidate drug targets. NEU1 α=0.08, FDR<0.0001, heterogeneity=0.17; CLIC1 α = 0.40, FDR < 0.0001, heterogeneity = 0.12; FLOT1 (α=0.41, FDR<0.0001, heterogeneity=0.15). In the largest current sepsis high-throughput sequencing dataset (547 cases, 117 controls), the expression of all three candidate genes was significantly upregulated (Wilkerson test, P<0.05). Pharos tool showed that the TDL levels of the three targets were as follows: CLIC1 and FLOT1 For Tbio, NEU1 The Tchem. MGI database shows that mutations / alleles in three genes are associated with sepsis-related phenotypes: NEU1 Associated with acute respiratory distress syndrome CLIC1 Related to abnormal platelet regulation, FLOT1 It is associated with neutrophil dysfunction. DrugnomeAI evidence suggests that each candidate target has at least one experimental approach (antibody or small molecule) and is expressed at a moderate to high level in immune organs, among which... NEU1 It showed a high probability of being targeted by drugs in all four general models. Based on the above evidence, it was determined that... NEU1 , CLIC1 , FLOT1 It is a potential drug target for sepsis.

[0131] Large-scale deep learning-based virtual drug screening was conducted for these three targets, including 9,795 drugs (derived from a database after standardization and deduplication to form a final non-redundant compound set) and 247,696 natural products. TransformerCPI2.0 identified 1,515 drugs and 32,384 natural products with prediction scores >0.9. GraphBAN validation yielded 169 complex-protein pairs with CPI probabilities >0.5 (predicted by at least two GraphBAN models). Off-target safety checks were performed on 40 compounds. ADR enrichment analysis excluded one compound. NEU1 Targeted compound (significant ADR: acute kidney injury) and 2 types CLIC1 Targeted compounds (significant ADR: coma). The remaining 37 compounds showed no off-target effects, no intolerance-related ADRs, and were predicted to be non-toxic by NYAN. Deep-PK prediction showed that 18 compounds had high safety (no cardiotoxicity, no liver damage, no significant impact on toxicity-related pathways, and <50% inhibition probability of more than 3 major CYP450 subtypes). After ADMET-AI comparative analysis, 6 lead compounds were finally identified ( FLOT1 Target 1 type NEU1 Targeting 2 types CLIC1 (Targeting 3 types), these compounds meet the following criteria: acute toxicity, cardiotoxicity, and hepatic injury percentile <80%; plasma protein binding percentile <80%; and drug clearance percentile >50%.

[0132] Example 3: Development of a Deep Learning-Based Precision Management Model for Sepsis 1. Dataset and Samples Postoperative sepsis prediction model: This study included 267 prospective cohort participants (77 cases and 190 controls) recruited from 8 hospitals in Germany and the UK between November 2007 and February 2017. Those diagnosed with postoperative sepsis were classified as the case group; the control group included those with postoperative infection but no sepsis (49 cases), those with a single postoperative course without complications (110 cases), and those with non-infectious systemic inflammatory response (31 cases).

[0133] Early sepsis identification model: Subjects from 9 studies were included. Those without data on the expression of the 3 most likely causal genes or with uncertain diagnosis were excluded. A total of 547 sepsis patients were included in the case group, and 117 healthy participants were included in the control group.

[0134] Prognostic prediction model: After excluding those with undetected survival outcomes and gene expression, 402 participants from the Canadian COVID-19 Lung Injury Longitudinal Biomarker (COLOBILI) study and multinational studies were included (63 cases and 339 controls). Most studies defined sepsis and septic shock according to the third international consensus definition (Sepsis-3), defining sepsis as a new-onset organ dysfunction diagnosed with infection plus a Sequential Organ Failure Assessment (SOFA) score ≥2 points higher than baseline. The number of outcome events in each dataset met the minimum sample size requirement for developing a binary clinical prediction model (based on the number of events per variable, the model was developed using 5 optimal informative features, and the sample size requirement = EPV × number of features = 10 × 5 = 50). ComBat was used to correct for batch effects in each high-throughput batch RNA sequencing dataset.

[0135] 2. Feature Engineering First, graphical tests were used to identify extreme outliers, which were then clipped to the interval [Q1-k(Q3-Q1), Q3+k(Q3-Q1)] (k=3), i.e., Winsserification. Based on the original expression characteristics of the three most likely causal genes, eight new features were generated (three gene expression ratios, two statistical moments, and three log-transformed original expression values), and features with variance <1.0×10⁻⁶ were removed. -4 The low-information characteristics are addressed. Stratified sampling is used to partition the dataset into training, validation, and test sets at a ratio of 70%:15%:15%, preserving the distribution of outcome events. The training set is robustly standardized using scalar normalization to minimize the impact of residual outliers.

[0136] Calculation of gene-to-gene ratio characteristics for genes i and j: Prevent division by zero.

[0137] Calculation of mean expression and standard deviation of all genes: Logarithmic transformation of each gene: 3. Feature combination selection For each candidate model, the most informative panel is selected from all possible feature combinations, which include 3 features randomly selected from 11 features. 165 panel-model pairs were tested to evaluate the classification performance of each panel-model pair on the test set (using the Area Under the Receiver Operating Characteristic Curve (AUROC)). The optimal information panel for each model was selected first. A Bayesian hyperparameter optimization framework (tree-structured Parsons estimation algorithm from Optuna, 100 trials per panel-model pair) was used to optimize the model hyperparameters. The panel with the highest classification performance for each model was selected as the optimal information panel and input into subsequent model training and hyperparameter optimization.

[0138] 4. Data Augmentation and Quality Assessment Borderline-SMOTE was used to augment the training set data and address class imbalance. The Kolmogorov-Smirnov test was used to compare the feature distributions of the original and synthetic data to evaluate the quality of the synthetic samples. The quality score Q was calculated according to formula (5), and Q≥0.8 was the threshold for retaining synthetic data.

