Devices and methods for screening biomarkers of immune status

Abstract

This invention relates to an apparatus and method for screening biomarkers of immune status. The method includes: sampling biological samples from multiple individuals, the biological samples being divided into a first data set and a second data set; a genetic algorithm of the screening optimization device is used to screen the data sets, and then an orthogonal array design and simulated fire exchange strategy are used to construct a classification model using sample inference methods to predict high-risk groups before infection, thereby achieving preventive intervention, such as priority vaccination or targeted prevention and control.

Claims

1. An apparatus for screening biomarkers of immune status, comprising: The database has a first dataset and a second dataset, wherein the first dataset contains at least one first protein expression data and the second dataset contains at least one second protein expression data; A screening and optimization device, connected to the database, uses a genetic algorithm to screen the first or second dataset, and then employs orthogonal array design and simulated fire reduction exchange strategy to search for protein feature combinations with predictive performance to reduce the risk of overfitting to high-dimensional data and obtain samples. A construction and training device, connected to the screening and optimization device, uses the samples and a sample inference method to construct a classification model, trains it using the protein feature combinations, and evaluates its accuracy and stability through leave-one-out cross-validation to obtain the model. A validation and prediction device, connected to the construction and training device, inputs the protein feature combinations into the trained model to predict on an independent validation dataset, and calculates the average value through repeated calculations; this average value is an immune status biomarker.

2. The apparatus for screening immune status biomarkers as described in claim 1, wherein, The screening and optimization device has a genetic calculus and an array calculus.

3. The apparatus for screening immune status biomarkers as described in claim 1, wherein, The constructed training device has a classifier and a cross-validator.

4. The apparatus for screening immune status biomarkers as described in claim 1, wherein, The verification and prediction device has a verification and prediction unit.

5. A method for screening biomarkers of immune status, comprising the following steps: Step S01, Exploration cohort: Constructing an exploration cohort for a training device, the cohort having N samples, and proceeding to step S02, or the construction and training device skipping training and proceeding directly to step S05; Step S02, Training with N-1 samples: The construction and training device using leave-one-out cross-validation to evaluate the performance of the N-1 samples; Step S03, Genetic computation: Using one sample in each loop, a screening and optimization device performs genetic computation feature selection, repeating the loop until the genetic computation feature selection for the Nth sample is completed, to generate the optimal subset record; Step S04, Model: Using the N-1 samples as the test in each loop, repeating the loop until the N-1 samples are tested, to generate a prediction result; Step S05, Testing the i-th sample: The construction and training device testing with the i-th sample to generate the prediction result; Step S06, Record the prediction result and the best feature subset: The construction and training device records the prediction result and the best feature subset into the database; Step S07, Is this the last sample: If yes, proceed to step S08; if no, proceed to step S01, the construction and training device provides the (i+1)th sample; Step S08, Evaluate the leave-one-out cross-validation result: The construction and training device evaluates each loop with leave-one-out cross-validation, and repeats the loop until all samples have completed the leave-one-out cross-validation evaluation, and the construction and training device integrates the test result and the best feature subset.

6. The method for screening immune status biomarkers as described in claim 5 further includes a genetic algorithm, comprising the following steps: Step S09, Initial Chromosome: Biological samples from multiple individuals are collected, and these samples are divided into a first dataset and a second dataset; Step S10, Adaptability Assessment: The screening optimization device uses a genetic algorithm to screen the first dataset or the second dataset, and then uses orthogonal array design and simulated fire reduction exchange strategy to search for protein feature combinations with predictive performance to reduce the risk of overfitting to high-dimensional data, and obtain at least one calculation result; Step S11, Maximum Generation: The construction and training device constructs a classification model using the calculation result and a sample inference method, trains it using the protein feature combination, and evaluates the accuracy and stability of the classification model through leave-one-out cross-validation, and obtains the model; The construction and training device determines whether the model is the maximum generation, and if so, proceeds to step S12; Step S12, Discovering the optimal chromosome: The construction training device discovers the optimal chromosome, which is further validated by external validation and performance evaluation methods. The average value after validation is used as an immune status biomarker.

7. The method for screening immune status biomarkers as described in claim 6, wherein, If step S11 is not successful, proceed to step S13; Step S13, Selection: The screening and optimization device selects at least two groups of proteins from the first dataset or the second dataset; Step S14, Exchange: The screening and optimization device uses a genetic algorithm combined with orthogonal array design and simulated fire-reduction exchange strategy to optimize the protein feature combination of the at least two groups of proteins, and obtains a first optimized dataset and a second optimized dataset; Step S15, Mutation: The screening and optimization device continues to use the genetic algorithm combined with orthogonal array design and simulated fire-reduction exchange strategy to optimize the protein feature combination of the first optimized dataset and the second optimized dataset, so that the first optimized dataset and the second optimized dataset mutate to form a first new dataset and a second new dataset; Step S16, New Chromosome: The first new dataset and the second new dataset are new chromosomes, and return to step S10.

8. The method for screening immune status biomarkers as described in claim 5 further includes a model external validation and performance evaluation method, comprising the following steps: Step S17, exploring groups: the validation prediction device explores groups for the optimal chromosome; Step S18, using key protein feature combinations: key protein feature combinations are selected as input features from the prediction results and the optimal feature subset record, and proceed to step S20; Step S19, validating groups: the validation prediction device predicts the groups to generate external validation, and proceeds to step S20; Step S20, model: the validation prediction device predicts the key protein feature combinations based on the external validation and obtains a model; Step S21, evaluating the model: the model is evaluated to obtain accuracy, sensitivity, specificity, and area under the feature curve, and the average value is obtained, which is the immune status biomarker.

9. The method for screening immune status biomarkers as described in claim 6, wherein, The biological sample can be a cell or tissue extract, blood, serum, plasma, saliva, urine, sputum, cerebrospinal fluid, tears, sweat, or feces.

Images