Microbial community optimal microhabitat identification method based on machine learning

Abstract

The invention discloses a machine learning-based microbial community optimal microhabitat identification method, which comprises the following steps of: firstly, acquiring microbial community information and microhabitat information, and calculating a microbial community evaluation index; secondly, training a machine learning model by taking the microflora evaluation index as a predicted value and the microhabitat information as a characteristic value; evaluating the performance of the machine learning model, performing importance ranking on the microhabitat features, and screening key microhabitat features; and finally, based on the machine learning model, obtaining the optimal range of the key microhabitat features by adopting individual condition expectation. According to the method, non-linear and complex relations in microbial communities are captured through machine learning, and the optimal range of visual and available key microhabitat features is obtained through individual condition expectation.

Description

Technical Field

[0001] The present invention relates to a method for identifying an optimal microhabitat of a microbial community, and in particular to a method for identifying an optimal microhabitat of a microbial community based on machine learning, and belongs to the field of bioinformatics and biotechnology. Background Art

[0002] Microbial communities are complex ecosystems composed of different microorganisms. They play a key role in processes such as material circulation and energy flow, and play an important role in sustainable agriculture, ecological restoration, and human health. Microhabitats refer to the specific environment in which microorganisms live, including physical, chemical, and biological factors such as temperature, pH, humidity, and nutrients. These factors in the microhabitat not only affect the microbial diversity and community structure of the microbial community, but also determine the ecological function and adaptability of the microbial community. Therefore, identifying microhabitat conditions is of great significance for stabilizing or regulating microbial communities.

[0003] Traditional microhabitat identification methods are mainly statistical methods, such as regression analysis, analysis of variance (ANOVA), principal component analysis (PCA), structural equation analysis, etc. However, as the study of microbial communities enters the era of big data, these methods can no longer meet the requirements for accurate identification of microhabitats: regression analysis cannot capture nonlinear and complex relationships; variance analysis cannot deeply explore the interactions between environmental factors; principal component analysis is unreliable when dealing with nonlinear or complex interactions; similarly, correlation analysis cannot capture nonlinear and complex relationships, and cannot infer causal relationships based on correlation results. At present, there is no method that can analyze the complex nonlinear relationship between microbial communities and microhabitat characteristics and identify the optimal microhabitat conditions. Summary of the invention

[0004] Purpose of the invention: The purpose of the present invention is to provide a microhabitat identification method based on machine learning, which uses machine learning to capture the nonlinear and complex relationships in microbial communities, and identifies the optimal range of intuitively available key microhabitats through individual conditional expectation analysis.

[0005] Technical solution: The present invention provides a method for identifying the optimal microhabitat of a microbial community based on machine learning, comprising the following steps: The first step is to obtain microbial community information and microhabitat information, and calculate microbial community evaluation indicators based on the microbial community information; In the second step, the microbial community assessment index was used as the prediction value and the microhabitat information as the feature value to train the machine learning model; The third step is to evaluate the performance of the machine learning model, rank the importance of microhabitat features, and screen key microhabitat features; The fourth step is to use individual conditional expectations to obtain the optimal range of the key microhabitat characteristics based on the machine learning model.

[0006] The present invention uses machine learning to establish a microhabitat identification method, which can capture the nonlinear and complex relationship between microbial communities and microhabitat characteristics, and derive key microhabitat characteristics and optimal ranges based on individual conditional expectation analysis. In the first step, the microbial community information includes species type and abundance, and the microhabitat information includes physical and chemical factors and geographical information that affect the structure of the microbial community. In the second step, the microbial community evaluation indicators include commonly used indicators in this field such as α diversity; the machine learning model includes commonly used models such as the gradient boosting regression tree algorithm. The third step is to evaluate the performance of the trained model, which helps to optimize the model to improve the prediction accuracy. The fourth step uses the constructed machine learning model to identify the optimal microhabitat characteristics, and through the individual conditional expectation (ICE) analysis that can reflect the relationship between the predicted value of each individual and a single variable, the optimal range of intuitively available key microhabitat characteristics is identified.

