CN-121983349-A - UC plant medicine intervention system integrating multidimensional biomarkers

Abstract

The invention discloses a UC plant medicine intervention system integrating a multidimensional biomarker, which belongs to the field of clinical medicine and comprises an acquisition processing module, an analysis layering module, a virtual screening module, a simulation verification module, a closed-loop optimization module, a preparation control module and a twin prediction module; the invention avoids the problem of treatment mismatch caused by single symptoms or experience compatibility, is obviously superior to the traditional mode of static scheme or 'one-time decision long-term use', reduces artificial subjective interference, improves the authenticity and repeatability of screening and optimizing the plant medicine formula, avoids failure in the later clinical or amplified preparation stage, improves the overall research and development and conversion efficiency, and obviously improves the refinement level of chronic disease management.

Inventors

• WANG SHAOKANG
• CHEN XIANGJUN
• Hao Zhengyang
• PENG QING

Assignees

• 西藏民族大学

Dates

Publication Date

20260505

Application Date

20260225

Claims (6)

1. 1. The UC plant medicine intervention system integrating the multidimensional biomarker is characterized by comprising an acquisition processing module, an analysis layering module, a virtual screening module, a simulation verification module, a closed-loop optimization module, a preparation control module and a twin prediction module; the acquisition processing module is used for acquiring and preprocessing basic information of a patient in an intervention initial stage; The analysis layering module is used for analyzing each piece of basic information after pretreatment, carrying out patient typing and risk layering, and identifying patient biomarkers; the virtual screening module designs virtual plant medicine components according to the biomarker characteristics of the patient and screens out candidate plant medicine matching schemes; the simulation verification module is used for constructing UC in-vitro microenvironment, testing candidate plant drug proportioning schemes and generating multiple groups of experimental response data; The closed-loop optimization module corrects the components and the component weights of the plant medicine proportioning scheme in real time according to in-vitro experimental data; the preparation control module automatically regulates and controls the preparation parameters of the plant medicine according to the optimized plant medicine proportioning scheme; The twin prediction module dynamically updates a patient simulation model according to patient real-time monitoring data, patient genome information and clinical indexes, and predicts disease evolution under different intervention strategies.
2. 2. The UC plant drug intervention system integrating multidimensional biomarkers according to claim 1 wherein said resolution stratification module performs patient typing and risk stratification, the specific steps of identifying patient biomarkers are as follows: S1.1, carrying out quality control processing on a sample layer on each group of preprocessed basic information of a patient, carrying out corresponding data transformation on each basic information after the quality control processing, carrying out Z-standardization processing on each basic information after the transformation, and recording and storing statistics used by mRNA, protein and metabolite molecules in each basic data after each transformation; S1.2, grouping mRNA, protein and metabolite molecules in basic information of each patient according to clinical labels, respectively calculating the mean value and sample variance of the molecules among the groups, adopting parameter test to obtain statistic and bilateral p values, performing multiple test correction on all the molecules, extracting the groups of molecules with the statistic meeting a preset threshold after the correction is completed, and establishing a corresponding candidate set; S1.3, sorting candidate molecules in a candidate set from high to low according to statistics, enriching each candidate molecule based on sorting results, calculating enrichment scores of each prior path set, evaluating concentration degree of molecules in the paths in full sorting, calculating significance of enrichment scores of each path through replacement test, carrying out multiple test correction, selecting paths with significance higher than a preset threshold, and establishing corresponding path sets; S1.4, based on the unified scale, merging the candidate molecules from each group into corresponding feature matrixes, modeling the UC target through a regularization method, then carrying out scale adjustment on the inside of each group, then adaptively adjusting the cross-group learning weight, carrying out final feature weighted ranking on the adjusted weight, and simultaneously outputting the integrated importance score of each molecule or channel, and constructing a corresponding sample space for each patient based on the established feature matrixes; S1.5, calculating a distance matrix among samples in each patient sample space through weighted Euclidean distance, establishing a plurality of groups of sample center clusters through a hierarchical clustering method, dividing the patient into a plurality of groups of subtypes based on the established sub-sample center clusters, generating characteristic pedigrees of each subtype, establishing and training a risk model for each subtype which is already typed by utilizing follow-up result data, calculating risk scores of each patient through the trained model, and dividing the patient into low, medium or high risk groups according to a preset threshold interval; S1.6, performing joint interpretation on the risk model and the parting result, identifying molecules or paths corresponding to high risks to determine preferential intervention targets, selecting clinically-intervened or pharmaceutically-accessible molecules from the identified candidate molecules and paths as processing candidates, then constructing sample-level weights, simulating randomization processing allocation, calculating average processing effects based on the allocation results, and evaluating potential causal effects of the corresponding molecules or paths on rehabilitation conditions of corresponding patients according to the calculation results.
