Method and system for regulating and controlling nitrogen circulation of constructed wetland by using root exudates

Abstract

The invention provides a method and a system for regulating and controlling nitrogen circulation of a constructed wetland by using root exudates, and relates to the technical field of constructed wetland sewage treatment. The method comprises the following steps: 1) constructing a plant root exudate database; 2) training a machine learning prediction model and optimizing parameters by adopting a machine learning algorithm and using the data of the plant root exudate database; 3) screening the compounds in the natural product database by using the machine learning prediction model in the step 2), predicting the denitrification potential of the compounds, and selecting candidate substances; and 4) constructing a microcosm culture system, in the microcosm culture system, detecting nitrogen removal and microbial response by adding the candidate substances obtained in the step 3) with different concentration gradients, and verifying the denitrification effect of the candidate substances. The system disclosed by the invention is high in datamation degree, simple and convenient to operate, accurate in prediction and low in cost, provides a new technical means for enhanced denitrification of the constructed wetland, and has a good industrial application prospect.

Description

Technical Field

[0001] This invention relates to the field of constructed wetland wastewater treatment technology, and in particular to a method and system for regulating nitrogen cycle in constructed wetlands using root exudates. Background Technology

[0002] Constructed wetlands are ecosystems that mimic natural wetlands, utilizing the synergistic effects of substrates, microorganisms, and plants to purify wastewater. Compared to traditional wastewater treatment methods, constructed wetlands offer advantages such as low energy consumption, low operating costs, and high treatment efficiency, making them a popular wastewater treatment technology. However, constructed wetlands still have some limitations in nitrogen removal, mainly manifested in unstable nitrogen removal efficiency and an unclear nitrogen removal mechanism.

[0003] Constructed wetland nitrogen removal primarily relies on a coupled nitrification-denitrification process, where ammonia nitrogen is oxidized to nitrate nitrogen under aerobic conditions and then reduced back to nitrogen gas under anaerobic or anoxic conditions. Many factors influence this process, such as dissolved oxygen, carbon-to-nitrogen ratio, temperature, and pH. Currently, methods to improve the nitrogen removal efficiency of constructed wetlands mainly include optimizing wetland design parameters (such as hydraulic loading and hydraulic retention time), enhancing artificial aeration, and adding exogenous carbon sources. However, these methods are either energy-intensive, costly, or prone to causing secondary pollution, making large-scale application difficult.

[0004] Plants, as an important component of constructed wetlands, play an irreplaceable role in maintaining system stability and improving purification efficiency. Increasing research indicates a close relationship between plant root exudates and nitrogen removal. Plant roots secrete various compounds into the rhizosphere, including sugars, amino acids, organic acids, and phenolic acids. These compounds provide carbon and energy sources for microorganisms and can also influence microbial community composition and metabolic activity, thereby affecting nitrogen transformation processes. Therefore, root exudates hold promise as a novel biomaterial for regulating nitrogen cycling in constructed wetlands.

[0005] Currently, research on enhancing nitrogen removal in constructed wetlands using root exudates is still in its early stages. Existing studies mainly focus on the compositional analysis of root exudates and the evaluation of the physiological and ecological effects of single compounds. Summary of the Invention

[0006] To address the technical problems existing in the prior art, embodiments of the present invention provide a method and system for regulating nitrogen cycling in constructed wetlands using root exudates. The technical solution is as follows:

[0007] A method for regulating nitrogen cycling in constructed wetlands using root exudates, the method comprising the following steps:

[0008] 1) Construct a plant root exudate database, which collects the following information about plant root exudates: compound name, molecular formula, structure, physicochemical properties and biological activity, and performs SMILES normalization and activity binary labeling preprocessing on the data;

[0009] 2) A machine learning algorithm is used, with simplified molecular linear canonical strings of compounds as input and the ability to promote nitrification and denitrification as output. The data from the plant root exudate database is used to train the machine learning prediction model and optimize the parameters.

