Method for typing microbial flora by using ecological index and artificial intelligence technology

Abstract

The invention relates to the field of female genital tract microbial flora typing, and discloses a method for typing female genital tract microbial flora by applying a microbial flora ecological index and an artificial intelligence technology. Ecological indexes comprise a diversity index, a dominance index (community dominance and species dominance) and species abundance, and female genital tract microbial florae are divided into simple florae and complex florae according to an ecological index clustering result. An artificial intelligence model is trained and validated using the typing results, and the artificial intelligence model can be used for typing the microbial flora of the female genital tract so as to continuously optimize the artificial intelligence model. Due to the fact that pathogenic mechanisms of different types of microbial florae may be different, the method and technology provided by the invention contribute to personalized diagnosis and treatment of bacterial vaginitis. The method is not only suitable for female genital tract microbial flora typing, but also can be used for typing microbial flora of human intestinal tracts, oral cavities, skin, reproductive systems, lungs, breast milk, secretions and the like.

Description

Technical Field

[0001] The present invention relates to a method for typing microbial flora, and in particular, a method for typing female genital tract microbial flora using ecological indices (diversity index, dominance index, relative abundance) and artificial intelligence technology (including machine learning and deep learning models), which divides female genital tract microbial flora into simple communities and complex communities. A simple community refers to a community with significant dominant species, low diversity index and high dominance index; a complex community refers to a community with no significant dominant species, high diversity and low dominance index. The artificial intelligence model is trained and verified by the results of microbial flora typing, and the model can be applied to type other female genital tract microbial flora. Since microbial flora of different types play different roles in the occurrence and development of bacterial vaginitis, it can provide a theoretical basis for personalized diagnosis and treatment of female bacterial vaginitis. Background Art

[0002] Bacterial vaginosis is a highly recurrent vaginal disease with an incidence / recurrence rate of approximately 30%. Traditionally, it has been argued that a community lacking dominant species, especially bacteria related to the genus Lactobacillus, is prone to bacterial vaginitis, and this argument has been extended to the view that high community diversity is prone to vaginitis. These hypotheses are supported by many studies that highly dominant bacterial species, especially Lactobacillus, can protect women's health. However, the above phenomenon is not absolute, and non-Lactobacillus-dominated vaginal microorganisms can also maintain a healthy state. This phenomenon was confirmed in a study involving the genital microbial flora of more than 1,000 healthy women. The applicant's team has found that most of the anaerobic bacteria associated with bacterial vaginitis in the non-Lactobacillus-dominated microbial flora are inhibited by other bacteria. It may be this inhibition that establishes a protective mechanism for the host and maintains the host's health.

[0003] Existing studies have divided the female reproductive tract microbial flora into five types, from type I to type V, and type IV is further divided into two subtypes. The results showed that types I, II, III and V are dominated by bacteria of the genus Lactobacillus, and type IV is not dominated by Lactobacillus, but contains a small number of strict anaerobic bacteria, most of which are associated with female bacterial vaginitis. The patent applicant team has further quantified these five types. Types I, II, III, and V can be identified by the species specificity of the indicator species (Lactobacillus), and their specificity values ​​are the highest. Type IV is characterized by significantly higher species diversity than the other four types, but this type lacks indicator species. Therefore, a healthy microbial community may be a community dominated by lactobacillus, or a community without lactobacillus dominance. If there is a dominant species in the microbial community, the diversity of the community is low, which we call a simple community. On the contrary, when there is a lack of a dominant species, the community will have a higher community (species) diversity, which we call a complex community. Simple communities and complex communities may play completely different roles in the occurrence and development of bacterial vaginitis. Therefore, the typing of female reproductive tract microbial flora provides a theoretical basis and guidance for the personalized diagnosis and treatment of bacterial vaginitis. Summary of the invention

[0005] The objects of the present invention are: The purpose of the present invention is to use ecological indexes (diversity index, dominance index, relative abundance) and artificial intelligence technology (machine learning and deep learning) to type the female reproductive tract microbial flora. The female reproductive tract microbial flora is divided into simple communities and complex communities. Simple communities refer to communities with significant dominant species, low diversity index and high dominance index; complex communities refer to communities with no significant dominant species, high diversity and low dominance index. And train and verify the artificial intelligence model through the results of microbial flora typing, and use the trained artificial intelligence model to type other female reproductive tract microbial flora. Since microbial flora of different types play different roles in the occurrence and development of bacterial vaginitis, it can provide a theoretical basis for personalized diagnosis and treatment of female bacterial vaginitis. The method of the present invention can also be used for typing of microbial flora such as human intestinal tract, oral cavity, skin, reproductive system, breast milk, lungs, secretions, etc.