3. 3. The multi-dimensional biomarker integrated UC plant drug intervention system of claim 2 wherein said virtual screening module designs virtual plant drug components based on patient biomarker characteristics and screens candidate plant drug matching schemes by the specific steps of: S2.1, mapping multidimensional biomarkers of a patient into a unified feature vector set, filling missing values of each type of biomarkers, screening required biomarkers based on preset feature selection rules, carrying out numerical transformation on each biomarker, and carrying out scale treatment on each biomarker based on the mean value and standard deviation of the numerical transformation to generate a patient condition vector; S2.2, using known compound descriptors and corresponding targets or path labels to supervise and train the condition variable self-encoder, inputting condition vectors of all patients into the condition variable self-encoder as conditions after training, generating a plurality of groups of potential vectors through multiple sampling, performing posterior verification on each generated group of potential vectors, and integrating all the potential vectors passing the verification to generate corresponding candidate virtual molecules; S2.3, inputting each candidate virtual molecule into a plurality of groups of independent predictors, outputting the activation probability of each candidate virtual molecule to each signal path through each predictor, and carrying out probability calibration on each output result to obtain a corresponding reliable probability value and estimate uncertainty, and when a plurality of groups of predictors exist in conflict signals, weighting and fusing according to preset credibility to obtain a comprehensive effect spectrum of each virtual molecule; s2.4, calculating a synthesis accessibility index of each candidate virtual molecule, evaluating each safety endpoint predicted value of each candidate virtual molecule based on a toxicity prediction model, integrating the safety endpoint predicted values and the comprehensive effect spectrum into corresponding safety-availability scores according to the synthesis accessibility index, each safety endpoint predicted value and the comprehensive effect spectrum by priority and configurable weight, and screening out candidate virtual molecules with the safety-availability scores lower than a preset threshold; S2.5, carrying out similarity retrieval on the screened virtual molecules and a description subset of the real plant compounds in a chemical description subspace by a cosine similarity method, and based on retrieval results, each virtual molecule corresponds to a plurality of groups of candidate real plant compounds with similarity higher than a preset threshold value, and retrieving availability information of each candidate real plant compound, if the candidate real plant can not be directly obtained, recording alternative candidates and extracting steps; S2.6, establishing a candidate element library according to the comprehensive effect spectrum, the safety-availability score and the candidate real plant availability and cost information, searching component proportions by a Pareto front search method, calculating a comprehensive target score of each component proportion, simultaneously arranging each component proportion from high to low according to the comprehensive target score based on the known producibility constraint, establishing a candidate plant medicine formula list, and adding operation suggestions prepared by corresponding plant sources for each candidate plant medicine formula.
4. 4. The UC plant drug intervention system integrating multidimensional biomarkers according to claim 1 wherein said simulated verification module generates a plurality of sets of experimental response data in the specific steps of: S3.1, according to the current experimental purpose, selecting human-derived intestinal organoids, intestinal epithelial cells and immune cells for co-culture, or adopting inflammatory factors to induce on single-layer epithelial cells, recording the source, batch and culture medium components, carrying out standardized inoculation on each culture medium, setting a judgment standard reaching a baseline stable state, measuring a plurality of groups of baseline indexes before entering drug treatment, and eliminating the culture medium which does not reach the standard; S3.2, preparing corresponding plant mother liquor based on candidate plant drug formulas, calculating the volume of the mother liquor required by each group of culture media, the concentration of solvent and final holes, designing multiple concentration gradients, setting the repeated group number of each group of candidate plant drug formulas on a plate, using an automatic liquid processor to sample, randomizing the positions of different formulas in the plate, and setting positive control and negative control at the plate edge; s3.3, measuring and recording inflammatory factor expression, epithelial barrier, cell activity and morphological multi-class endpoint data of each sample under each candidate plant drug formula in a preset time period, carrying out in-batch standardization on the multi-class endpoint data, and recording the original and standardized results of each time point of each hole; s3.4, measuring TEER of each sample at each time point before and after treatment, recording measurement original values, collecting each sample, measuring fluorescence intensity, converting the fluorescence intensity into permeability through a standard curve, calculating barrier function indexes of the corresponding samples based on the TEER and the permeability, and carrying out normalization treatment on each barrier function index according to holes; S3.5, collecting culture supernatant and cell samples of each candidate plant drug formula at a preset time point, respectively performing metabolome and transcriptome analysis, simultaneously recording sample volume, dilution factors and batches, performing quantization treatment on metabolites or transcripts according to a standard curve in the samples to obtain corresponding expression quantity time sequence data, and quantifying dynamic response by adopting an AUC method; And S3.6, carrying out in-batch normalization processing on the data acquired in each preset time point, calculating Z 'factors read out by each group of culture mediums, if the Z' factors are smaller than a preset threshold value, eliminating the corresponding culture mediums, calculating corresponding combination scores according to various endpoint data, marking candidate plant medicine formulas with the combination scores meeting the hit rules as hits based on preset hit rules, generating a candidate plant medicine formula experiment verification list according to the combination scores from high to low, and recording the original values, standardized values and combination scores of each candidate plant medicine formula at each endpoint.
