Method and system for establishing disease prediction model
By extracting multi-level features from bacterial phase data and training disease prediction models, the shortcomings of intestinal bacterial phase disease prediction in the prior art are solved, the accuracy of disease prediction and ability to understand pathogenic mechanisms are improved, and the treatment and Development of prevention methods.
Patent Information
- Application Number
- CN202310751073.3
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2023-06-20
- Publication Date
- 2024-12-20
AI Technical Summary
The prior art lacks effective disease prediction models, especially in the case of disease prediction based on intestinal bacterial phases, and it is difficult to accurately predict diseases and understand pathogenic mechanisms.
By extracting bacterial species-level, interactive growth and cluster-level features from bacterial phase data from multiple samples, selecting key features using feature screening models, and training disease prediction models to improve the accuracy of disease prediction and understand pathogenic mechanisms. .
It improves the accuracy of disease prediction and the richness of characteristic information, helps to understand the pathogenesis of the disease, and helps to develop more effective treatment and prevention methods.
Smart Images
Figure CN119170241A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to feature engineering technology, and in particular to a method and system for establishing a disease prediction model. Background Art
[0002] "Bacteria phase" refers to the sum of microorganisms such as bacteria, viruses, fungi and yeast living in the human body. Together, they form a complex ecosystem that allows the body to function normally. Like other ecosystems, if one species is damaged, it may impact the ecological balance. The most complex flora in the human body is the intestinal flora, and the intestinal flora has a significant impact on human health. It not only affects intestinal function, but also further affects human health and body balance by participating in human metabolism and regulating the immune system. Many diseases are related to the imbalance of intestinal flora, such as brain neurodegeneration, cardiovascular disease, diabetes, autoimmune diseases and cancer. For example, animal and human experiments have pointed out that "Fusobacterium nucleatum" may be a causative factor of colorectal cancer. In experiments, using antibiotics to kill Fusobacterium nucleatum can reduce the occurrence of tumors. However, due to the high complexity of the intestinal flora, there is currently a lack of solutions for disease prediction based on the intestinal flora.
[0003] Therefore, a method and system for establishing a disease prediction model is needed, wherein the established disease prediction model can effectively predict the disease by using bacterial phase characteristics, so as to serve as a warning before the disease occurs and further understand its pathogenic mechanism. Summary of the invention
[0004] An embodiment of the present invention provides a method for establishing a disease prediction model executed by a computer system. The method comprises the following steps: for each of a plurality of samples, extracting feature values of a plurality of bacterial phase characteristics from the bacterial phase data of the sample; selecting a plurality of bacterial phase characteristics from the plurality of bacterial phase characteristics as selected features; and training a disease prediction model. Each training data used to train the disease prediction model comprises (1) the disease data of each sample and (2) the feature value of the selected feature of the sample. The plurality of bacterial phase characteristics comprises bacterial species level features, interactive growth and decline features, and bacterial flora level features.
[0005] In one embodiment, the bacterial species level features include relative abundance data and presence / absence data of each bacterial species.
[0006] In one embodiment, the mutual fluctuation feature includes a hierarchical ratio between two taxa at a taxonomic level.
[0007] In one embodiment, the flora-level features include a Beta diversity matrix.
[0008] In one embodiment, a plurality of selected features are selected from a plurality of bacterial phase features, comprising: inputting disease data and a plurality of bacterial phase features into a plurality of feature screening models to obtain a plurality of feature pools; sorting the bacterial phase features based on the number of times each bacterial phase feature is selected into the feature pools by the feature screening models to obtain a feature ranking; and based on the feature ranking, selecting a specified number of bacterial phase features from the plurality of bacterial phase features as the selected features.
[0009] The embodiment of the present invention provides a system for establishing a disease prediction model. The system includes a storage device and a processing device. The storage device stores disease data and bacterial phase data of multiple samples. The processing device loads a program from the storage device to execute each step of the above method.
[0010] The method and system for establishing a disease prediction model provided by the present disclosure improves the richness and representativeness of feature information by establishing bacterial phase features from multiple aspects of microorganisms. This not only helps to improve the accuracy of disease prediction, but also helps to explain the analysis results. In addition, it can help humans to have a deeper understanding of the pathogenic mechanism of the disease and help to develop more effective treatment and prevention methods. BRIEF DESCRIPTION OF THE DRAWINGS
[0011] The present disclosure will be better understood from the following illustrative embodiments with accompanying drawings. In addition, it should be understood that in the flowchart of the present disclosure, the execution order of each block may be changed, and / or some blocks may be changed, deleted or merged.