[0139] D_i is the KS statistic for feature i, d is the feature dimension, and r_orig and r_synth are the proportions of positive categories in the original and synthetic data, respectively.

[0140] 5. Deep Learning Model Architecture The task is based on three features generated from three most likely causal genes of sepsis in a small sample dataset. Therefore, four candidate deep learning models (Transformer, TabNet, contrastive self-supervised learning model, and a CNN-LSTM-GRU hybrid model based on attention mechanism) are used instead of conventional machine learning models to avoid the risk of overfitting. In the Transformer architecture, features are treated as sequences (single tokens). They are embedded into a high-dimensional space by learning linear projection, followed by layer normalization and dropout regularization. Sine positional encoding is used to enable the model to learn relative positional information. The embedded sequences are processed through the Transformer encoder layer (multi-head self-attention layer, feedforward network, GELU activation, compression-expansion mode, residual connections, and pre-normalization). Attention pooling is applied after the encoder layer, and the importance weight of each position is calculated through a q-parameterized attention mechanism, allowing the model to dynamically weight features and their relevance to the classification task. The classification head uses a three-layer multilayer perceptron for progressive dimensionality reduction, intermediate normalization and activation, and sigmoid transformation to calculate the predicted probability.

[0141] In TabNet, batch normalization is used to standardize the feature scale, and a projection layer (two-layer feedforward network, GELU activation, batch normalization) expands the feature dimension. Each decision step t∈{1,…,n_steps} includes attention-based feature selection, feature processing, and decision aggregation. Each feature converter uses a shared fully connected layer (batch normalization, GELU activation, discarding) to generate attention features and decision features. Feature selection is achieved through attention masks; the higher the mask value, the higher the importance of the feature in the current decision step. The attention mask M^(t) for step t is calculated according to formula (6): f_att is the attention transformation network, ⨀ is element-wise multiplication, a^(t-1) is the aggregation representation from the previous step, and P^(t-1) is the prior scale factor. The prior scale is initialized to P^(0)=1 and updated according to formula (7): Here, γ represents the relaxation parameter, used to control the degree of feature reuse in the decision-making step. A classification network with multiple fully connected layers, batch normalization, and GELU activation is applied to map the aggregated decision representations to predicted probabilities.

[0142] The CNN-LSTM-GRU model integrates multi-scale convolutional feature extraction, bidirectional recursive temporal modeling, and attention-based aggregation. Parallel convolutional blocks with different kernel sizes are used to extract multi-scale local patterns. Each convolutional block consists of two one-dimensional convolutional layers with progressive channel expansion, batch normalization, GELU activation, and loss regularization. To integrate complementary information from different receptive fields, 1×1 convolutional layers are used for feature fusion, integrating multi-scale representations into a unified latent dimension. Bidirectional LSTM and GRU layers are used to model temporal dependencies, and their outputs are concatenated to form a relatively comprehensive temporal representation. Exit between recurrent layers is applied to reduce the risk of overfitting. Multi-head self-attention layers with learned positional encoding are employed to estimate feature importance. Global average pooling is used to aggregate temporal information. Classification probabilities are predicted by a multi-layer network with progressive dimensionality reduction.

[0143] The contrastive self-supervised learning model comprises two modules: contrastive pre-training for learning representations and subsequent supervised fine-tuning for classification tasks. The backbone network consists of four fully connected layers with asymptotic dimensionality, batch normalization, GELU activation, and exit regularization. During pre-training, two augmented views are generated for each sample by adding Gaussian noise (σ=0.01). The embedding representation of each augmented view is used to generate an encoder using the backbone, and a projection head maps the representation to a low-dimensional embedding space. Subsequently, a prediction head is applied to transform one embedding to predict the representation of the other. A SimSiam-inspired contrastive loss leverages a stopping gradient operation to employ negative cosine similarity on the target representation. Where p_i represents the prediction from view I, and z_j represents the separation projection from view j. During the fine-tuning phase, the projection and prediction heads are discarded, and a supervised classification head consisting of three layers with asymptotic dimensionality, batch normalization, GELU activation, and loss regularization is appended to the fine-tuning backbone. The learning rate in this phase is scaled by 0.1 to preserve the learned representations.

[0144] 6. Model Training and Evaluation During training, a hybrid loss function (combining focus loss and binary cross-entropy) was employed, and label smoothing was applied to improve generalization ability. Bayesian hyperparameter optimization was performed using TPE sampling implemented in Optuna, with the optimization objective being the validation set AUROC. Optimizers included AdamW or Adam, and learning rate schedulers included cosine annealing with hot restart, a single-cycle strategy with dynamic looping, or a platform-reduction-based scheduler. The best model for each task (the one with the highest AUROC in the test set) was selected from models optimized over 100 rounds of hyperparameter optimization. The classification cutoff was calculated using the Youden index. The classification performance of each model in the test set was evaluated using AUROC, Area Under the Precision-Recall Curve (AUPRC), accuracy, precision, recall, F1 score, specificity, and sensitivity. The 95% confidence intervals for the classification metrics were estimated using 1,000 Bootstrap simulations.

[0145] For the sepsis risk prediction task, based on FLOT1 Gene expression, NEU1 Logarithmic transformation of expression and CLIC1 The panel-trained Transformer model with the logarithmic transformation of the expression showed the best classification performance compared to the other three candidate models (AUROC=0.85; number of cases=12, number of controls=29). Figure 4 In the early identification task of sepsis, based on CLIC1 Express, CLIC1 : FLOT1The TabNet model trained with the combination of gene ratios and three-gene standard deviations showed the best classification performance among the four candidate models (AUROC=0.90; number of cases=82, number of controls=18). Figure 5 Among the four candidate models, the attention-based CNN-RNN-GRU model has the following training features: NEU1 : FLOT1 Gene ratio, CLIC1 : FLOT1 Gene ratio and CLIC1 The logarithmic transformation values ​​of the expression showed the best classification performance in the clinical deterioration detection task (AUROC=0.86; number of cases=10, number of controls=51); Figure 6 In short, these three most likely pathogenic genes show high potential to form a versatile combination suitable for precision sepsis management. This combination is widely available because it can likely be detected using low-cost detection methods such as real-time PCR.