5. 5. The UC plant drug intervention system integrated with multidimensional biomarkers according to claim 3 wherein said closed loop optimization module modifies in real time the plant drug proportioning scheme components and component weights as follows: S4.1, receiving in-vitro experimental data, converting the in-vitro experimental data into a structured form by adopting a predefined analyzer, recording metadata of hole sites, formula IDs, processing concentration and time points, detecting missing values, abnormal values and consistency with standard curves of the experimental board in the newly input in-vitro experimental data in real time, meanwhile, marking the abnormal data, eliminating or correcting the abnormal data according to a preset processing rule, and carrying out real-time standardized processing on each group of in-vitro experimental data according to the standard curves in the experiment; S4.2, collecting predicted values output by predictors corresponding to the plant drug formulas to the same end point, mapping the predicted values according to the same scale as the experiment, calculating predicted-observed residual errors of the end points, respectively giving different weights according to importance of the end points, then counting weighted residual errors of all the end points to form corresponding inconsistent metrics, updating corresponding predictor parameters by adopting an adaptive optimizer based on the inconsistent metrics, and then correcting and quantifying uncertainty of each predictor by adopting Bayes according to small samples and adjusting learning step length, and storing parameter snapshots and observation-predicted residual errors of the wheel after the predictors are updated; S4.3, predicting the effect mean value and uncertainty of each potential ratio by using each updated predictor, calculating the acquisition value of each potential ratio, simultaneously arranging each potential ratio according to the acquisition value from high to low, and selecting each potential ratio with the acquisition value higher than a preset threshold value for in-vitro verification; S4.4, mapping the theoretical ratio into an experimentally-realized mass ratio, correcting potential ratios which do not meet the known experimental constraint, then packaging the executable ratios in batches, issuing the executable ratios to an experimental platform, generating unique IDs for the executable ratios in LIMS, and evaluating the relative improvement rate of the optimal comprehensive scores of a plurality of rounds after each closed loop iteration is finished; And S4.5, if the improvement is lower than a preset threshold, judging convergence and triggering termination, if the change of the acquired value is detected to be higher than the preset threshold in iteration, automatically triggering rollback to the previous round of security parameter snapshot and submitting the condition to manual expert examination, and simultaneously adjusting weight, correcting QC rules or introducing external verification data to recalibrate the predictor by the expert.
6. 6. The multi-dimensional biomarker integrated UC plant drug intervention system of claim 1 wherein said twin prediction module dynamically updates patient simulation models, predicting disease evolution under different intervention strategies as follows: s5.1, pulling each observation data of continuous monitoring flow, discrete sampling group, baseline genome variant information and clinical electronic medical record from each data source, aligning each observation data according to a uniform time reference, reserving time stamps and batch metadata, and then respectively carrying out corresponding modularized preprocessing on the observation data of different modes to generate corresponding mode characteristics; S5.2, generating corresponding fusion feature vectors by weighting and summing multi-modal features in the same time window, adding corresponding time window identifiers and credibility weights for each fusion feature, establishing a patient simulation model containing short-term pathological load indexes, access activation degrees, barrier function indexes and predictive risk quantity mixed sub-vectors, and then performing multiple look-ahead simulation on candidate intervention strategies through the patient simulation model in a preset time window to obtain time sequence output, and storing simulation results as a track set; S5.3, after the new fusion feature vector is generated, a patient simulation model is developed from the last time point to the prior state at the current moment through prospective simulation, the prior state of the current model and the latest fusion feature vector are subjected to weighted correction according to the confidence level, a latest track set is generated, then an end point and an intermediate index corresponding to clinic are set, and a corresponding scalar is calculated for each track according to a time window; S5.4, converting each scalar on the track into a risk score through a result mapping function, then sequencing each candidate intervention strategy according to the risk score from low to high to form a strategy priority list, carrying out sliding window analysis on the simulation track of the sliding window analysis strategy on the simulation set of each candidate strategy, calculating expected gains and implementation costs of different lengths and different starting time windows from the current moment, acquiring expected net gains of each time window according to the acquired expected gains and implementation costs, taking the time window with the highest expected net gain as an optimal adjustment window, and generating corresponding clinical prompts based on the optimal adjustment window.

