Model training method for depression detection, detection system, program, and storage medium

Abstract

A kind of algorithm model for depression detection, detection system, program and storage medium.The training method of the artificial intelligence model includes: constructing artificial intelligence model, its input includes the concentration of 3-4 kinds of to-be-detected compounds selected from 5-hydroxyanthranilic acid, methionine, gamma-aminobutyric acid, 3,4-dihydroxyphenylacetic acid, 2-picolinic acid, quinolinic acid, kynurenine, 5-hydroxytryptamine, 3-hydroxykynurenine, 3-hydroxyanthranilic acid, dopamine, norepinephrine, homocysteine, and the output is the risk index of suffering from severe depression;Large sample data is used to train the artificial intelligence model.The AUC of the detection model of the application is greater than or equal to 0.95, even 0.99;Accuracy rate reaches more than 90%;High sensitivity;Algorithm is simple, iterative convergence is fast, is convenient for transplantation, can be embedded into detection system, also can be used as a single data processing unit, works using encryption key or external U disk mode.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of intelligent detection, in particular to a model training method for depression detection, a detection system adopting the same, an executable program and a storage medium. BACKGROUND

[0002] In recent years, with the continuous development and cross-fusion of genomics, transcriptomics, proteomics and polypeptidomics technologies, specific biomarkers closely related to the occurrence and development of depression have also been continuously screened and discovered. These biochemical substances that can mark the changes or possible changes in the system, organs, tissues, cells and subcellular structures or functions during the occurrence and development of depression and during the treatment of antidepressants are defined as depression markers (DM). However, there are few depression markers used as diagnostic indicators for clinical depression, and most of the research on biomarkers closely related to depression and their detection methods remains in the basic research stage.

[0003] The development of artificial intelligence technology brings new changes to various industries. Artificial intelligence technology has also been applied to the diagnosis of diseases, such as the diagnosis of depression. CN113052113B discloses a depression recognition method and system based on a compact convolutional neural network, the method comprising: acquiring electroencephalogram data of a plurality of subjects; wherein the subjects include depression subjects and normal subjects; performing data preprocessing on the electroencephalogram data, and dividing the preprocessed electroencephalogram data into a training data set and a test data set according to a predetermined proportion; inputting the training data set into a pre-constructed compact convolutional neural network to train the compact convolutional neural network, and generating a depression recognition model when the compact convolutional neural network reaches a predetermined convergence state; inputting the test data set into the depression recognition model for recognition and outputting depression recognition results and normal recognition results, respectively. By implementing the present application, the dependence of the recognition model on data quality can be effectively reduced, and the accuracy of recognition can be improved.

[0004] CN112232191B discloses a depression recognition system based on micro-expression analysis. It belongs to the field of computer vision; the specific steps are: 1, training a deep multi-task recognition network; 2, dividing the important local area of the face, and eliminating the area irrelevant to micro-expression; 3, training an adaptive double-flow neural network to locate the start frame, Apex frame and end frame of micro-expression movement; 4, according to the analysis of micro-expression in different backgrounds, judge whether the person has depression. The present application is based on a deep multi-task neural network, pre-processes the image, divides the important local area of the face, improves the recognition speed of the double-flow neural network, and meets the real-time requirements; and through the BLSTM-CNN neural network based on attention mechanism, the important frame picture features are extracted, and the double-flow features extracted by the adaptive fusion double-flow neural network are improved to improve the positioning of the micro-expression movement frame, and then the speed and accuracy of micro-expression recognition are improved.

[0005] US20220341945A1 discloses a marker composition for diagnosing major depression, which comprises ZA2G and thrombin as markers, a method for determining information necessary for determining the occurrence of major depression using the marker composition, a composition for determining the occurrence of major depression comprising reagents for measuring the expression level of the marker, and a kit for determining the occurrence of major depression comprising a device for measuring the expression level of the marker. Among them, the combination is realized using analysis methods selected from linear or nonlinear regression analysis methods, analysis of variance, neural network analysis method, genetic analysis method, support vector machine analysis method, hierarchical clustering analysis or clustering analysis method, hierarchical algorithm using decision tree, kernel principal component analysis method, Markov blanket analysis method, recursive feature elimination or entropy-based recursive feature elimination analysis method, forward floating search or backward floating search analysis method and combination thereof. The combination is performed using a computer algorithm.