[0010] 3) Use the machine learning prediction model described in step 2) to screen compounds in the natural product database, predict their denitrification potential, and select candidate substances;

[0011] 4) Construct a microcosm culture system. In the microcosm culture system, by adding the candidate substances from step 3) at different concentration gradients, nitrogen removal and microbial response are detected to verify the denitrification effect of the candidate substances. The microcosm culture system includes: a culture device, a sample introduction device, and a monitoring device. The culture device includes: an plexiglass device, a soil substrate, and wetland plants. The sample introduction device includes: a sample pump, a sample inlet, a connecting tube, and a simulated wastewater storage tank. The monitoring device includes: a dissolved oxygen monitoring device, a temperature monitoring device, and a pH probe.

[0012] Optionally, the method further includes:

[0013] 5) The experimental data obtained in step 4) are integrated using the machine learning algorithm to reveal the nitrogen cycle process and microbial driving mechanism of the constructed wetland mediated by the candidate substances, and to screen out key active substances and their optimal concentrations.

[0014] Optionally, in step 1), the plant root exudate database includes compound information from literature reports and natural product databases, wherein the number of active substances is 200-500.

[0015] And / or, in step 2), the machine learning prediction model is constructed using the Chemprop deep learning model, and the optimal model parameters are determined by verification through key indicators, wherein the key indicators include: the average test score of the model training results, the model prediction and recognition accuracy, and the number of consecutive correct recognitions, and the optimal model parameters include: the average test score of the model training results > 0.8, and the preferred model training results average test score > 0.9.

[0016] Optionally, in step 3), the candidate substances include: fumaric acid, naringenin, genistein, and rosinin secreted by plant roots.

[0017] Optionally, in step 4), the micro-universe cultivation system includes plexiglass material, a soil matrix is ​​laid at the bottom of the plexiglass device, simulated sewage is configured, wetland plants are planted, and an artificial wetland pilot system is constructed.

[0018] And / or, periodically monitor the concentrations of ammonia nitrogen, nitrite nitrogen, nitrate nitrogen and / or total nitrogen in the system, and assess the contribution of nitrification and denitrification processes to total nitrogen removal;

[0019] And / or, use amplicon sequencing technology to analyze changes in microbial community structure;

[0020] And / or, using real-time quantitative PCR, analyze the effects of candidate substances on key functional genes of the nitrogen cycle.

[0021] A system for enhancing nitrogen removal in constructed wetlands using root exudates, the system comprising key active substances obtained by screening using the method and / or formulations prepared from the key active substances.

[0022] A system for regulating nitrogen cycling in constructed wetlands using root exudates, the system comprising:

[0023] (1) Plant root exudate database: The database collects data on plant root exudates, including the following information: compound name, molecular formula, structure, physicochemical properties and biological activity, and performs SMILES normalization and activity binary labeling preprocessing on the data to provide data support for screening candidate substances;

[0024] (2) Machine learning prediction unit: The machine learning prediction unit includes a machine learning prediction model, which is trained by using data from the plant root exudate database, with the simplified molecular linear canonical string of the compound as the independent variable and the ability to promote nitrification and denitrification as the dependent variable, and is used to screen compounds in the natural product database, predict their denitrification potential, and screen candidate substances.

[0025] (3) Microcosm culture system: The microcosm culture system is used to simulate the artificial wetland environment. By adding the candidate substances, its impact on key processes of wetland nitrogen cycle and functional microorganisms is evaluated. The microcosm culture system includes: a culture device, a sample injection device, and a monitoring device. The culture device includes: an plexiglass device, a soil substrate, and wetland plants. The sample injection device includes: a sample injection pump, a sample inlet, a connecting pipe, and a simulated wastewater storage tank. The monitoring device includes: a dissolved oxygen monitoring device, a temperature monitoring device, and a pH probe.

[0026] Optionally, the machine learning prediction model is constructed using the Chemprop deep learning model, and the optimal model parameters are determined through key indicators. The key indicators include: the average test score of the model training results, the model prediction accuracy, and the number of consecutive correct recognitions. The optimal model parameters include: the average test score of the model training results > 0.8, and the preferred model training results average test score > 0.9.