Description

UC plant medicine intervention system integrating multidimensional biomarkers Technical Field The invention relates to the field of clinical medicine, in particular to a UC plant medicine intervention system integrated with a multidimensional biomarker. Background Ulcerative colitis (Ulcerative Colitis, UC) is an inflammatory bowel disease characterized by chronic, recurrent inflammation of the colonic mucosa, whose pathogenesis involves complex interactions of multiple factors such as genetic susceptibility, immune imbalance, impaired intestinal barrier function, microecological disorders and environmental factors. At present, clinical treatment mainly depends on 5-aminosalicylic acid preparations, glucocorticoids, biological preparations and small molecular targeted drugs, but still has the problems of obvious individual difference of curative effects, insufficient long-term safety, high recurrence rate, drug resistance or intolerance of partial patients and the like, and an intervention strategy with more individuation and long-term management capability is needed. The botanical drug has the advantages of multiple targets, systemic modulation and relative safety in UC intervention, and has been widely used for adjuvant or long-term maintenance therapy. However, the traditional herbal medicine application depends on empirical compatibility, lacks accurate matching based on the molecular characteristics of patients, has limited preparation process, in-vivo distribution and curative effect evaluation means, and is difficult to adapt to the disease characteristics of high heterogeneity of UC. Furthermore, current studies focus on single biomarker or static endpoint assessment, making prospective predictions of disease dynamic evolution and risk of recurrence difficult to achieve. The existing UC plant medicine intervention system has the problems of treatment mismatch caused by single symptom or experience compatibility, has higher artificial subjective interference components, reduces the authenticity and repeatability of plant medicine formula screening and optimizing, has higher failure risk in the later clinical or amplified preparation stage, and reduces the overall research and transformation efficiency, so we propose the UC plant medicine intervention system integrating multidimensional biomarkers. Disclosure of Invention The invention aims to solve the defects in the prior art, and provides a UC plant medicine intervention system integrated with a multidimensional biomarker. In order to achieve the above purpose, the present invention adopts the following technical scheme: The UC plant medicine intervention system integrating the multidimensional biomarker comprises an acquisition processing module, an analysis layering module, a virtual screening module, a simulation verification module, a closed-loop optimization module, a preparation control module and a twin prediction module; the acquisition processing module is used for acquiring and preprocessing basic information of a patient in an intervention initial stage; The analysis layering module is used for analyzing each piece of basic information after pretreatment, carrying out patient typing and risk layering, and identifying patient biomarkers; the virtual screening module designs virtual plant medicine components according to the biomarker characteristics of the patient and screens out candidate plant medicine matching schemes; the simulation verification module is used for constructing UC in-vitro microenvironment, testing candidate plant drug proportioning schemes and generating multiple groups of experimental response data; The closed-loop optimization module corrects the components and the component weights of the plant medicine proportioning scheme in real time according to in-vitro experimental data; the preparation control module automatically regulates and controls the preparation parameters of the plant medicine according to the optimized plant medicine proportioning scheme; The twin prediction module dynamically updates a patient simulation model according to patient real-time monitoring data, patient genome information and clinical indexes, and predicts disease evolution under different intervention strategies. As a further aspect of the present invention, the analysis layering module performs patient typing and risk layering, and the specific steps for identifying patient biomarkers are as follows: S1.1, carrying out quality control processing on a sample layer on each group of preprocessed basic information of a patient, carrying out corresponding data transformation on each basic information after the quality control processing, carrying out Z-standardization processing on each basic information after the transformation, and recording and storing statistics used by mRNA, protein and metabolite molecules in each basic data after each transformation; S1.2, grouping mRNA, protein and metabolite molecules in basic information of each