[0006] Therefore, with the development of technology, there are many artificial intelligence assisted depression diagnosis technologies at present, but these technologies have complicated input signals, are not easy to collect, or have unsatisfactory diagnosis accuracy, and there is no mature technology that can be really applied to clinical. SUMMARY

[0007] Therefore, the main purpose of the present application is to provide a depression detection model training method, detection system, program and storage medium, in order to at least partially solve the above technical problems.

[0008] In order to achieve the above purpose, as the first aspect of the present application, a training method of an artificial intelligence model for detecting depression is provided, comprising the following steps:

[0009] constructing an artificial intelligence model, an input of the artificial intelligence model comprising concentrations of 3-4 kinds of to-be-detected compounds in a to-be-detected sample, the to-be-detected compounds being selected from 5-hydroxyanthranilic acid, methionine, gamma-aminobutyric acid, 3, 4-dihydroxyphenylacetic acid, 2-pyridinecarboxylic acid, quinolinic acid, kynurenine, 5-hydroxytryptamine, 3-hydroxykynurenine, 3-hydroxyanthranilic acid, dopamine, norepinephrine, homocysteine, an output of the artificial intelligence model being a risk assessment index of a subject of the to-be-detected sample suffering from major depressive disorder;

[0010] The artificial intelligence model is trained by using large sample data, wherein healthy persons are positive samples, and persons suffering from major depressive disorder are negative samples.

[0011] As a second aspect of the present application, a major depressive disorder detection system is further provided, comprising:

[0012] A data processing unit is configured to determine a risk of a subject of a to-be-tested sample suffering from major depressive disorder based on a detection result of a detection device and an artificial intelligence model trained by the method for training an artificial intelligence model for detecting depressive disorder according to any one of claims 1-4.

[0013] An extraction reagent combination comprises calibrators, quality control samples and extraction solutions of to-be-detected compounds as biomarkers matched with an input of a pre-trained artificial intelligence model;

[0014] A detection device is configured to detect a sample of a subject treated by the extraction reagent combination to obtain concentrations of the to-be-detected compounds.

[0015] As a third aspect of the present application, an executable program is further provided, which can be executed by a computing device to implement the following method:

[0016] Input or obtain concentrations of to-be-detected compounds in a sample; wherein the to-be-detected compounds are biomarkers matched with an input of a pre-trained artificial intelligence model;

[0017] Determine a risk of a subject of the sample suffering from major depressive disorder based on the concentrations of the to-be-detected compounds and an artificial intelligence model trained by the method for training an artificial intelligence model for detecting depressive disorder.

[0018] As a fourth aspect of the present application, a storage medium is further provided, which stores the executable program as described above.

[0019] According to the above technical solutions, the model training method for detecting depressive disorder, the detection system, the program and the storage medium of the present application have at least one of the following beneficial effects relative to the prior art:

[0020] (1) Precision problem, high accuracy, high sensitivity, AUC greater than or equal to 0.95, even 0.99; The accuracy rate is more than 90%.

[0021] (2) Simple algorithm, fast iterative convergence, easy to transplant, can be embedded in the detection system, or can be used as a separate data processing unit, and works in the way of encryption key or external U disk. BRIEF DESCRIPTION OF DRAWINGS

[0022] The method and device of the present application will be further described below in combination with the drawings and examples:

[0023] Figure 1 is a flow chart of the training method of the artificial intelligence model of the present application;

[0024] Figures 2-14 is the relative content distribution (content comparison) of the thirteen biomarkers in the plasma samples of different groups of the present application;

[0025] Figures 15-34 is the ROC curve diagram of the various combinations of the thirteen preferred biomarkers of the present application for diagnosing severe depression and healthy controls. DETAILED DESCRIPTION

[0026] In order to make the purpose, technical scheme and advantages of the present application more clear and explicit, the present application will be further described in detail below in combination with specific examples and with reference to the drawings.