[0006] The technical solution adopted by the present invention is: For N female reproductive tract microbial flora data, the present invention uses the following process to type the flora. S1: Calculate the diversity index, dominance index, and relative abundance of each female reproductive tract microbial flora; S2: Clustering N ecological indices of female reproductive tract microbial flora, and dividing the female reproductive tract microbial flora into simple communities and complex communities according to the clustering results; S3: Simple community refers to a community with significant dominant species, characterized by high dominance, low diversity, and high relative abundance of dominant species; S4: A complex community is a community without a significant dominant species, characterized by low dominance, high diversity, and no dominant species with high relative abundance; S5: Based on the typing results, train and verify artificial intelligence models that can type the female reproductive tract microbial flora, including machine learning and deep learning models, which can be used to type other female reproductive tract microbial flora and optimize and improve the artificial intelligence models.

[0007] The effects of the present invention are: A method for typing the female reproductive tract microbial flora using ecological indices (diversity index, dominance index, relative abundance) and artificial intelligence techniques (including machine learning and deep learning) has been invented. The female reproductive tract microbial flora can be divided into simple and complex communities, providing a theoretical basis for personalized diagnosis and treatment of female bacterial vaginitis. The method of the present invention can also be used for typing the microbial flora of the human intestine, oral cavity, skin, reproductive system, breast milk, lungs, secretions, etc. BRIEF DESCRIPTION OF THE DRAWINGS

[0008] Attached photos Figure 1 The present invention is an implementation flow chart of the method of the present invention. The present invention divides healthy population samples and vaginitis patient population samples into simple communities and complex communities respectively by diversity index, dominance index (including community dominance index and species dominance index) and relative abundance. Then, machine learning and deep learning models are applied to train and test the results of flora typing. Attached photos Figure 2 It is a fully connected neural network model of deep learning model. Figure 2 The figure above is the neuron structure, including input, weight, activation function and output value. Neuron is the basic unit of neural network. Figure 2 The figure below is a fully connected neural network, including input layer, hidden layer and output layer. Each hidden layer contains several neuron structures. DETAILED DESCRIPTION

[0009] The present invention takes 3265 reproductive tract microbial flora samples of healthy women and 3617 reproductive tract microbial flora samples of vaginitis patients as examples to demonstrate the typing method process for healthy flora samples and vaginitis patient flora. S1: Calculate the diversity index, dominance index, and relative abundance of each female reproductive tract microbial flora sample, and calculate the diversity of the microbial flora according to formula 1, where p i represents the relative abundance of bacterial species i, N represents the number of bacterial species, and q represents the order of diversity: The dominance index at the species level and the dominance index at the community level were calculated according to formula 2-5, where m represents the average relative abundance of all bacteria in the community, and D c Represents the dominance index at the community level: D s Represents the dominance index at the species level. D s =D c -D SD . (5) S2: The ecological indices of genital tract microbial flora samples of 3265 healthy women and 3617 genital tract microbial flora samples of vaginitis patients were clustered respectively. According to the clustering results, the female genital tract microbial flora of healthy people was divided into simple communities and complex communities, and the genital tract microbial flora of vaginitis patients was divided into simple communities and complex communities. S3: Simple community refers to a community with significant dominant species, characterized by high dominance, low diversity, and high relative abundance of dominant species (Healthy-S and BV-S in Table 1). S4: A complex community is a community without a significant dominant species, characterized by low dominance, high diversity, and no dominant species with high relative abundance (Healthy-C and BV-C in Table 1). S5: Based on the typing results, train and verify artificial intelligence models that can type the female reproductive tract microbial flora, including machine learning and deep learning models, which can be used to type other female reproductive tract microbial flora. Apply the following artificial intelligence models: (1) Linear support vector machines (SVM) are a two-class classification model. Its basic model is a linear classifier with the largest margin defined in the feature space. The learning strategy is to maximize the margin, which is formalized as solving a convex quadratic programming problem: ξ i ≥0, i=1, 2, ... n (2) Random forest is a supervised machine learning algorithm constructed from a large number of decision trees. It is an ensemble learning method for classification, regression and other tasks. Random forest establishes results based on the predictions of decision trees. It makes predictions by calculating the average output of various decision trees. Increasing the number of trees can improve the accuracy of the results. The decision tree algorithm adopts a tree structure and uses layer-by-layer reasoning to achieve the final classification. The decision tree consists of the following elements: 1. Root node: contains the full set of samples. 2. Internal nodes: corresponding to feature attribute tests. 3. Leaf nodes: represent the results of the decision. (3) k-nearest neighbor (k-NN) classification is performed by measuring the distance between different feature values. The distance is generally measured using Euclidean distance or Manhattan distance: or (4) Logistic regression is simply a Sigmoid function added to the linear regression model: That is, replace t with the linear regression model w T x * , and get the logistic regression model: (5) Gradient Boosting Decision Tree is an iterative decision tree algorithm that consists of multiple decision trees and is an algorithm model for multi-classification. Gradient boosting constructs trees in a continuous manner, and each tree attempts to correct the errors of the previous tree. (6) Fully Connected Neural Network Deep learning uses the most basic neural network architecture. A fully connected neural network consists of an input layer, a hidden layer, and an output layer (such as Figure 2 as indicated).