[0012] Figure 1A is a flow chart of a method for establishing a disease prediction model according to an embodiment of the present invention.
[0013] Figure 1B Corresponding to Figure 1A An embodiment of the present invention is a schematic diagram of a method for establishing a disease prediction model.
[0014] Figure 2 It is a flowchart of an embodiment of the feature screening stage.
[0015] Figure 3 Corresponding to Figure 2 An embodiment of the present invention is a schematic diagram showing multiple feature screening models and feature pools.
[0016] Figure 4 The present invention is a system block diagram of a system for establishing a disease prediction model according to an embodiment of the present invention.
[0017] The reference numerals are described as follows:
[0018] 100: Methods
[0019] S101-S103: Steps
[0020] P101: Feature extraction stage
[0021] P102: Feature screening stage
[0022] P103: Model training phase
[0023] 10: Bacteria data
[0024] 11: Fungal characteristics
[0025] 11A: Characteristics at the strain level
[0026] 11B: Interactive growth and decline characteristics
[0027] 11C: Characteristics at the flora level
[0028] 12: Training Materials
[0029] 14: Selected Features
[0030] 15: Disease data
[0031] 16: Disease prediction model
[0032] S201-S203: Steps
[0033] M(1)-M(N): Feature screening model
[0034] FP(1)-FP(N): Feature Pool
[0035] 400: System
[0036] 401: Processing Device
[0037] 402: Storage Device DETAILED DESCRIPTION
[0038] The following description lists various embodiments of the present invention, but is not intended to limit the content of the present invention. The actual scope of the invention is defined by the claims.
[0039] In the various embodiments listed below, the same reference numerals will be used to represent the same or similar elements or components.
[0040] Serial numbers in this specification and the claims, such as "first", "second", etc., are only for the convenience of explanation and have no sequential relationship with each other.
[0041] The following description of the embodiments of the apparatus or system is also applicable to the embodiments of the method, and vice versa.
[0042] Figure 1AFIG. 1 is a flow chart of a method 100 for establishing a disease prediction model according to an embodiment of the present invention. Figure 1A As shown, the method 100 includes steps S101-S103. Figure 1B Corresponding to Figure 1A , a schematic diagram of method 100 is shown. Figure 1B As shown, method 100 includes a feature extraction stage P101, a feature selection stage P102, and a model training stage P103. Figure 1A and Figure 1B , for a better understanding of this embodiment.
[0043] Figure 1A Step S101 in is equivalent to Figure 1B In step S101 and the feature extraction stage P101, feature values of multiple bacterial phase features 11 are extracted from the bacterial phase data 10 of each sample. Then, the method 100 proceeds to step S102.
[0044] like Figure 1B As shown, the microbial features 11 include three types: species-level features 11A, microbiota interaction features 11B, and community-level features 11C, each of which includes one or more microbial features.
[0045] Further explanation, in the embodiments of the present disclosure, the "sample" may refer to various specimens collected from the human body or animal body, such as blood, urine, saliva, feces, tissue, etc., which can be used for further testing, detection or diagnosis. Each sample has corresponding bacterial phase data 10 and disease data 15, which respectively describe the bacterial phase of the specimen and the disease type of the subject (i.e., the source of the specimen, human or animal). Since the intestinal flora is one of the most complex flora in the human body and has a significant impact on human health, in the embodiments of the present disclosure, the human intestinal flora is often used as an example of the bacterial phase data 10, but the present disclosure is not limited to this.
[0046] In the example of human intestinal flora, the flora data 10 can be collected by testing human fecal samples or by gastrointestinal endoscopy. Disease data 15 can be collected using various medical testing methods, such as blood tests, imaging examinations, etc. However, the present disclosure does not limit the collection methods of flora data 10 and disease data 15. Other practices, such as blood tests, urine tests, or tissue biopsies, can also be used to collect relevant data. In addition, the present disclosure does not limit the types of diseases that can be included in the disease data 15. Depending on the specific application and needs, the types of diseases in the disease data 15 may include various diseases, such as infections, chronic diseases, and cancer. Assuming that the subject is very healthy, the disease data 15 may also indicate that the subject does not have any disease. Alternatively, the disease data 15 may reflect the overall health of the subject, such as the goodness of metabolic function or the immune system, but the present disclosure is not limited to this.
