Search

CN-122000060-A - Short-term prognosis prediction method for liver failure by fusing multiscale characteristics

CN122000060ACN 122000060 ACN122000060 ACN 122000060ACN-122000060-A

Abstract

The invention discloses a liver failure short-term prognosis prediction method with multi-scale characteristics, which relates to the technical field of medical prognosis prediction, and comprises the steps of collecting serum proteome and clinical data of patients at the time of admission and 24 hours later; the method comprises the steps of constructing a hepatocyte core metabolism network calculation model, converting serum protein concentration change into protein characteristic energy values, disturbing the metabolism model by using the protein energy values and extracting virtual metabolism flux characteristics, extracting multi-scale characteristics from clinical time sequence data, carrying out nonlinear fusion on the metabolism flux characteristics and the clinical characteristics to generate a prognosis index, and outputting death risk classification results according to the prognosis index. According to the invention, by constructing the hepatocyte specific metabolic network model, the serum proteome time sequence is mapped into the quantitative flux characteristic reflecting the liver core metabolic function state, and nonlinear coupling is carried out with the clinical time sequence index, so that quantitative evaluation of liver dynamic metabolic disorder can be realized, and comprehensive judgment of short-term prognosis of patients is completed.

Inventors

  • GAO SHUO
  • ZHOU JIN
  • LIU ZHI

Assignees

  • 天津财经大学

Dates

Publication Date
20260508
Application Date
20260126

Claims (8)

  1. 1. A method for short-term prognosis of liver failure incorporating multiscale features, the method comprising the steps of: Acquiring patient serum proteomic data and clinical temporal data; constructing a hepatocyte specific metabolic network model based on the liver specific function information; Calculating a protein characteristic energy value according to the dynamic change of the protein concentration in the serum proteomics data and the functional importance of the protein concentration in the metabolic network model; Based on the protein characteristic energy value, personalized constraint is applied to the hepatocyte-specific metabolic network model, and metabolic flux characteristics reflecting the metabolic function state of the liver are generated by solving a flux balance problem under a preset physiological function target; Extracting multi-scale clinical features of the clinical temporal data; Fusing the metabolic flux features and the multi-scale clinical features to generate coupling features and calculating a prognosis index; and outputting a prognosis risk classification result based on the prognosis index.
  2. 2. The method for short-term prognosis of liver failure with fusion of multiscale features according to claim 1, wherein the obtaining of patient serum proteomic data and clinical temporal data specifically comprises: respectively collecting serum samples at a first time point after the patient is admitted and a second time point which is separated by a preset time interval; Performing a proteomic test on the serum sample to obtain protein expression profile data corresponding to two time points; and synchronously acquiring clinical parameter data at the two time points.
  3. 3. The method for short-term prognosis of liver failure with fusion of multiscale features according to claim 1, wherein the constructing a hepatocyte-specific metabolic network model specifically comprises: extracting metabolic reactions related to the core metabolic functions of the liver cells from a genome scale metabolic network model to form an initial reaction set; Integrating liver-specific proteomic information, correlating proteins involved in the serum proteomic data with their catalyzed metabolic reactions to activate and constrain the corresponding reactions in the initial set of reactions; Defining a metabolite associated with serum detection as an exchange metabolite of the model; based on the reaction set and the exchange metabolites, a hepatocyte-specific metabolic network model with steady state mass balance constraints is formed.
  4. 4. A method for short-term prognosis of liver failure in combination with multiscale features according to claim 1, wherein the generating metabolic flux features specifically comprises: calculating enzyme constraint adjustment factors of each metabolic reaction in the hepatocyte-specific metabolic network model according to the protein characteristic energy value, wherein the adjustment factors represent the modulation effect of corresponding protein changes on the maximum flux of the reaction; constructing a flux balance analysis model with a target of maximizing a preset liver physiological function by taking the modulated enzyme constraint as a boundary condition; solving the flux balance analysis model to obtain personalized steady-state metabolic flux distribution; extracting values representing key metabolic pathway fluxes from the personalized steady state metabolic flux distribution, forming the metabolic flux characteristics.
  5. 5. The method for short-term prognosis of liver failure with fusion of multiscale features according to claim 1, wherein the calculation of protein feature energy values specifically comprises: calculating the protein concentration variation of the same patient at two time points; Calculating, for each protein, a functional centrality weight based on the stoichiometric scale of all metabolic reactions it catalyzes in the hepatocyte-specific metabolic network model, and the statistical significance of its concentration variation in the patient population; And calculating the characteristic energy value of the protein by combining the direction and the amplitude of the protein concentration variation and the functional center weight.
  6. 6. The method for short-term prognosis of liver failure with fusion of multiscale features according to claim 1, wherein the extracting multiscale clinical features of the clinical temporal data specifically comprises: extracting a parameter value of a first time point from the clinical time sequence data as a point scale feature; Calculating the absolute change quantity and the relative change rate of the parameter between the two time points to be used as a difference scale characteristic; Calculating a change trend estimated value based on the parameter values of the two time points to serve as a track scale feature; and splicing the point scale features, the difference scale features and the track scale features to form the multi-scale clinical features.
  7. 7. A method of short-term prognosis of liver failure in combination with a multiscale feature according to claim 1, wherein the fusing of the metabolic flux feature with the multiscale clinical feature generates a coupling feature, comprising: respectively carrying out standardization treatment on the metabolic flux characteristics and the multi-scale clinical characteristics; Respectively carrying out linear projection on the normalized metabolic flux characteristics and the multi-scale clinical characteristics; Performing element-by-element multiplication operation on the projected features to simulate interaction among different modal features; and applying nonlinear transformation to the interaction result to generate the coupling characteristic.
  8. 8. The method for short-term prognosis of liver failure with fusion of multiscale features according to claim 7, wherein the calculating the prognostic index specifically comprises: Splicing the normalized metabolic flux characteristic, the normalized multi-scale clinical characteristic and the coupling characteristic; inputting the spliced characteristics into a prediction model for processing; outputting the prognostic index, wherein the prognostic index is a probability value representing the risk of death of the patient within a predetermined period of time.