[0007] Preferably, in the first step, the microbial community information is obtained through open data sources or sequencing technology, such as 16S rRNA amplicon sequencing.

[0008] Preferably, in the first step, the microhabitat information includes physical and chemical factors and geographical information that significantly affect the microbial community structure, such as total nitrogen, ammonia nitrogen, nitrate nitrogen, nitrite nitrogen, water-soluble organic nitrogen, water temperature, pH, total phosphorus, organic carbon, longitude and latitude, etc.

[0009] Preferably, in the first step, the microbial community evaluation index includes α diversity, β diversity or a microbial community evaluation index based on the topological structure of the microbial co-occurrence network.

[0010] Preferably, in the second step, the machine learning model is a gradient boosted regression tree algorithm (GBRT). The GBRT algorithm is suitable for large data sets, can process various types of data, and has high prediction accuracy and generalization ability.

[0011] Preferably, in the second step, the machine learning model uses 75% to 85% of the predicted values ​​and feature values ​​as a training set, and the rest as a validation set. Preferably, 75% of the data is used as a training set.

[0012] Preferably, in the second step, the machine learning model uses Bayesian optimization to perform hyperparameter optimization.

[0013] Preferably, in the second step, the machine learning model is cross-validated several times, preferably 10 times.

[0014] Preferably, in the third step, the evaluation includes using R 2Evaluate the model fit, or use mean absolute error (MAE) and root mean square error (RMSE) for evaluation.

[0015] Preferably, in the third step, the importance ranking is obtained based on permutation feature importance (PFI), and the top 5 features in importance ranking are preferably used as key microhabitat features.

[0016] Beneficial effects: Compared with the prior art, the present invention has the following significant advantages: the present invention uses machine learning models to capture the nonlinear and complex relationships in microbial communities, and identifies the optimal range of intuitively available key microhabitat characteristics through individual condition expectations. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] Figure 1 A schematic diagram of the process of the method for identifying the optimal microhabitat of a microbial community provided by the present invention; Figure 2 Identification and performance evaluation of key microhabitat features using a gradient boosted regression tree algorithm for embodiments of the present invention; Figure 3 is the optimal range of the five key microhabitat characteristics in the embodiment of the present invention; Figure 4 This is the verification result of the optimal range of key microbial characteristics of the embodiment of the present invention. DETAILED DESCRIPTION

[0018] The technical solution of the present invention is further described below in conjunction with the accompanying drawings.

[0019] This embodiment provides a method for identifying the optimal microhabitat of a microbial community based on machine learning. The microbial community was collected from the sewage biological treatment system, and its process parameters, water samples and sludge samples were collected from various biochemical pools of 177 sewage treatment plants across the country, including anaerobic pools, anoxic pools and aerobic pools, with a total of 1068 water samples and sludge samples.

[0020] The optimal microhabitat identification method for microbial communities based on machine learning is as follows Figure 1 As shown, the specific steps include: The first step is to obtain microbial community information and microhabitat information, and calculate microbial community evaluation indicators based on the microbial community information.

[0021] Microbial community information, including species type and abundance, was derived from sequencing and literature research. The sludge samples were sequenced for 16S rRNA amplicon to determine the bacterial 16S rRNA V3-V4 region. The 16S primers used for sequencing were: 341F (CCTAYGGGRBGCASCAG); 806R (GGACTACNNGGGTATCTAAT). The sulfamethoxazole (SMX) degradation community composed of functional species, structural species and co-metabolizing species was determined through literature research and statistical analysis. The microbial network topology coefficient was selected as the microbial community evaluation indicator, and the microbial network topology coefficient was obtained by constructing the biological association network using Spiec-Easi.

[0022] Microhabitat information includes physical and chemical factors and geographical information that affect the structure of microbial communities: total nitrogen (TN), ammonia nitrogen (NH4+), temperature (T), pH, total phosphorus (TP), total organic carbon (TOC), dissolved oxygen (DO), sludge age (SRT) and hydraulic retention time (HRT), etc. It is mainly analyzed with reference to water and wastewater monitoring and analysis methods; the process parameters refer to the sewage biological treatment system for specific reference.