[0047] In one embodiment, the bacterial phase data 10 may be represented by the number or abundance information of various bacterial species in the sample. It is worth noting that the present disclosure does not limit the bacterial phase data 10 to include all bacterial species on the sample. In one embodiment, the bacterial species included in the bacterial phase data 10 may be determined based on the prevalence of the bacterial species in the sample. For example, the bacterial phase data 10 may only consider bacterial species with a prevalence of more than 10% in the sample. However, the present disclosure does not limit the selection of bacterial species in the bacterial phase data 10.
[0048] In one embodiment, the bacterial species-level feature 11A provides more detailed information about each bacterial species, including relative abundance data and presence / absence data of each bacterial species.
[0049] Relative abundance data indicates the relative number of a bacterial species in a sample, and can usually be represented by the ratio of the DNA sequence of the bacterial species in the sample. For example, if there are 10 million bacterial cells in the sample, of which 1 million are species A and 1.5 million are species B, then the relative abundance value of species A is 100 / 1000=0.1, and the relative abundance value of species B is 150 / 1000=0.15. Therefore, the relative abundance value of species A, 0.1, and the relative abundance value of species B, 0.15, can be used as two feature values of species-level feature 11A.
[0050] The presence or absence data indicates whether a bacterial species exists in a sample, and can usually be expressed in binary form. For example, if bacterial species C exists in a sample and bacterial species D does not exist in the sample, the presence or absence data value of bacterial species C is 1, and the presence or absence data value of bacterial species D is 0. Therefore, the presence or absence data value of bacterial species C, 1, and the presence or absence data value of bacterial species D, 0, can be used as two feature values of the bacterial species level feature 11A.
[0051] In one embodiment, the mutual increase / decrease feature 11B includes a hierarchical ratio between two taxonomic units at a taxonomic level.
[0052] The stratification ratio is an indicator used to describe the relationship between the growth and decline of two bacterial species. Usually, all bacterial species are divided according to the taxonomic hierarchy such as phylum, class, order, family, genus, and species. Then, the relative abundance ratio between the two bacteria in each hierarchy is calculated to describe the composition structure of the bacterial community at each level in the sample. Assume that at a specific hierarchy (H Hierarchy ) contains two taxa: Taxon A and Taxon B. Taxon A has m species, and their relative abundances are X1, X2, X3…X m ; Taxon B has n species, and their relative abundances are Y1, Y2, Y3…Y n , then the calculation formula of the stratification ratio is as follows:
[0053]
[0054] The following provides an example of the bacterial profile data of subject 01.
[0055]
[0056]
[0057] As shown in Table 1, Phylum A contains three bacterial species, a, b, and c, and the relative abundance of the three species is 0.1. Species a and b belong to Class A under Phylum A, and species c belongs to Class B under Phylum A. Phylum B contains four bacterial species, d, e, f, and g, and their relative abundances are 0.1, 0.1, 0.2, and 0.3, respectively, and all four belong to Class C under Phylum B. The aforementioned "classification level" can be any one of phylum, class, order, family, and genus, and the interactive growth and decline feature 11B can include a stratification ratio obtained from one or more classification levels. The following will provide an example of calculating the stratification ratio using the two classification levels of "phylum" and "class".
[0058] In terms of the classification level of "phylum", the stratification ratio of Phylum A to Phylum B is the ratio of the sum of the phyla of Phylum A to the sum of the phyla of Phylum B in , which is calculated as follows:
[0059]
[0060] In terms of the classification level of "class", the stratification ratio between Class A, Class B and Class C is calculated based on the sum of the classes of Class A, Class B and Class C in , as follows:
[0061]
[0062]
[0063]
[0064] The four values of the stratification ratio calculated above are It can be used as one of the four eigenvalues of the interactive growth and decline feature 11B.
[0065] In one embodiment, the flora level feature 11C includes a Beta Diversity Matrix. The Beta Diversity Matrix is a mathematical tool used to compare the similarities or differences between different floras (i.e., overall floras). Each element in the matrix represents the degree of difference between two floras, usually measured by distance or similarity.