[0027] Some terms in the present application have the following meanings:

[0028] Depression is a common mental disorder that involves long-term low mood, loss of pleasure or interest in activities, and often affects an individual's work, study and social function.

[0029] Standard, standard article, standard content in content determination. Standard includes stoichiometric standard and drug testing standard.

[0030] Quality control product is a stable substance used to check the performance of analytical instruments or methods. Quality control product includes standard and substance similar to actual sample matrix.

[0031] Extract, chemical reagent or chemical reagent containing standard for quantitative determination; used to separate target analyte and biological matrix, reduce the interference of biological matrix on quantitative determination of target analyte.

[0032] Biomarker, generally refers to a characteristic biochemical indicator of a general physiological or pathological or therapeutic process that can be objectively measured and evaluated, through which the current process of the organism in the biological process can be known. In the present invention, the preferred is the compound, more preferably the metabolite, which is differentially present (i.e. increased or decreased) in the biological sample of the subject or group of subjects with the first phenotype (e.g. suffering from a disease) compared to the subject with the biological phenotype. The biomarker is preferably differentially present (i.e. p-value less than 0.05, determined using Welch's T test or Wilcoxon rank sum test) at a statistically significant level.

[0033] "Metabolite" or "small molecule" refers to organic and inorganic molecules present in cells. The term excludes macromolecules, such as large proteins, large nucleic acids, or large polysaccharides. Small molecules in cells are typically found in solution in the cytoplasm or other organelles (e.g. mitochondria), where they form a pool of intermediates that can be further metabolized or used to make macromolecules. The term "small molecule" includes signaling molecules and intermediates in chemical reactions that convert energy in food into usable forms. Examples of small molecules include sugars, fatty acids, amino acids, nucleotides, intermediates formed in cellular processes, and other small molecules found within cells.

[0034] Specificity, in a medical test for diagnosing a disease, refers to the ability to correctly identify those who are not diseased.

[0035] Sensitivity, in a medical test for diagnosing a disease, refers to the ability to correctly identify those who are diseased.

[0036] Kit, a box for containing chemical reagents for detecting chemical components, drug residues, virus species, etc. Some kits also include sample plates for conveniently dropping samples and presenting results. The kit is produced to enable the experimenter to get rid of the tedious reagent preparation and optimization process, so the kit generally comes with a corresponding instruction manual, and the user can get satisfactory results according to the instruction manual without or with only a small amount of optimization.

[0037] The present inventors have conducted a lot of research on the diagnosis of depression, and through retrieval, it is found that there are several metabolic pathways related to depression, and there are three or four thousand small molecule compounds involved. Although there are many articles, reports and patent results related to the detection and diagnosis method of depression, there are still few schemes that can be applied in clinical practice. At present, there are few reagent kit products for detecting depression on the market, and the existing technologies reported in the public domain are still in the state of clinical improvement and cannot enter the practical state. In addition, with the rapid development of artificial intelligence technology, the combination with various disciplines has also achieved fruitful results, but in the detection of depression, although there are many existing technologies that use artificial intelligence technology for detection, these schemes still have the defect that the diagnosis results are difficult to use in clinical practice and difficult to promote. After in-depth research on the detection method, repeated optimization of the algorithm model and clinical verification by multiple parties, the present inventors propose several model training methods, detection systems, programs and storage media for detecting depression, which can quickly and accurately diagnose whether the test object has depression through machine learning by standardized detection means, and with the increase of data samples, the specificity and accuracy are also gradually improved.