[0146] Example 4: NEU1 / CLIC1 / FLOT1 The therapeutic effect of gene knockout on septic mice 1. Construction of lentiviral vectors For mice NEU1 , CLIC1 and FLOT1 For each gene, a specific target sequence was designed. The target sequence and primer sequences are as follows: Table 1. Target sequence and primer sequence The above target sequences were cloned into the pLKO.1-TRC-puro vector (element sequence: U6 promoter-shRNA-hPGK-puro; the recombination plasmid map is shown below). Figure 7 As shown in the figure, construct a recombinant lentiviral plasmid.

[0147] 2. Cell experiments 2.1 Lentiviral Packaging and Infection (1) Paving Take healthy RAW264.7 cells, gently scrape them off with a cell scraper, count them, and seed them into 24-well plates at 1×10⁶ cells per well. 5 Add 500 μL of complete culture medium to each cell and incubate overnight at 37°C in a 5% CO2 incubator until the cell confluence reaches 30%–50% the next day.

[0148] (2) Viral infection Lentiviral titer is 1×10 10Add 240 μL of fresh complete culture medium, 10 μL of diluted virus solution (MOI=50), and 2.5 μL of 1×HiTransGP working solution to each well at TU / mL. Incubate for 72 hours without changing the medium.

[0149] (3) Screening and maintenance Reporter gene expression was observed under a fluorescence microscope 72 hours after infection. The culture medium was then replaced with fresh complete medium containing 3 μg / mL puromycin for selection, with the medium changed every 2-3 days until all uninfected control cells died. After selection, the puromycin concentration was halved to 1.5 μg / mL for maintenance culture. Subsequent passages were always performed using a cell scraper, without trypsin digestion.

[0150] 2.2 Low-efficiency verification Cells from each group were collected, total RNA was extracted using the Trizol method and reverse transcribed into cDNA, and the cDNA was detected by SYBR Green qRT-PCR. NEU1 , CLIC1 , FLOT1 mRNA expression level (2^-ΔΔCt method); total protein was extracted by RIPA lysis, and target protein expression was detected by Western blot to verify knockout efficiency.

[0151] 2.3 Cell viability assay Stable knockdown cells were seeded into 96-well plates (1×10⁶ cells). 4 / well), after stimulation with LPS (100ng / mL) for 24h, add 10μL of CCK-8 reagent to each well, incubate at 37℃ for 2h, and measure OD value at 450nm.

[0152] 2.4 Detection of inflammatory factors The secretion levels of TNF-α and IL-6 in cell supernatant were detected by ELISA; total RNA was extracted from cells using the Trizol method, and the expression levels of TNF-α and IL-6 mRNA were detected by qRT-PCR.

[0153] 3. Effects of shRNA targets on septic mice 3.1 Animal grouping and model establishment SPF-grade male BALB / c mice, 6–8 weeks old and weighing 20–25 g, were selected. After arrival, the mice were placed in an SPF-grade animal facility for 7 days of acclimatization. The conditions were: temperature 22–25℃, humidity 40%–60%, 12-hour light / dark cycle, and free access to food and water. A stratified block randomization method was used, with mice randomly assigned proportionally to 10 groups: sham group, sepsis model group, negative control shNC, single knockout groups (shNEU1, shCLIC1, shFLOT1), double knockout groups (shNEU1+shCLIC1, shNEU1+shFLOT1, shCLIC1+shFLOT1), and triple gene knockout group (shNEU1+shCLIC1+shFLOT1), with 15 mice in each group. A sepsis model was established using the cecal ligation and puncture (CLP) method: After anesthesia, an incision was made along the midline of the abdomen, and the cecum was ligated at the distal half of the ileocecal valve. A 21G needle was used to puncture the cecum and expel a small amount of intestinal contents. The cecum was then returned to its original position and sutured layer by layer. In the sham surgery group, only the cecum was exposed and returned to its original position after an incision. Postoperatively, preheated saline (1 mL / cattle) was injected subcutaneously for fluid resuscitation.

[0154] 3.2 Lentiviral Transfection Protocol Using lentivirus-mediated short hairpin RNA (shRNA) interference technology, targeted mice were used. NEU1 , CLIC1 , FLOT1 The shRNA sequence of the gene and the negative control sh-NC. The titer of each lentivirus was 1×10⁻⁶. 10 TU / mL, aliquot and store at -80°C, avoiding repeated freeze-thaw cycles. Before use, remove and thaw slowly on ice.

[0155] The transfection experiment was conducted according to the following group design: The negative control group (sh-NC) received 10 μL of sh-NC stock solution per mouse; each single knockout group (shNEU1, shCLIC1, and shFLOT1) received 10 μL of the corresponding targeted virus stock solution; the double knockout groups (shNEU1+shCLIC1, shNEU1+shFLOT1, and shCLIC1+shFLOT1) received a mixture of 5 μL of each of the two targeted virus stock solutions before injection; and the triple gene knockout group (shNEU1+shCLIC1+shFLOT1) received a mixture of 3.33 μL of each of the three targeted virus stock solutions before injection. The total amount of virus stock solution used in all groups was 10 μL, and the total viral load was 1 × 10⁻⁶. 8 TU.