[0066] In one embodiment, the Beta diversity matrix can use Jaccard similarity as a measurement indicator. Jaccard similarity is based on the aforementioned presence or absence value, and calculates the ratio of the number of common bacterial species in the two bacterial communities to the total number of bacterial species, and its value is between 0 and 1. The closer the value is to 1, the more common bacterial species there are between the two samples, and the smaller the difference in bacterial phases. Conversely, the closer the value is to 0, the fewer common bacterial species there are between the two samples, and the greater the difference in bacterial phases. In practice, the bacterial phase data 10 of each sample can be used as a reference benchmark in turn to calculate the bacterial phase differences between other samples and the reference sample. In addition to using samples as reference objects, pre-defined bacterial phases (such as the bacterial phases of diabetic patients) can also be used as reference benchmarks. The calculation formula for Jaccard similarity is as follows:
[0067]
[0068] Among them, |X| and |Y| represent the number of bacterial species in each of the two bacterial colonies, |X∪Y| represents the total number of bacterial species in the two bacterial colonies, and |X∩Y| represents the number of bacterial species in common.
[0069] The following provides examples of the disease categories and the presence / absence values of each bacterial species for Subject 01-05.
[0070]
[0071] Disease Type Subjects Bacteria A Bacteria Bacteria C 0 Subject 01 0 1 1 0 Subject 02 0 1 0 0 Subject 03 1 1 1 1 Subject 04 0 0 0 1 Subject 05 1 0 1
[0072] First, taking Subject 01 as the reference, the Jacquard similarity of Subject 02 relative to Subject 01 is calculated as follows:
[0073]
[0074] Among them, |Subject 01| and |Subject 02| represent the number of bacterial species in the samples of subject 01 and subject 02 respectively, and |Subject 01∩Subject 02| represents the number of bacterial species shared by the samples of subject 01 and subject 02. According to , it can be seen that the sample of subject 01 has two bacterial species (bacterial species B and bacterial species C), so |Subject 01|=2; the sample of subject 02 has only one bacterial species (bacterial species B), so |Subject02|=1; the samples of subject 01 and subject 02 have one bacterial species (bacterial species B), so |Subject 01∩Subject 02|=1.
[0075] According to the calculation method introduced in the previous paragraph, the Jacquard similarity values between all subjects Subject 01-05 can be calculated successively. The Beta diversity matrix composed of these values can be used as the eigenvalue of the flora level characteristics.
[0076] In another embodiment, the Beta diversity matrix can use Bray-Curtis similarity as a measurement indicator. Bray-Curtis similarity is based on the aforementioned relative abundance value, and is calculated by dividing the absolute value of the difference in relative abundance of each species in the two bacterial flora by the sum of the relative abundance of each species. Its value is between 0 and 1. The closer the value is to 1, the smaller the difference in bacterial phases between the two samples, and the closer the value is to 0, the greater the difference in common bacterial phases between the two samples. The calculation formula of Bray-Curtis similarity is as follows:
[0077]
[0078] Among them, M i1 With M i2 They represent the relative abundance of the i-th bacterial species in the two samples, and S represents the total number of bacterial species.
[0079] Please refer to Figure 1A and Figure 1B . Figure 1A Step S102 in is equivalent to Figure 1B In step S102 and the feature screening stage P102, a plurality of bacterial phase features are selected from the plurality of bacterial phase features as selected features 14. Then, the method 100 proceeds to step S103.
[0080] In one embodiment, in step S102 and the feature screening stage P102, a feature screening model is used to select the selected feature 14. The feature screening model can be implemented using various statistical models or machine learning models. The statistical model can, for example, use the least absolute shrinkage and selection operator (Lasso) algorithm, stepwise logistic regression, statistical test, etc. The machine learning model can, for example, use a decision tree, logistic regression, naive Bayes, random forest, support vector machine (SVM), fully-connected neural network, etc., and the present disclosure is not limited to this.
[0081] Figure 2 yes Figure 1A FIG. 1 is a flowchart of an embodiment of step S102 in FIG. 1 . In this embodiment, multiple feature screening models are used to select features. Figure 2 As shown, step S102 may further include steps S201-S203. Figure 3 Corresponding to Figure 2 The embodiment of FIG. 1 shows a schematic diagram of multiple feature screening models M(1)-M(N) and feature pools FP(1)-FP(N). Please refer to FIG. Figure 2 and Figure 3 , for a better understanding of this embodiment.