[0038] Specifically, as shown in Figure 1 The present application discloses a training method of an artificial intelligence model for detecting depression, comprising the following steps:

[0039] An artificial intelligence model is constructed, the input of the artificial intelligence model comprises the concentration of a preset to-be-detected compound in a to-be-detected sample, the preset to-be-detected compound is selected from 3-4 kinds of 5-hydroxyanthranilic acid, methionine, gamma-aminobutyric acid, 3, 4-dihydroxyphenylacetic acid, 2-picolinic acid, quinolinic acid, kynurenine, 5-hydroxytryptamine, 3-hydroxykynurenine, 3-hydroxyanthranilic acid, dopamine and norepinephrine; the output of the artificial intelligence model is the risk index of the testee from whom the to-be-detected sample is collected for suffering from severe depression;

[0040] The artificial intelligence model is trained by using large sample data, wherein the healthy people are positive samples and the patients with severe depression are negative samples.

[0041] Figures 2-14 The relative content distribution (content comparison) of the 13 biomarkers in the plasma samples of different groups is shown in the figure, and it can be seen from the figure that these biomarkers are very helpful for distinguishing and detecting patients with severe depression and healthy people as markers.

[0042] The artificial intelligence model is a vector machine model, a K-means clustering model, a decision tree model, a principal component analysis model, or a neural network model. Preferably, the artificial intelligence model / neural network model is, for example, a support vector machine analysis method, a hierarchical clustering analysis or a clustering analysis method (K-means clustering model), a decision tree hierarchical algorithm, a principal component analysis method (PCA), and the like, preferably a neural network model, for example, a convolutional neural network model (CNN), more preferably a multilayer perceptron (MLP).

[0043] In a preferred embodiment, the input of the artificial intelligence model is set to, for example, six parameters, namely the age and gender of the testee from which the sample to be detected is collected as general parameters, and the concentrations of kynurenine, 2-pyridinecarboxylic acid, 3-hydroxyanthranilic acid, and 3-hydroxykynurenine in the sample to be detected as special parameters.

[0044] The activation function of the artificial intelligence model is, for example, a Relu function.

[0045] The artificial intelligence model is, for example, adjusted in weight by using the adam chance gradient descent method to quickly converge.

[0046] The present application also discloses a major depressive disorder detection system, which is also a major depressive disorder diagnosis system or data processing system, comprising:

[0047] A data processing unit is configured to determine the risk of major depressive disorder of the testee from which the sample to be detected is collected based on the detection results of the detection device and the artificial intelligence model trained by the training method of the artificial intelligence model for detecting depression as described above.

[0048] An extraction reagent combination comprises calibration samples, quality control samples, and extraction solutions of the test compounds as biomarkers matched with the input of the pre-trained artificial intelligence model;

[0049] A detection device is configured to detect the plasma sample of the testee treated by the extraction reagent combination to obtain the concentration of the test compound.

[0050] The detection device is, for example, a chromatography, high-performance liquid chromatography (HPLC), or the like, or a combination thereof, which can accurately detect the content (concentration) of the test compound, and then determine the risk of major depressive disorder of the testee by using the pre-trained artificial intelligence model.

[0051] The extraction reagent combination can further comprise an isotopic internal standard for mass spectrometry detection, so that the content of the test compound can be accurately detected by mass spectrometry or chromatography-mass spectrometry.

[0052] The reagent combination is, for example, a kit provided to medical diagnostic personnel in the form of a kit to facilitate accurate and rapid operation and diagnosis.

[0053] The application also discloses an executable program which can be executed by a computing device to implement the method.

[0054] Input or acquire the concentration of the to-be-detected compound in the sample; wherein the to-be-detected compound is a biomarker matched with the input of the pre-trained artificial intelligence model;

[0055] Based on the concentration of the to-be-detected compound and the artificial intelligence model trained according to the training method of the artificial intelligence model for detecting depression, the risk of the tested subject from which the to-be-detected sample is collected to suffer from major depressive disorder is determined.

[0056] The computing device is, for example, selected from a single-chip microcomputer, a single-board computer, a desktop computer, a notebook computer, a server, a programmable logic controller (PLC) and a field programmable gate array (FPGA), etc. The computing device can be part of a detection device, and can perform computing function expansion through a special external interface and / or platform; or the computing device can be independent of the detection device, and can obtain or input the detection result of the detection device through wired, wireless, network or storage medium, and then calculate the risk index by using the pre-trained artificial intelligence model stored in the computing device.