[0156] 3.3 Survival rate and survival status assessment Mice were observed daily for 28 days post-surgery to assess survival and plot Kaplan-Meier survival curves. Hair condition, climbing ability, and activity level (0–3 points) were dynamically assessed on days 5, 9, 13, 17, 21, 25, and 28 post-surgery, along with body weight and rectal temperature.

[0157] 3.4 Serum inflammatory factor detection On the 28th day after surgery, the final sample was taken by enucleation of the eyeball, blood was collected, serum was separated by centrifugation, and the levels of TNF-α and IL-6 were detected by ELISA.

[0158] 3.5 Validation of target knockout in spleen tissue Spleen tissue was collected, total RNA was extracted using the Trizol method, and detected by qRT-PCR. NEU1 , CLIC1 , FLOT1 mRNA expression; total protein was extracted by RIPA lysis, and the expression level of each target protein was detected by Western blot.

[0159] 3.6 Detection of inflammatory factors in spleen tissue Total RNA was extracted from spleen tissue using the Trizol method, and the expression levels of TNF-α and IL-6 mRNA were detected by qRT-PCR.

[0160] 3.7 Pathological examination of multiple organs Mice were sacrificed on the 28th day after surgery. Heart, liver, spleen, lung, kidney and cecum were taken, fixed in 4% paraformaldehyde, embedded in paraffin, sectioned, stained with hematoxylin and eosin (HE), and observed under a light microscope. The degree of damage to each organ was assessed by a semi-quantitative scoring system (0-4 points).

[0161] 4. Experimental Results 4.1 Validation of shRNA target knockout efficiency in RAW264.7 cells like Figure 7-8 The qRT-PCR results shown in the figure indicate that, compared with the LPS+sh-NC group, the mRNA expression levels of NEU1, CLIC1, and FLOT1 in each specific shRNA single knockout group and combined knockout group were significantly reduced (P<0.01), indicating that the target genes were effectively silenced at the transcriptional level, and the combined knockout did not cause mutual interference. Western blot results were consistent with the mRNA change trend, showing that the expression levels of the corresponding target proteins in each knockout group were significantly lower than those in the LPS+sh-NC group (P<0.05 or P<0.01). These results demonstrate that each specific shRNA can efficiently silence the gene and protein expression of three target sites in RAW 264.7 cells, with stable and reliable knockout efficiency.

[0162] 4.2 Effects of shRNA targets on LPS-induced RAW264.7 cell viability, inflammatory factors, and target mRNA expression 4.2.1 Cell viability (CCK-8) like Figure 9 CCK-8 assay results showed that cell viability in both the LPS and LPS+sh-NC groups was significantly lower than that in the normal group (P<0.01). Cell viability in each knockout group was higher than that in the LPS+sh-NC group to varying degrees: in the single knockout group... NEU1 The knockout group showed the most significant recovery. CLIC1 and FLOT1 The knockout group showed the second best recovery; the double knockout group showed limited recovery, with no statistically significant difference compared to the LPS+sh-NC group (P>0.05); the triple gene knockout group showed the most significant recovery of cell viability (P<0.01), and was superior to each single and double knockout group. These results indicate that the combined knockout of the three targets has a synergistic protective effect against LPS-induced cell damage.

[0163] 4.2.2 Inflammatory factor secretion levels (ELISA) like Figure 10 ELISA results showed that the levels of TNF-α and IL-6 secretion in the LPS group and the LPS+sh-NC group were significantly higher than those in the normal group (P<0.01). The levels of inflammatory factors in each knockout group were lower to varying degrees than those in the LPS+sh-NC group: in the single knockout group... NEU1 The knockout group showed the most significant decrease; the double knockout group showed a further decrease; and the triple gene knockout group showed the most significant decrease (P<0.01), approaching the level of the normal group. These results indicate that the combined knockout of the three targets can synergistically inhibit LPS-induced inflammatory cytokine secretion.

[0164] 4.2.3 Expression of inflammatory cytokines mRNA (qRT-PCR) like Figure 11 qRT-PCR results showed that the expression of TNF-α and IL-6 mRNA in the LPS group and the LPS+sh-NC group was significantly higher than that in the normal group (P<0.01). The expression levels of inflammatory factor mRNA in each knockout group were lower than those in the LPS+sh-NC group to varying degrees. NEU1 The single-gene knockout group showed the most significant reduction, while the combined knockout of three genes demonstrated the best reduction effect, consistent with the trend observed in ELISA results. These results indicate that knockout of the three targets can inhibit LPS-induced inflammatory cytokine gene expression at the transcriptional level.

[0165] 4.3 Validation of the effect of target gene knockout in mice like Figure 12-13 qRT-PCR results showed that, compared with the sham-operated control group, the spleen tissue of the sepsis model group mice contained [a certain amount of...]. NEU1 , CLIC1 , FLOT1 The mRNA expression levels of all three target genes were significantly increased (P<0.01), indicating that the three target genes were significantly activated in sepsis. After specific shRNA intervention, the mRNA expression levels of each target gene were significantly decreased compared with the model group (P<0.01), among which... NEU1 The knockout group showed the most significant decrease in mRNA levels at each target site, indicating a significant knockout effect. Western blot analysis showed that, compared with the sham-operated control group, the spleen tissue of the sepsis model group mice showed a significantly lower reduction in mRNA levels at each target site. NEU1 , CLIC1 , FLOT1 Protein expression levels were significantly increased (P<0.01), consistent with the trend of mRNA changes. After specific shRNA intervention, the protein expression levels of each target site were significantly decreased compared with the model group (P<0.01), consistent with the trend of mRNA level changes, further validating the effective knockout of each target site at the protein level.

[0166] 4.4 Effects of shRNA targets on serum inflammatory cytokine secretion in septic mice like Figure 14 ELISA results showed that serum TNF-α and IL-6 levels in the sepsis model group were significantly higher than those in the sham-operated control group (P<0.01). The levels of inflammatory factors in each target knockout group were lower than those in the model group to varying degrees. NEU1 The single knockout group showed the most significant decrease, the double knockout group showed better inhibition than the single knockout group, and the triple gene knockout group showed the most significant decrease, which was close to the level of the sham-operated control group (P>0.05).