[0082] In step S201, the disease data 10, as well as various bacterial features such as bacterial species level features 11A, interactive growth and decline features 11B and bacterial flora level features 11C are input into N feature screening models M(1)-M(N) to obtain N feature pools FP(1)-FP(N). Then, the process proceeds to step S202.
[0083] like Figure 3 As shown, each of the feature screening models M(1)-M(N) selects one or more bacterial phase features from multiple bacterial phase features such as bacterial species level features 11A, interactive growth and decline features 11B, and bacterial flora level features 11C to form a corresponding feature pool. Figure 3In the example, feature screening model M(1) outputs feature pool FP(1), feature screening model M(2) outputs feature pool FP(2), and so on. Different feature screening models M(1)-M(N) will screen out different features, so the bacterial phase features in feature pools FP(1)-FP(N) are not the same.
[0084] Further explanation, the feature screening model M(1)-M(N) can be implemented by various statistical models or machine learning models. The statistical model can be, for example, the least absolute shrinkage and selection operator (Lasso) algorithm, stepwise logistic regression, statistical test, etc. The machine learning model can be, for example, a decision tree, logistic regression, naive Bayes, random forest, support vector machine (SVM), fully-connected neural network, etc., and the present disclosure is not limited to this.
[0085] In one embodiment, each of the feature screening models M(1)-M(N) calculates at least one statistical index for each bacterial phase feature, and compares the statistical index with the corresponding critical value to determine whether to select this bacterial phase feature to the corresponding feature pool. The critical value is a pre-set fixed value, which depends on the feature screening model itself. The statistical index can be, for example, a P value, an odds ratio, a correlation coefficient, a fold change, etc.
[0086] In one embodiment, the number of features to be selected into the corresponding feature pool FP(1)-FP(N) can be determined based on the accuracy of each feature screening model M(1)-M(N). For example, a scree plot or an elbow method can be used to determine the number of bacterial phase features selected by the feature screening model M(1)-M(N) into the feature pool FP(1)-FP(N). In the feature curve graph drawn when using the scree plot or the elbow method, the horizontal axis usually represents the number of features used by the model, and the vertical axis represents the performance index of the model, such as accuracy or area under the curve (AUC). The optimal number of features depends on the change in the performance index. Conceptually, when the number of features increases, the performance index will increase accordingly, but after a certain number of features, if the improvement rate of the performance index slows down significantly, then the number of features can be selected as the optimal number of features.
[0087] Please refer to Figure 2 In step S202, based on the number of times each bacterial phase feature is selected into the feature pool FP(1)-FP(N) by the feature screening models M(1)-M(N), the bacterial phase features are sorted to obtain a feature ranking. Then, the process proceeds to step S203.
[0088] The aforementioned "number of times each bacterial phase feature is selected into the feature pool FP(1)-FP(N) by the feature screening model M(1)-M(N)" is equivalent to the number of feature pools in the feature pools FP(1)-FP(N) that contain the bacterial phase feature. The larger the number, the higher the importance or reference value of the bacterial phase feature. Therefore, the bacterial phase feature that ranks higher in the feature ranking has a higher importance or reference value.
[0089] In step S203, based on the feature ranking, a specified number of bacterial phase features are selected from a plurality of bacterial phase features as selected features. The specified number can be a predefined value or can be determined according to the accuracy of the feature screening model. Specifically, multiple test data can be used to obtain multiple accuracy rates of each feature screening model when different numbers of bacterial phase features are selected. The test data is only used to detect the accuracy of the feature screening model. After obtaining the accuracy of each feature screening model when different numbers of bacterial phase features are selected, one of the feature screening models is selected based on these accuracy rates. For example, the feature screening model with the highest accuracy rate can be selected. Assuming that 10 different numbers of bacterial phase features are selected to test the accuracy rate of the feature screening model, a feature screening model will obtain 10 accuracy rates. X feature screening models will have 10X accuracy rates. Therefore, the highest accuracy rate can be found among the 10X accuracy rates to select the feature screening model with the highest accuracy rate. Then, based on the number and accuracy rate of the selected feature screening models, the above-mentioned steep slope diagram or elbow method can be used to determine the specified number.