[0057] The application also discloses a storage medium on which the executable program is stored. The storage medium is, for example, a floppy disk, a solid state disk, a mechanical hard disk, a U disk, a mobile hard disk, an erasable storage medium, an optical disk, a cache, an EPPRAM, etc.

[0058] The application will be further described and illustrated below by means of specific examples. It should be noted that the following examples are only illustrative and are not intended to limit the application.

[0059] For the convenience of the test, the content of the to-be-detected compound, such as 5-hydroxyanthranilic acid, methionine, gamma-aminobutyric acid, 3, 4-dihydroxyphenylacetic acid, 2-picolinic acid, quinolinic acid, kynurenine, 5-hydroxytryptamine, 3-hydroxykynurenine, 3-hydroxyanthranilic acid, dopamine, norepinephrine, homocysteine, etc. in blood will be detected simultaneously in the example part, and then the corresponding model or fitting formula will be constructed by statistical analysis or algorithm in each example.

[0060] I. The kit is taken as an example of experimental drugs in each example, and the kit comprises:

[0061] (1) Instrument test: polar end-capped C18 column, 1% formic acid (volume fraction) and 0.03% trifluoroacetic acid (volume fraction) mixed aqueous solution as mobile phase additive A; 1% formic acid (volume fraction) and 90% acetonitrile aqueous solution (1% (v / v) formic acid solution) as mobile phase additive B.

[0062] (2) Sample test: reagent for detecting 5-hydroxyanthranilic acid, methionine, gamma-aminobutyric acid, 3, 4-dihydroxyphenylacetic acid, 2-picolinic acid, quinolinic acid, kynurenine, 5-hydroxytryptophan, 3-hydroxykynurenine, 3-hydroxyanthranilic acid, dopamine, norepinephrine, homocysteine in plasma.

[0063] ① Calibration: mixed solution containing 5-hydroxyanthranilic acid, methionine, gamma-aminobutyric acid, 3, 4-dihydroxyphenylacetic acid, 2-picolinic acid, quinolinic acid, kynurenine, 5-hydroxytryptophan, 3-hydroxykynurenine, 3-hydroxyanthranilic acid, dopamine, norepinephrine, homocysteine;

[0064] ② Quality control: plasma solution containing 5-hydroxyanthranilic acid, methionine, gamma-aminobutyric acid, 3, 4-dihydroxyphenylacetic acid, 2-picolinic acid, quinolinic acid, kynurenine, 5-hydroxytryptophan, 3-hydroxykynurenine, 3-hydroxyanthranilic acid, dopamine, norepinephrine, homocysteine;

[0065] ③ Internal standard solution: 5% methanol aqueous solution containing 5-hydroxyanthranilic acid-D5, methionine-D3, gamma-aminobutyric acid-D3, 3, 4-dihydroxyphenylacetic acid-D3, 2-picolinic acid-D4, quinolinic acid- 13 C3 15 N, kynurenine- 13 C6, 5-hydroxytryptophan-D4, 3-hydroxykynurenine- 13 C6, 3-hydroxyanthranilic acid-D3, dopamine-D4, norepinephrine-D5, homocysteine-D4;

[0066] ④ Extract: 2% dithiothreitol (DTT) with mass number and 15-30% 5-sulfosalicylic acid aqueous solution with volume fraction;

[0067] ⑤ Diluent: 0.2% formic acid (volume fraction) and 0.006% trifluoroacetic acid (volume fraction) mixed solution.

[0068] Of course, the kit form can not be used, but the reagents purchased or prepared by oneself can also be used as long as the purpose of the present application can be achieved.