[0167] 4.5 Effects of shRNA targets on the mRNA levels of inflammatory factors in the spleen tissue of septic mice like Figure 15 qRT-PCR results showed that the expression levels of TNF-α and IL-6 mRNA in the spleen tissue of mice in the sepsis model group were significantly higher than those in the sham-operated control group (P<0.01), indicating that sepsis induced the activation of local inflammatory response in the spleen tissue. After targeted silencing, the expression of inflammatory factor mRNA in each knockout group was reduced to varying degrees compared with the model group. In the single knockout group, NEU1 The knockout group showed the lowest expression levels of TNF-α and IL-6, indicating the most significant inhibitory effect (P<0.01). In the double knockout groups, the expression of inflammatory factors was further reduced in each combined knockout group, showing superior results compared to each single knockout group (P<0.01). The triple gene knockout group showed the lowest expression levels of TNF-α and IL-6 (P<0.01), approaching the levels of the sham-operated control group (P>0.05).

[0168] 4.6 Effect of shRNA targets on 28-day survival rate of septic mice like Figure 16 Kaplan-Meier survival analysis showed that the 28-day survival rate in the sepsis model group was only 26.7%, and each target knockout group improved the survival rate to varying degrees. Among the single knockout groups, the shNEU1 group had the best effect (60.0%), followed by the shFLOT1 group (53.3%), and the shCLIC1 group (46.7%) was the worst. The survival rate of the double knockout groups was further improved (66.7%–80.0%). The survival rate of the triple gene knockout group reached 86.7%, the highest among all intervention groups, an improvement of 60.0 percentage points compared with the model group. These results indicate that silencing the three target genes has a significant protective effect on septic mice, and the effect of multi-gene combined intervention is better than that of single gene, showing a clear synergistic effect.

[0169] 4.7 Effects of shRNA targets on the survival status of septic mice 4.7.1 Hair condition The condition of a mouse's fur is a direct indicator of its general health. A higher score indicates dirtier, more disheveled fur and a poorer health condition. For example... Figure 17 As shown, the hair follicle score in the sepsis model group reached 2.71 on day 5, close to the worst state. The three-gene combination group had a score of only 0.86 on day 5, and completely recovered to 0 on day 28, comparable to the sham surgery group. All treatment groups were significantly better than the sepsis group (P<0.05).

[0170] 4.7.2 Climbing ability Climbing ability reflects a mouse's muscle strength and motor coordination; a higher score indicates a lower ability. For example... Figure 18 As shown, sepsis severely impaired the climbing ability of mice, with a score of 2.86 on day 5, indicating a near-complete loss of climbing ability. The three-gene combined group had a score of 1.14 on day 5 and fully recovered on day 28, significantly better than the sepsis group (P<0.01).

[0171] 4.7.3 Activity Activity level reflects the overall mental state and activity level of mice; a higher score indicates less activity and poorer mental condition. For example... Figure 19 As shown, the activity level of septic mice was significantly reduced, with a score as high as 2.86 on day 5, indicating extreme lethargy. The three-gene combination group had a score of only 0.86 on day 5, but recovered to normal by day 17, showing the most significant improvement (P<0.001). Notably, the shNEU1+shFLOT1 dual genome also completely recovered to normal after day 13.

[0172] 4.7.4 Changes in body temperature Body temperature is an important physiological indicator reflecting the severity of sepsis; sepsis is often accompanied by hypothermia (a drop in body temperature). For example... Figure 20 As shown, the sepsis model group experienced a sharp drop in body temperature to 35.40℃ on day 5, a decrease of 2.16℃, indicating severe hypothermia. The three-gene combined group showed a maximum temperature decrease of only 0.68℃, maintaining a temperature of 37.16℃ on day 5, close to normal. All treatment groups had significantly higher body temperatures than the sepsis group (P<0.01), suggesting that gene intervention can effectively maintain body temperature homeostasis.

[0173] 4.7.5 Weight Changes Weight change is a comprehensive indicator reflecting nutritional status and disease severity. For example... Figure 21 As shown, the sepsis model group experienced a sharp weight loss of 19.1% on day 5 and had not yet returned to baseline levels by day 28. The combined three-gene group achieved a maximum weight loss of only 7.6%, and by day 28, their weight had exceeded baseline (21.74g), showing a normal growth trend. The dual-gene group achieved approximately 10-11% weight loss, and the single-gene group achieved approximately 15-16% weight loss, both significantly better than the sepsis group (P<0.01).

[0174] 4.8 Effects of shRNA targets on the pathological morphology of various organs in septic mice (HE staining) A mouse model of sepsis was established using the cecal ligation and puncture method, and the results were systematically evaluated. NEU1 , CLIC1 , FLOT1 The pathological effects of single, double, and triple knockout of three genes on multi-organ damage caused by sepsis were investigated. All tissue sections were stained with hematoxylin and eosin (HE) and semi-quantitatively scored (0–4 points). The results are as follows: Figure 22-24 As shown.

[0175] In the sham surgery control group, all organ structures remained intact, with no obvious pathological changes observed. The heart showed neatly arranged myocardial fibers with clear striations, and no interstitial edema or inflammatory cell infiltration. The liver exhibited regular hepatic cords, clear sinusoids, and homogeneous cytoplasm. The spleen showed a clear boundary between the white and red pulp, and intact splenic nodules. The lungs showed good alveolar structure with no widening of septa and clean cavities. The kidneys showed clear glomerular structure and intact brush borders of the renal tubules. The cecal mucosa and glands were normal, with no inflammatory reaction observed.