[0090] Although in Figure 3 In the illustrated embodiment, the strain-level feature 11A, the interactive growth and decline feature 11B, and the flora-level feature 11C are input into the feature screening model M(1)-M(N) together to obtain the feature pools FP(1)-FP(N), but the present disclosure is not limited thereto. In another embodiment, the strain-level feature 11A, the interactive growth and decline feature 11B, and the flora-level feature 11C can be input into the feature screening model M(1)-M(N) separately to obtain N*3 feature pools. The bacterial phase features in these feature pools will be integrated to generate a feature set, and then the features corresponding to the strain-level feature 11A, the interactive growth and decline feature 11B, and the flora-level feature 11C can be selected from the feature set in a similar manner to S202-S203 as the selected features 14.
[0091] Please refer to Figure 1A and Figure 1B . Figure 1A Step S103 in is equivalent to Figure 1B In step S103 and the model training stage P103, the disease prediction model 16 is trained. The trained disease prediction model can receive the above-mentioned selected features of the subject as input data and output the result of disease prediction. Depending on the purpose of disease prediction, the disease prediction model 16 can be a classification model that outputs the type of disease, or a regression model that outputs a metabolic function index, an immunity index or the probability of a specific disease, which is not limited by the present disclosure.
[0092] To further illustrate, each piece of training data 12 used to train the disease prediction model 16 includes (1) the disease data 15 of each sample and (2) the feature value of the selected feature 14 of the sample. During the model training stage P103, a loss function such as mean square error (MSE), mean absolute error (MAE) or cross-entropy can be used to calculate a loss value representing the difference between the prediction result output by the disease prediction model 16 and the actual disease data 15. Furthermore, an optimizer can be used to recursively adjust the parameters of the disease prediction model 16 so that the loss value converges, thereby optimizing the disease prediction model 16. The optimizer can be implemented using algorithms such as gradient descent (GD), stochastic gradient descent (SGD) or adaptive moment estimation (Adam), but the present disclosure is not limited thereto. Taking the optimizer using the gradient descent method as an example, the gradient can be obtained by performing partial derivative calculation on the loss function, and then the parameters of the disease prediction model 16 are adjusted according to the gradient to reduce the loss value. Through the training process of repeated result feedback and parameter update, the loss value is gradually reduced until the loss value converges to the minimum value.
[0093] Figure 4 FIG. 4 is a system block diagram of a system 400 for establishing a disease prediction model according to an embodiment of the present invention. Figure 4 As shown, the system 400 includes a processing device 401 and a storage device 402 .
[0094] System 400 may be a personal computer (such as a desktop computer or a laptop computer) or a server computer running an operating system (such as Windows, Mac OS, Linux, UNIX, etc.), or a mobile device such as a tablet computer or a smart phone, but the present disclosure is not limited thereto.
[0095] The processing device 401 may include one or more hardware components for executing instructions, such as a central processing unit (CPU), a graphics processing unit (GPU), a microprocessor, a controller, a microcontroller, an application specific integrated circuit (ASIC), a field programmable gate array (FPGA), a single chip system (SoC), etc., and the present disclosure is not limited thereto. In an embodiment of the present disclosure, the processing device 401 loads a program from the storage device 402 to execute steps S101-S103 of the method 100.
[0096] The storage device 402 may be any device having a non-volatile memory (such as a read only memory, an electrically-erasable programmable read-only memory (EEPROM), a flash memory, a non-volatile random access memory (NVRAM), such as a hard disk (HDD) array, a solid state drive (SSD) or an optical disk, but the present invention is not limited thereto. In an embodiment of the present disclosure, the storage device 402 stores a program, and disease data and bacterial profile data of a plurality of samples. The program includes a plurality of instructions for implementing the above method 100. When the processing device 401 loads the program from the storage device 402, the instructions are executed to implement the above method 100.
[0097] The bacterial phase feature engineering scheme provided by the present invention enables disease prediction to take into account multi-level and multi-faceted bacterial phase features. Specifically, by establishing the bacterial species level feature 11A, the effect of bacterial quantity and the presence or absence of bacterial species on the disease is taken into account, which can better highlight the relationship between key pathogenic bacteria and diseases compared with the current method; by establishing the interactive growth and decline feature 11B, the correlation between the growth and decline of bacterial species and the disease is taken into account, and in the case of unknown microbial functionality, the establishment of bacterial phase features using different biological levels can further understand the functional unit of the microorganism and provide an introduction to subsequent analysis and pathogenic mechanism research; by establishing the bacterial flora level feature 11C, the "quality" and "quantity" of the bacterial flora and the relationship between the ecological balance between bacterial flora and the disease are taken into account at the same time.