[0069] II. Detection process

[0070] (1) Experimental subjects

[0071] Through cooperation with two city-level hospitals with certain geographical distance in Jiangsu, the following experimental data were collected and collated. Among the total of 400 subjects, the age range was 18-66 years old, including 150 cases of depressive patients (MDD) and 120 healthy controls (HC). The detailed information of the subjects is shown in Table 1:

[0072] Table 1 Detailed information of subjects

[0073]

[0074] a: Chi-square test; b: K-W test

[0075] It should be noted that since depression is still diagnosed by means of scales and other means, there is no gold standard for instrument detection at present, so only severe patients with clear manifestations are selected for the above-mentioned depressive patients to ensure the accuracy of the results. In future research, the study of mild and moderate patients will be gradually carried out to explore more scientific, objective and gold standard detection means.

[0076] In addition, the above-mentioned detection has informed the tested subjects who collected the samples and obtained their consent.

[0077] (2) Sample collection

[0078] Collect venous blood into vacuum blood collection tubes containing anticoagulant. Slowly invert and mix evenly, place at 2-8℃, complete centrifugal separation of plasma within 30 min, transfer the plasma to EP tubes, and store at -80℃ for standby use.

[0079] (3) Sample pretreatment

[0080] Accurately take 50 μL of plasma into a 1.5 mL EP tube, add 20 μL of internal standard, then add 8 μL of 30% (volume fraction) 5-sulfosalicylic acid solution containing 2% (mass fraction) DTT, high-frequency oscillate for 5 min for protein precipitation; centrifuge at 4℃, 11000 rpm for 10 min; take 50 uL of supernatant and place it in a 96-well plate, then add 50 μL of 0.2% formic acid water (volume fraction) and 0.015% trifluoroacetic acid (volume fraction) mixed solution to each empty position, low-frequency oscillate for 2 min, and detect on the machine.

[0081] (4) Ultra-high performance liquid chromatography-mass spectrometry detection parameter setting

[0082] Separation was performed on a Thermofisher aQ column, column temperature was 35℃, injection volume: 5uL, mobile phase A: 0.0075% trifluoroacetic acid-0.1% formic acid water; mobile phase B: 0.1% formic acid water 90% acetonitrile water; mobile phase gradient was set as shown in Table 2:

[0083] Table 2 mobile phase gradient setting

[0084]

[0085] Mass spectrometry device parameter settings are shown in Table 3:

[0086] Table 3 Mass spectrometry device parameter settings

[0087]

[0088] (5) Data analysis and diagnosis

[0089] A. Based on the software (such as MassLynx software) provided with the liquid chromatograph tandem mass spectrometer, a calibration curve is established, and the corresponding factors of each compound in the on-machine liquid are substituted into the linear equation to output the content of each compound in the plasma;

[0090] B. For the combined index, the content of a single compound is introduced into the self-built system, and the combined index is calculated according to the previously constructed data model;

[0091] (6) Actual measurement data

[0092] The reagent is used to distinguish between healthy controls and severe depression; the content of the biomarker in the blood of 120 healthy people and 150 severe depression patients is detected, and single factor variance analysis is performed to evaluate the difference in the content of the selected biomarker in healthy people and severe depression patients; multiple logistic regression is used to establish a diagnostic model, and ROC analysis is further performed to quantify the diagnostic performance of the combination containing multiple selected biomarkers (metabolites).

[0093] Example 1

[0094] The present embodiment discloses a reagent combination for detecting depression, which comprises calibration samples, quality control samples, extraction solutions, dilution solutions, internal standard samples containing 3-hydroxyanthranilic acid-D3 kynurenine-quinolinic acid- 13 C6, quinolinic acid- 13 C3 15 N, mobile phase additive.

[0095] The blood sample of the testee is detected by the above-mentioned extraction reagent combination, the plasma in the blood sample is extracted, and the contents of 3-hydroxyanthranilic acid, kynurenine and quinolinic acid in the plasma are detected by liquid chromatography tandem mass spectrometry. The experimental results, the detection accuracy for depression, and the difference data are shown in Table 5.

[0096] Examples 2-19

[0097] The specific reagent composition, test instrument, test process, etc. of Example 2-19 are the same as those of Example 1, and the only difference is that the biomarkers selected therein are shown in Table 4 as follows.

[0098] The experimental results, the detection accuracy for depression, and the difference data are shown in Table 5.