[0176] In the sepsis model group, all organs showed severe damage, with a score of 4. The heart showed extensive myocardial cell necrosis, nuclear pyknosis or dissolution, myocardial fiber rupture, significant interstitial edema with abundant inflammatory cell infiltration and hemorrhage foci; the liver showed lobular structure destruction, extensive hepatocyte swelling, vacuolar degeneration, and large-scale necrosis, with blurred boundaries between necrotic and normal tissues, dilated and congested sinusoids, and abundant inflammatory cell infiltration in the portal areas; the spleen showed structural disorder, significantly reduced white pulp, highly congested red pulp, macrophage proliferation, and hemorrhage; the lungs showed severe alveolar destruction, significantly widened septa, abundant inflammatory cell infiltration, hyaline membrane formation within the alveolar cavities, and consolidation; the kidneys showed extensive necrosis and shedding of renal tubular epithelium, luminal occlusion, abundant protein casts, glomerular atrophy, and interstitial edema with inflammatory cell infiltration; the cecum showed transmural inflammatory necrosis, extensive necrosis of the mucosa and submucosa, destruction of crypts, abundant neutrophil infiltration, and signs of perforation in some areas.

[0177] The single-gene knockout group (sepsis + shNEU1) scored approximately 2 points, showing significantly reduced damage to various organs. Mild myocardial swelling, mild cloudy swelling of hepatocytes with focal necrosis, mild widening of the pulmonary septa, mild cloudy swelling of the renal tubules, and only mild sloughing of the cecum mucosal epithelium without transmural necrosis were observed. The sepsis + shCLIC1 group also scored approximately 2 points, with pathological changes similar to the shNEU1 group, but with significant individual variability; some animals scored as high as 4 points, suggesting that the protective effect of CLIC1 knockout is not stable. The sepsis + shFLOT1 group scored 3.67 points, with the degree of organ damage similar to the model group, exhibiting widespread myocardial necrosis, extensive hepatocyte necrosis, diffuse inflammatory infiltration of the lungs with hyaline membranes, renal tubular necrosis and sloughing with numerous casts, and transmural inflammation of the cecum, suggesting that FLOT1 knockout not only has no protective effect but may even exacerbate the damage.

[0178] The sepsis + shNEU1 + shCLIC1 group (double gene knockout group) scored 1.33 points, with only mild pathological changes in each organ, such as mild interstitial edema of the myocardium, mild hepatocellular turbidity, and slightly widened alveolar septa; the overall structure was basically preserved. The sepsis + shNEU1 + shFLOT1 group scored only 1 point, with HE staining of each organ showing essentially normal tissue structure, very similar to the sham-operated control group. Only a few fields showed mild interstitial edema, with no inflammatory cell infiltration or structural destruction. This result was unexpected, suggesting that the potent protective effect mediated by NEU1 knockout can completely offset the harmful effects of FLOT1 knockout. The sepsis + shCLIC1 + shFLOT1 group scored approximately 2 points, with moderate damage in each organ, similar to the single knockout group.

[0179] In the triple gene knockout group (shNEU1+shCLIC1+shFLOT1), all organ scores were 0. The histological manifestations of the heart, liver, spleen, lungs, kidneys, and cecum were similar to those of the sham-operated control group, with no obvious pathological changes, indicating that triple gene knockout may reverse multi-organ damage caused by sepsis.

[0180] 5. Experimental Conclusions This embodiment systematically compares... NEU1 , CLIC1 , FLOT1 Synergistic effects in multi-organ damage caused by sepsis. Single-gene knockout results showed that... NEU1 and CLIC1 Knockout has a partial protective effect, while FLOT1 Knockout exacerbates the damage. In dual gene knockout, NEU1 and FLOT1 The significant protective effect of combined knockout suggests NEU1 Possible negative regulation FLOT1 Mediated inflammatory pathways. The complete protective effect of the three-gene knockout further suggests that there may be complex signaling network interactions among the three. NEU1 The inhibition may be achieved by downregulating CLIC1 and blocking FLOT1 Related pro-inflammatory pathways synergistically inhibit oxidative stress, apoptosis, and inflammatory cascades, ultimately achieving comprehensive protection of multiple organs. The above results fully demonstrate that... NEU1 Inhibitors CLIC1 Inhibitors and FLOT1 The inhibitory drug composition and the application of combined three-gene intervention in the preparation of antiseptic drugs exhibited unexpected synergistic therapeutic effects. These results provide new experimental evidence for the study of the molecular mechanisms of sepsis-related organ damage and lay a theoretical foundation for the development of multi-target combined intervention strategies, demonstrating significant potential for translational medicine.

[0181] Example 5: Validation of the synergistic interaction effect of shRNA target silencing Based on the results of Example 4, the synergistic inhibitory effect of multi-target combined intervention on the inflammatory response of sepsis was evaluated by statistical analysis.

[0182] 1. Experimental Design and Statistical Methods A three-factor completely randomized factorial design (2×2×2) was used, with three intervention factors: NEU1, CLIC1, and FLOT1 (each factor had two levels: "silenced" and "non-silenced"), resulting in eight intervention groups: ① sh-NC group; ② shNEU1 group; ③ shCLIC1 group; ④ shFLOT1 group; ⑤ shNEU1+shCLIC1 group; ⑥ shNEU1+shFLOT1 group; ⑦ shCLIC1+shFLOT1 group; ⑧ shNEU1+shCLIC1+shFLOT1 group. Three biological replicates were set up in each group for target gene expression detection (mRNA and relative protein expression levels); n=6 was used for each observation group in animal and cell functional assays.

[0183] The detection indicators include: (1) Expression level - the relative protein expression level and mRNA expression level of NEU1, CLIC1 and FLOT1; (2) Animal inflammatory factors - the levels of inflammatory factors IL-6 and TNF-α in mouse peripheral samples; (3) Cellular inflammatory factors and vitality - the levels of IL-6 and TNF-α in cells and the vitality of CCK-8 cells.