[0098] In summary, the method and system for establishing a disease prediction model provided by the present disclosure improves the richness and representativeness of feature information by establishing bacterial phase features from multiple aspects of microorganisms. This not only helps to improve the accuracy of disease prediction, but also helps to explain the analysis results. In addition, it can help humans to have a deeper understanding of the pathogenic mechanism of the disease and help to develop more effective treatment and prevention methods.
[0099] The above paragraphs are described using multiple aspects. Obviously, the teachings of this article can be implemented in multiple ways, and any specific architecture or function disclosed in the examples is only a representative case. According to the teachings of this article, it should be understood in the art that each aspect disclosed in this article can be implemented independently, or two or more aspects can be combined to implement.
[0100] Although the present disclosure has been described above with reference to the embodiments, it is not intended to limit the present disclosure. Anyone skilled in the art may make slight changes and modifications without departing from the spirit and scope of the present disclosure. Therefore, the scope of protection of the invention shall be determined by the definition of the appended claims.
Claims
1. A method for establishing a disease prediction model, executed by a computer system, the method comprising: Extracting feature values of multiple bacterial phase characteristics from bacterial phase data of each of the multiple samples; Selecting a plurality of bacterial phase features from the plurality of bacterial phase features as selected features; and Training a disease prediction model, wherein each piece of training data used to train the disease prediction model includes (1) disease data of each of the plurality of samples and (2) the feature value of the selected feature of the sample; The various bacterial phase characteristics include species-level characteristics, interactive growth and decline characteristics, and bacterial flora-level characteristics.
2. The method of claim 1, wherein the species-level features include relative abundance data and presence / absence data of each of a plurality of bacterial species.
3. The method of claim 1, wherein the interactive growth and decline feature comprises a hierarchical ratio between two taxa at a taxonomic level. The method of claim 1 , wherein the flora-level feature comprises a Beta diversity matrix.
5. The method of claim 1, wherein selecting a plurality of selected features from the plurality of bacterial phase features comprises: Inputting the disease data and the plurality of bacterial phase characteristics into a plurality of feature screening models to obtain a plurality of feature pools, wherein each of the plurality of feature screening models selects one or more of the plurality of bacterial phase characteristics to form a corresponding one of the plurality of feature pools; sorting the plurality of bacterial phase features based on the number of times each bacterial phase feature is selected into the plurality of feature pools by the plurality of feature screening models to obtain a feature ranking; and Based on the feature ranking, a specified number of bacterial phase features are selected from the multiple bacterial phase features as the multiple selected features.
6. A system for establishing a disease prediction model, comprising: a storage device for storing disease data and bacterial profile data of a plurality of samples; as well as A processing device, loading a program from a storage device to execute: Extracting feature values of multiple bacterial phase characteristics from the bacterial phase data of each of the multiple samples; Selecting a plurality of bacterial phase features from the plurality of bacterial phase features as selected features; and Training a disease prediction model, wherein each piece of training data used to train the disease prediction model includes: (1) disease data of each of the plurality of samples; and (2) the feature value of the selected feature of the sample; The various bacterial phase characteristics include species-level characteristics, interactive growth and decline characteristics, and bacterial flora-level characteristics.
7. The system of claim 6, wherein the species-level features include relative abundance data and presence / absence data of each of a plurality of bacterial species.
8. The system of claim 6, wherein the interactive fluctuation feature comprises a hierarchical ratio between two taxa at a taxonomic level.
9. The system of claim 6, wherein the flora-level feature comprises a Beta diversity matrix.
10. The system of claim 6, wherein the processing device further performs: Inputting the disease data and the plurality of bacterial phase characteristics into a plurality of feature screening models to obtain a plurality of feature pools, wherein each of the plurality of feature screening models selects one or more of the plurality of bacterial phase characteristics to form a corresponding one of the plurality of feature pools; Sort the plurality of bacterial phase features based on the number of times each bacterial phase feature is selected into the plurality of feature pools by the plurality of feature screening models to obtain a feature ranking; and Based on the feature ranking, a specified number of bacterial phase features are selected from the multiple bacterial phase features as the selected features.