[0099] Table 4: Selected biomarkers in each test example

[0100]

[0101] Table 5: Experimental detection results of each test example

[0102]

[0103] Among them: p<0.05 p<0.01

[0104] Figures 15-34 is the ROC curve diagram of the various combinations of the thirteen preferred biomarkers of the present application for diagnosing major depressive disorder and healthy controls. As shown in Figures 15-34 and Table 5, the above-mentioned Examples 1-19 are used to illustrate that 3-4 biomarkers are selected from the preferred biomarkers of the present application, good specificity and sensitivity can be obtained, the data is convergent, and can be well fitted by a multiple linear, quadratic, exponential, etc. fitting formula, and is also suitable for clustering / classification by an artificial intelligence model.

[0105] In order to further prove the advantages of the present application, the present application selects the four biomarkers of Example 7, kynurenine, 2-pyridinecarboxylic acid, 3-hydroxyanthranilic acid, and 3-hydroxykynurenine, and performs an artificial intelligence model parameter optimization test.

[0106] Example 20

[0107] 1. Collect the venous blood of the patient, and collect the age, gender, etc. information of the patient, which is defined as general parameters;

[0108] 2. The tester processes the patient's blood sample and collects data through integrated equipment to obtain the results of four specific indicators, which are defined as specific parameters; the specific indicators are kynurenine, 2-pyridinecarboxylic acid, 3-hydroxy-o-aminobenzoic acid, and 3-hydroxykynurenine.

[0109] 3. Preprocess datasets with general and specific parameters and map them to low-dimensional latent space vectors. Data preprocessing includes handling missing values ​​and standardization.

[0110] (1) Handling missing values

[0111] For missing values, a deletion strategy is adopted to ensure data integrity;

[0112] (2) Standardization (normalization) processing

[0113] The data for the above six indicators are standardized and adjusted to the same scale.

[0114] 4. Select the Multilayer Perceptron (MLP) data model. The parameter descriptions and settings are as follows:

[0115] Table 6. Parameter Description and Settings for the Multilayer Perceptron (MLP) Data Model

[0116]

[0117] 5. Evaluate the effectiveness of the data model and calculate the AUC index of the ROC curve.

[0118] Examples 21-25

[0119] The only difference between the different embodiments lies in the artificial intelligence algorithms used. The specific detection performance of the artificial intelligence models in embodiments 21-25 is shown in Table 7 below.

[0120] Table 7. Specific detection performance of the artificial intelligence models in Examples 21-25

[0121]

[0122] Therefore, by selecting appropriate inputs and outputs, the artificial intelligence model of this invention can adopt various algorithm types, such as multilayer perceptron (MLP), support vector machine analysis, KNN, Naive Bayes, logistic regression, etc., and can obtain satisfactory AUC values. It is expected to be further used in clinical testing and trained and optimized with a larger amount of large sample data.

[0123] The above-described specific embodiments further illustrate the purpose, technical solutions and beneficial effects of the present application, and it should be understood that the above-described is only a specific embodiment of the present application and is not intended to limit the present application, and any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present application should be included in the protection scope of the present application.