Description

Short-term prognosis prediction method for liver failure by fusing multiscale characteristics Technical Field The invention belongs to the technical field of medical prognosis prediction, and particularly relates to a liver failure short-term prognosis prediction method integrating multiscale characteristics. Background Short-term prognosis of liver failure, particularly in the early window period (usually several days to one month) after patient admission, is assessed and predicted based on its clinical and biological data for the risk of occurrence of adverse clinical outcomes such as death. The core goal of this prediction is to identify high-risk patients with possible severe exacerbations based on a combination of criteria including mainly static serum biochemical indicators (e.g., bilirubin, prothrombin time) reflecting liver synthesis, detoxification and excretion functions, and integrated multiple indicators. In the prognosis of liver failure, the prior art methods rely mainly on statistical analysis of patient clinical indicators (e.g. blood clotting functions, bilirubin levels) and single or combined serum protein concentrations. These methods typically treat the detected proteins as independent biomarkers, looking for statistical associations between their concentration and patient prognosis by constructing multiple regression models or machine learning classifiers. However, liver failure, particularly slow-onset acute liver failure, has its pathophysiological core involving complex systemic metabolic disorders caused by massive necrosis or dysfunction of hepatocytes. The existing modeling paradigm, directly from protein concentration to prognostic signature, has a fundamental limitation that it cannot translate discrete, static protein abundance measurements into quantitative descriptions of dynamic, systemic dysfunction of the liver, a core metabolic organ, in a disease state. Specifically, the concentration change of proteins (such as coagulation factors and acute phase proteins) detected in serum is the comprehensive result of liver synthesis, secretion, clearance and other functional disorders, and is also the manifestation of pathological processes such as systemic inflammatory reaction and the like. The prior art, while able to identify proteins associated with poor prognosis, fails to explain how changes in these proteins specifically and quantitatively perturb the metabolic network inside hepatocytes, such as the flux of key biochemical pathways necessary for life maintenance, e.g., urea cycle, gluconeogenesis, etc. This results in a model that is more like a "black box" and that predicts based on statistical correlations, but lacks an explicit, interpretable link to the core pathophysiological mechanism of the decline in the actual metabolic function reserves of the liver. Accordingly, the following scheme is proposed for the above problems. Disclosure of Invention The invention aims to provide a liver failure short-term prognosis prediction method with multi-scale characteristics, which is characterized in that a liver cell specific metabolic network model is constructed, serum proteome time sequence is mapped into quantitative flux characteristics reflecting the metabolic function state of a liver core, and nonlinear coupling is carried out on the quantitative flux characteristics and clinical time sequence indexes, so that quantitative evaluation of dynamic metabolic disorder of the liver can be realized, comprehensive judgment of short-term prognosis of a patient is completed, and the problem that the dynamic process of systemic dysfunction of the liver cannot be quantitatively represented due to the fact that the prior art depends on statistical association of static biomarkers is solved. In order to solve the technical problems, the invention is realized by the following technical scheme: The invention discloses a liver failure short-term prognosis prediction method with multi-scale characteristics fused, which comprises the following steps: Acquiring patient serum proteomic data and clinical temporal data; constructing a hepatocyte specific metabolic network model based on the liver specific function information; Calculating a protein characteristic energy value according to the dynamic change of the protein concentration in the serum proteomics data and the functional importance of the protein concentration in the metabolic network model; Based on the protein characteristic energy value, personalized constraint is applied to the hepatocyte-specific metabolic network model, and metabolic flux characteristics reflecting the metabolic function state of the liver are generated by solving a flux balance problem under a preset physiological function target; Extracting multi-scale clinical features of the clinical temporal data; Fusing the metabolic flux features and the multi-scale clinical features to generate coupling features and calculating a prognosis index; and outputting a prog