[0184] Data are expressed as mean ± standard error (Mean ± SEM). Ordinary three-way ANOVA (GraphPad Prism 9.4.1) was used to test for main effects (NEU1, CLIC1, FLOT1 silence), pairwise second-order interactions, and three-way third-order interactions. The effect size was the percentage of total variance (% of total variance). Tukey's method was used for multiple comparisons between groups. The significance level was set at α = 0.05; P < 0.05 was considered statistically significant (significance markers: **** for P < 0.0001, *** for P < 0.001, ** for P < 0.01, * for P < 0.05, ns for no statistical significance).

[0185] 2. Statistical Analysis Results 2.1 Expression Regulation After Target Silencing Whether protein or mRNA is used as the readout indicator, NEU1 , CLIC1 , FLOT1 All three main effects of silencing were highly significant (all P < 0.0001). The combined main effects at the protein level accounted for 94%–98% of the total variation, and at the mRNA level, they accounted for 92%–94%, indicating that each of the three factors is an independent and potent regulatory node in this signaling network.

[0186] Protein-level interaction patterns vary depending on the readout target. NEU1 When the protein is read, CLIC1 × FLOT1 The interaction was significant (P=0.0026, accounting for 0.690% of the total variance), suggesting that in NEU1 Protein regulation CLIC1 and FLOT1 There is functional synergy. CLIC1 When the protein is read out, only third-order interaction occurs ( NEU1 × CLIC1 × FLOT1 The result was significant (P<0.0001, accounting for 3.581% of the total variation), indicating that... CLIC1 Protein expression regulation requires the simultaneous presence of all three factors to produce a combined effect that transcends the main effect. FLOT1 When the protein is read, NEU1 × CLIC1 (P=0.0002) NEU1 × FLOT1 Both the third-order interaction (P=0.0370) and the third-order interaction (P=0.0005) were significant, indicating that... NEU1 It plays a central hub role in this interactive network.

[0187] At the mRNA level, with CLIC1 and FLOT1 When mRNA was used as the readout indicator, all pairwise interactions and tertiary interactions were highly significant (all P < 0.0001); NEU1 When mRNA is the readout indicator, NEU1 × FLOT1 Interaction P < 0.0001, NEU1 × CLIC1 (P=0.0006) CLIC1 × FLOT1 Both the first-order interaction (P=0.0031) and the third-order interaction (P=0.0045) were significant. These results suggest that any two factors or the combination of any three factors at the transcriptional level can significantly affect the expression of mRNA at various target sites, and that the three factors are in a global synergistic regulatory network.

[0188] 2.2 Interactions of peripheral inflammatory factors in animals All three main effector factors significantly affected peripheral IL-6 and TNF-α levels in mice (all P < 0.0001). All pairwise and tertiary interactions of IL-6 were significant (all P < 0.0001), among which... NEU1 Silent individuals contributed 45.37% of the total variance. FLOT1 Silent contribution: 28.11%. CLIC1The silencing effect contributed 12.81%. All main effects of TNF-α and the interactions between the two factors were significant (P<0.0001 or P=0.0030), and the third-order interaction was also significant (P=0.0048), with an overall trend consistent with IL-6. These results indicate that the factors do not act independently but rather have complex synergistic regulatory relationships.

[0189] 2.3 Interaction between cytoinflammatory factors and cell viability All major effects, pairwise interactions, and third-order interactions of cellular IL-6 and TNF-α were significant (all P < 0.0001), and the effect variation patterns were highly consistent with those in animals, indicating that this set of regulatory pathways can occur at the cellular level.

[0190] Among the CCK-8 cell viability indicators, all main effects were significant (all P < 0.0001); the interaction between the two factors was significant. NEU1 × FLOT1 The P-value was 0.0019, while the P-values ​​for all other pairwise and third-order interactions were <0.0001. Notably, the (AB vs CD)×(AC vs BD) and third-order interactions contributed 29.03% and 27.17% of the total variance, respectively, totaling over 56%, which is significantly higher than their proportion in inflammatory cytokine markers. This indicates that cell survival status is more susceptible to regulation by multi-factor synergistic interactions than inflammatory cytokine release.

[0191] 2.4 Synergistic inhibitory effect of three-target co-silencing Tukey's multiple comparisons showed that the three-target co-silencing group (shNEU1+shCLIC1+shFLOT1) was significantly superior to the other 7 groups at both the protein and mRNA levels (P<0.0001). The mean expression decrease was greater than that of all single and double silencing groups. There were no cross-zero phenomena in the 95% confidence interval, excluding differences from sampling fluctuations. The synergistic inhibitory effect was stable and reproducible.

[0192] 3. Conclusion NEU1 , CLIC1 , FLOT1All three were independent and significant regulatory factors (all main effects P < 0.0001). Their interaction effects exhibited a hierarchical characteristic of "global synergy at the transcriptional level and target specificity at the protein level." At the transcriptional level, they belong to the same synergistic regulatory network, and at the protein level, they form a target-specific functional regulatory network. Tukey's multiple comparisons confirmed that the inhibitory effect of the three-target co-silencing group was significantly better than that of all single-silencing and double-silencing groups, with a 95% confidence interval not crossing 0, indicating a statistically testable synergistic gain. These results support the use of these three as combined targets for multi-target joint intervention, or as combined biomarkers for the subtyping, prognostic assessment, and efficacy monitoring of inflammatory diseases such as sepsis.

[0193] Finally, it should be noted that the above content is only used to illustrate the technical solution of the present invention, and is not intended to limit the scope of protection of the present invention. Simple modifications or equivalent substitutions made by those skilled in the art to the technical solution of the present invention do not depart from the essence and scope of the technical solution of the present invention.