Claims

1. A depression detection system, characterized in that, include: A data processing unit is used to determine the risk of major depressive disorder (MDD) in the subjects from whom the test samples are collected, based on the detection results of the detection equipment and the training method of the artificial intelligence model for detecting depression. The artificial intelligence model is trained through the following steps: constructing an artificial intelligence model, the input of which includes the concentrations of 3-4 compounds selected from 5-hydroxy-an-aminobenzoic acid, methionine, γ-aminobutyric acid, 3,4-dihydroxyphenylacetic acid, 2-pyridinecarboxylic acid, quinolinic acid, kynurenine, 5-hydroxytryptophan, 3-hydroxykynurenine, 3-hydroxy-an-aminobenzoic acid, dopamine, norepinephrine, and homocysteine ​​in the test sample; the output of the artificial intelligence model is a risk assessment index for major depressive disorder in the subjects from whom the test samples are collected; training the artificial intelligence model using large sample data, the artificial intelligence model being a multilayer perceptron model, where healthy individuals are positive samples and individuals with major depressive disorder are negative samples; the input of the artificial intelligence model also includes the age and gender of the subjects from whom the test samples are collected. The activation function of the artificial intelligence model is the ReLU function; the artificial intelligence model uses the Adam opportunistic gradient descent method for weight adjustment; and the data model is evaluated by calculating the ROC curve and AUC index. An extraction reagent assemblies comprising calibrators, quality control samples, and extracts of a target compound serving as a biomarker, matched to the input of a pre-trained artificial intelligence model; the selected biomarker is chosen from any of the following combinations: 3-Hydroxy-anaminobenzoic acid, kynurenic acid, quinolinic acid; 2-Pyridinecarboxylic acid, 3-hydroxy-o-aminobenzoic acid, quinoline acid; 5-Hydroxytryptophan, 2-pyridinecarboxylic acid, 3-hydroxy-an-aminobenzoic acid; 3-Hydroxykynurenine, 2-pyridinecarboxylic acid, 5-hydroxytryptophan; 2-Pyridinecarboxylic acid, 3-hydroxy-an-aminobenzoic acid, 3-hydroxykynurenine; Quinolinic acid, 3-hydroxykynurenine, 2-pyridinecarboxylic acid; 5-Hydroxytryptophan, 2-pyridinecarboxylic acid, 3-hydroxy-an-aminobenzoic acid, kynurenine; 2-Pyridinecarboxylic acid, 3-hydroxy-an-aminobenzoic acid, 3-hydroxykynurenine, homocysteine; 5-Hydroxytryptophan, homocysteine, 2-pyridinecarboxylic acid, 3-hydroxy-an-aminobenzoic acid; 3,4-Dihydroxyphenylacetic acid, 3-hydroxy-o-aminobenzoic acid, quinoline acid; γ-aminobutyric acid, 3,4-dihydroxyphenylacetic acid, 3-hydroxy-o-aminobenzoic acid; Homocysteine, quinolinic acid, 3-hydroxy-an-aminobenzoic acid; Homocysteine, quinolinic acid, 3-hydroxy-2-aminobenzoic acid, 5-hydroxy-2-aminobenzoic acid; 3-Hydroxy-o-aminobenzoic acid, dopamine, 5-Hydroxy-o-aminobenzoic acid; Homocysteine, 3-hydroxy-an-aminobenzoic acid, kynurenine; Kyrenine, 2-pyridinecarboxylic acid, norepinephrine; 3-Hydroxy-o-aminobenzoic acid, dopamine, 2-pyridinecarboxylic acid; Methionine, 3,4-dihydroxyphenylacetic acid, 2-pyridinecarboxylic acid; 3-Hydroxy-anaminobenzoic acid, 3,4-dihydroxyphenylacetic acid, methionine; The detection device is used to detect the sample of the test subject after it has been treated with the above-described extraction reagent combination to obtain the concentration of the compound to be detected.

2. The depression detection system according to claim 1, characterized in that, The extraction reagent combination also includes an isotopic internal standard for mass spectrometry detection; and / or The extraction reagent combination is a kit.

3. The depression detection system according to claim 1, characterized in that, The detection equipment is a chromatography, high performance liquid chromatography, mass spectrometry, or chromatography-mass spectrometry detection equipment.

4. A storage medium, characterized in that, The storage medium stores a method executable by a computing device, the method being as follows: Input or obtain the concentration of the compound to be detected in the sample; wherein the compound to be detected is a biomarker matched with the input of a pre-trained artificial intelligence model; Based on the concentration of the compound to be detected and the artificial intelligence model trained by the training method for detecting depression as described in claim 1, the risk of the test subject from whom the sample was collected suffering from major depressive disorder is determined.

5. The storage medium according to claim 4, characterized in that, The computing device is selected from microcontrollers, single-board computers, desktop computers, laptops, servers, programmable logic controllers (PLCs), and field-programmable gate arrays (FPGAs).

Images