[0010] According to the above steps, the present invention divides the genital tract microbial flora samples of 3265 healthy women into simple communities and complex communities, and divides the genital tract microbial flora samples of 3617 vaginitis patients into simple communities and complex communities. As shown in Table 1, among the samples of healthy people, 1546 samples are complex communities (Healthy-C) and 1719 samples are simple communities (Healthy-S). Among patients with bacterial vaginitis, 1627 samples are complex communities (BV-C) and 1990 samples are simple communities (BV-S). Alpha Diversity is a diversity index, Community Dominance is a dominance index at the community level, Species Dominance is a dominance index at the species level, and Species abundance is the relative abundance of species. As can be seen from Table 1, the diversity of simple communities is significantly lower than that of complex communities, and the dominance index of simple communities is higher than that of complex communities. Table 1. Genital microbial flora typing results of healthy people and patients with bacterial vaginosis

[0011] After typing the reproductive tract microbial flora, the typing results are used to train the artificial intelligence model for typing other female reproductive tract microbial flora. The specific process is as follows: 1. 70% of the samples were randomly selected to train machine learning and deep learning models, including random forest, decision classification tree, K nearest neighbor classification, support vector machine, logistic regression and fully connected neural network, and the remaining 30% of the sample data were used to verify the accuracy of the model. 2. To eliminate the impact of sample differences on the results, repeat step 1 100 times. Table 2 is the average value of the 100 model validation accuracies. In Table 2, Groups is the classification method of each bacterial community, NN represents the fully connected neural network model of the deep learning method, RF is the random forest model, GBC is the gradient boosting decision tree, LR is the logistic regression model, KNN is the K nearest neighbor classification model, and SVM is the support vector machine model. As can be seen from Table 2, the accuracy of all models for all classifications is above 70%. The classification effect of the random forest model is better than that of other models, with a classification accuracy of about 90% and a maximum of about 95%. Explanation of the high accuracy of bacterial community classification prediction: (1) There are significant differences in the ecological characteristics of the bacterial flora of healthy people and those of patients with bacterial vaginosis. (2) There are significant differences in the ecological indicator characteristics of simple and complex communities of the microbiome in healthy people. (3) There are significant differences in the ecological indicator characteristics of simple and complex communities of bacterial vaginosis patients. (4) There are significant differences in the ecological indicator characteristics of the simple community microbiota of healthy people and that of patients with bacterial vaginosis. (5) There are significant differences in the ecological index characteristics of the complex community flora of healthy people and bacterial vaginosis patients. This result confirms the reliability of the application of ecological index typing and artificial intelligence model typing. Table 2. Average accuracy of artificial intelligence models from 100 sampling tests

Claims

1. A method for typing female reproductive tract microbial flora using ecological indexes and artificial intelligence technology, wherein the ecological indexes include diversity index, community dominance index, species dominance index, and relative abundance, and the artificial intelligence technology includes machine learning and deep learning models, and typing N microbial flora through the following steps: S1: Calculate the diversity index, dominance index, and relative abundance of each female reproductive tract microbial flora; S2: Clustering N ecological indices of female reproductive tract microbial flora, and dividing the female reproductive tract microbial flora into simple communities and complex communities according to the clustering results; S3: Simple community refers to a community with significant dominant species, characterized by high dominance and low diversity, and the relative abundance of dominant species is very high; S4: Complex communities are communities without significant dominant species, characterized by low dominance and high diversity, with few dominant species with high relative abundance. S5: Based on the typing results, train and verify artificial intelligence models capable of typing the female reproductive tract microbial flora, including machine learning and deep learning models. The obtained models can be used to type other female reproductive tract microbial flora.

2. The method according to claim 1, characterized in that: The female reproductive tract community is divided into simple community and complex community, and the simple community and complex community play different roles in the occurrence, development and pathogenic mechanism of bacterial vaginitis.

3. The method according to claim 1, characterized in that: The microbial flora typing method is not limited to the microbial flora of the female reproductive tract, and can be used to type the microbial flora of the human intestine, oral cavity, skin, breast milk, reproductive system, lungs, and secretions.

4. The method according to claims 1-3, characterized in that: Products that implement their algorithms and functions in any form of software, firmware and hardware to provide services.

Images