[0023] In the second step, the microbial network topology coefficient was used as the prediction value and the microhabitat information as the feature value to train the machine learning model gradient boosted regression tree (GBRT).

[0024] Bayesian optimization and 10-fold cross validation were used to optimize GBRT hyperparameters, and the microbial community assessment indicators and water quality characteristics were input into the machine learning model with a training set: predicted value ratio of 75%:25%.

[0025] The third step is to evaluate the performance of the machine learning model, rank the importance of microhabitat features, and screen key microhabitat features.

[0026] Using R 2 , mean absolute error (MAE) and root mean square error (RMSE) were used to evaluate the performance of GBRT. The top five microhabitat features in terms of importance are called key microhabitat features, such as Figure 2 and Figure 3 shown.

[0027] The fourth step is to use individual conditional expectation (ICE) to obtain the optimal range of the key microhabitat characteristics based on the machine learning model.

[0028] The key microhabitat characteristics and their optimal ranges were as follows: the optimal ranges of dissolved oxygen were 0.36-0.38 mg / L for anaerobic pools, 0.72-0.73 mg / L for anoxic pools, and 4.82-4.89 mg / L for aerobic pools; the optimal ranges of temperature were less than 15.73°C for anaerobic pools, less than 15.56°C for anoxic pools, and 13.60-15.70°C for aerobic pools. The optimal ranges of pH were 6.82-7.23 for anaerobic pools, 6.89-6.98 for anoxic pools, and 7.49-7.59 for aerobic pools. The total phosphorus was negatively correlated with the microbial evaluation indexes, and the total nitrogen was positively correlated with the microbial evaluation indexes.

[0029] Based on the difference between the optimal range and the worst range of microhabitat characteristics, the optimal range of key microhabitat characteristics was verified. Figure 4 As shown in the figure, the number of functional species with significantly increased abundance in the optimal range of key microhabitat characteristics of different biochemical pools is significantly higher than that with the worst range. The results show that the optimal range of key microhabitat characteristics obtained by the optimal microhabitat identification method for microbial communities based on machine learning can effectively promote the stability and functionality of the target microbial community.

Claims

1. A method for identifying the optimal microhabitat of a microbial community based on machine learning, characterized in that: The following steps are involved: The first step is to obtain microbial community information and microhabitat information, and calculate microbial community evaluation indicators based on the microbial community information; In the second step, the microbial community assessment index was used as the prediction value and the microhabitat information as the feature value to train the machine learning model; The third step is to evaluate the performance of the machine learning model, rank the importance of microhabitat features, and screen key microhabitat features; The fourth step is to use individual conditional expectations to obtain the optimal range of the key microhabitat characteristics based on the machine learning model.

2. The optimal microhabitat identification method according to claim 1, characterized in that: In the first step, the microbial community information is obtained through open data sources or sequencing technology.

3. The optimal microhabitat identification method according to claim 1, characterized in that: In the first step, the microhabitat information includes physicochemical factors and geographical information that significantly affect the structure of the microbial community.

4. The optimal microhabitat identification method according to claim 1, characterized in that: In the first step, the microbial community evaluation index includes α diversity, β diversity or a microbial community evaluation index based on the topological structure of the microbial co-occurrence network.

5. The optimal microhabitat identification method according to claim 1, characterized in that: In the second step, the machine learning model is a gradient boosting regression tree algorithm.

6. The optimal microhabitat identification method according to claim 1, characterized in that: In the second step, the machine learning model uses 75% to 85% of the predicted values ​​and feature values ​​as the training set, and the rest as the validation set.

7. The optimal microhabitat identification method according to claim 1, characterized in that: In the second step, the machine learning model is optimized for hyperparameters using Bayesian optimization.

8. The optimal microhabitat identification method according to claim 1, characterized in that: In the second step, the machine learning model is cross-validated several times.

9. The optimal microhabitat identification method according to claim 1, characterized in that: In the third step, the evaluation includes R 2 , mean absolute error, or root mean square error.

10. The optimal microhabitat identification method according to claim 1, characterized in that: In the third step, the importance ranking is obtained based on the permuted feature importance.

Images