Claims

1. A combination of molecular biomarkers for early diagnosis, prognostic prediction, prognostic assessment, and / or treatment selection of sepsis-related diseases, characterized in that, It includes PLA2G4A , FLOT1 , NEU1 , MSH5 , AGER , CCHCR1 , APOM , CLIC1 , HLA-DRA , NOTCH4 , PPT2 , PDIK1L and ABHD16A One or more of them.

2. The use of the molecular marker combination according to claim 1 in the preparation of reagents or kits for diagnosing and / or screening sepsis-related diseases; characterized in that, The reagents or kits are used for the early diagnosis, prognosis prediction, prognostic assessment, and / or treatment selection of sepsis-related diseases.

3. The use according to claim 2, characterized in that, The reagent or kit is used to detect samples from subjects who are diagnosed with or suspected of having sepsis; Preferably, the sample is derived from the subject's body fluids, cells, tissues, metabolites, and / or excretions; More preferably, the sample is derived from the subject's bodily fluids and is selected from at least one of blood, plasma, extracellular fluid, tissue fluid, urine, lymph, or cerebrospinal fluid; Preferably, the reagent or kit contains a substance capable of specifically detecting the molecular marker or its active fragment as described in claim 1 or 2; preferably, it is a primer, probe, antibody, biochip, or small molecule compound; And / or, the reagent is directly or indirectly labeled with a detectable tag, the detectable tag being selected from at least one of the following: Radioactive isotopes, fluorescent groups, chemiluminescent components, enzymes, enzyme substrates, enzyme cofactors, enzyme inhibitors, dyes, metal ions or ligands.

4. A product for diagnosing and / or screening sepsis-related diseases, characterized in that, The product contains a substance that is specific to or capable of specifically recognizing the molecular marker composition of claim 1 or 2; Preferably, the substance includes reagents, kits, and / or detection devices; Preferably, the substance comprises or is selected from at least one of the following: sequencing reagents, immunological detection reagents, molecular biology detection reagents, flow cytometry detection reagents, immunological detection kits, molecular biology detection kits, sequencing kits, biochip reagents, and biochip kits; More preferably, the substance comprises or is selected from at least one of the following: primers, probes, antibodies, biochips, or small molecule compounds; Preferably, the substance has or is coupled with a detectable marker; More preferably, the detectable marker is selected from at least one of the following: radioactive isotopes, fluorescent groups, chemiluminescent components, enzymes, enzyme substrates, enzyme cofactors, enzyme inhibitors, dyes, metal ions, or ligands.

5. A system for a combination of molecular biomarkers for predicting, assessing, and / or selecting treatment regimens for sepsis-related diseases, characterized in that, The system includes: (1) A data processing module that receives and / or processes molecular-level data comprising combinations of molecular markers as described in claim 1 or 2 of the object; (2) The analysis and calculation module predicts the prognosis of the object based on the data obtained in (1); Preferably, the system further includes a data acquisition module for acquiring a sample of the object, the sample containing data on the molecular marker composition level as described in claim 1 or 2; Preferably, the analysis and calculation module further includes a module for building a prediction model based on the obtained data; Preferably, the step of establishing the prediction model includes: (a) Determination of samples from the subject PLA2G4A , FLOT1 , NEU1 , MSH5 , AGER , CCHCR1 , APOM , CLIC1 , HLA-DRA , NOTCH4 , PPT2 , PDIK1L and / or ABHD16A The levels include the nucleic acid molecule level, preferably the gene level, the mRNA level, the mRNA cDNA level, and / or the protein molecule level; (b) Obtain predictive values ​​for sepsis-related diseases in the subjects; (c) Compare the predicted value with a preset threshold value; when the predicted value of the sample is greater than or equal to the threshold value, the patient is judged to have a high risk of sepsis-related disease; when the predicted value of the sample is less than the threshold value, the patient is judged to have a low risk of sepsis-related disease.

6. The system according to claim 5, characterized in that, The system also includes a module for predicting lead compounds based on the obtained data; Preferably, the lead compound prediction is based on a deep learning algorithm.

7. A computer-readable storage medium comprising a computer program, characterized in that, When the computer program is executed, it runs the system of claim 5 or 6 to perform the functions of the system.

8. A computer device comprising a memory and a processor, the memory storing a computer program, characterized in that, The processor is used to execute the computer program to achieve the functions of the system according to claim 5 or 6.

9. The use of the pharmaceutical composition in the preparation of a medicament for the prevention and / or treatment of sepsis-related diseases, characterized in that, The pharmaceutical composition comprises an active ingredient capable of regulating the levels of the molecular marker group of claim 1; Preferably, the pharmaceutical composition contains substances capable of regulating... NEU1 , CLIC1 and FLOT1 Levels of medication; Preferably, the pharmaceutical composition comprises NEU1 Inhibitors CLIC1 Inhibitors and FLOT1 Inhibitors; More preferably, the NEU1 inhibitor, CLIC1 inhibitor, and FLOT1 inhibitor are respectively targeted NEU1 Gene, CLIC1 Genes and FLOT1 Gene siRNA, shRNA, or antisense oligonucleotide (ASO), or The NEU1 Inhibitors CLIC1 Inhibitors and FLOT1 Inhibitors are targeted NEU1 protein, CLIC1 Protein and FLOT1 Antibodies against proteins or their antigen-binding fragments or small molecule compounds; Preferably, the pharmaceutical composition further comprises a pharmaceutically acceptable carrier or excipient.

10. The application according to claim 9, characterized in that, The pharmaceutical composition has at least one of the following applications: (a) Inhibits the expression of sepsis-related inflammatory factors; (b) Reduce mortality rates in individuals with sepsis-related diseases; (c) Reduce multi-organ damage in patients with sepsis-related diseases; (d) Improve the survival status of people with sepsis-related diseases.

Images