[0027] Optionally, the candidate substances include: fumaric acid, naringenin, genistein, and rosin, which are secreted by plant roots.

[0028] Optionally, the micro-universe cultivation system includes an plexiglass material, on which a soil matrix is ​​laid, wetland plants are planted, simulated wastewater is prepared, and a pilot-scale artificial wetland system is constructed.

[0029] And / or, periodically monitor the concentrations of ammonia nitrogen, nitrite nitrogen, nitrate nitrogen and / or total nitrogen in the system, and assess the contribution of nitrification and denitrification processes to total nitrogen removal;

[0030] And / or, use amplicon sequencing technology to analyze changes in microbial community structure;

[0031] And / or, using real-time quantitative PCR, analyze the effects of candidate substances on key functional genes of the nitrogen cycle.

[0032] The beneficial effects of the technical solutions provided in the embodiments of the present invention include at least the following:

[0033] This invention utilizes big data mining and machine learning technologies to screen key active substances from a massive natural product database and optimize their application schemes, thereby forming a new technology and process for enhanced nitrogen removal in constructed wetlands using root exudates. This not only significantly improves the nitrogen removal efficiency of constructed wetlands but also has advantages such as low cost and environmental friendliness, aligning with the concept of sustainable development and possessing broad application prospects. Therefore, the new technology for enhanced nitrogen removal in constructed wetlands based on root exudates developed in this invention is of great significance for improving the wastewater treatment efficiency of constructed wetlands, promoting wetland ecological restoration, and protecting the water environment. This invention aims to use advanced omics and information technology to systematically reveal the process and mechanism of root exudates regulating nitrogen cycling in constructed wetlands, screen key active substances, and optimize process parameters, providing new methods and approaches for nitrogen removal in constructed wetlands. This system can be used to study the process and mechanism of root exudates regulating nitrogen removal in constructed wetlands, providing new ideas and methods for enhanced nitrogen removal in constructed wetlands.

[0034] (1) The root exudate regulation artificial wetland nitrogen cycle system provided by the present invention has a high degree of integration. It realizes intelligent screening of compounds through machine learning prediction model. It can predict substances with unknown activity in natural product database (400,000 substances). Compared with traditional screening methods, the screening efficiency is greatly improved and the research and development cycle is greatly shortened.

[0035] In addition, the system adopts a standardized microcosmic culture device and is designed according to actual artificial wetland parameters to ensure that the experimental results have good scalability for engineering applications.

[0036] (2) This invention utilizes a root exudate regulation system to optimize the nitrogen removal performance of constructed wetlands without the need for additional chemical agents, which can significantly improve nitrogen removal efficiency and achieve efficient nitrogen removal treatment of constructed wetlands; the screened efficient root exudates can significantly promote nitrification and denitrification processes and enhance microbial activity; the established evaluation method can comprehensively reveal the microbial ecological regulation mechanism of root exudates.

[0037] (3) The present invention has a high degree of system data, is easy to operate, accurate in prediction and low in cost, providing new theoretical basis and technical means for enhanced denitrification in artificial wetlands, and has good prospects for industrial application. Attached Figure Description

[0038] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0039] Figure 1 This is a schematic diagram of a method and system for regulating nitrogen cycle in artificial wetlands using root exudates, provided in an embodiment of the present invention; wherein, 1-root exudate database, 2-machine learning prediction unit, 3-machine learning model, 4-screening active substance unit, 5-microcosm culture system, 6-wetland plants, 7-simulated wastewater, 8-soil and other substrates. Detailed Implementation

[0040] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0041] The terms “first,” “second,” “third,” “fourth,” etc. (if present) in the specification, claims, and accompanying drawings of this invention are used to distinguish similar objects and are not necessarily used to describe a particular order or sequence. It should be understood that such data can be interchanged where appropriate so that embodiments of the invention described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms “comprising” and “having,” and any variations thereof, are intended to cover a non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0042] This invention proposes a method and application for regulating nitrogen cycling in constructed wetlands using root exudates. This method integrates the construction of a root exudate compound database, the prediction and screening of key active substances, and the verification through micro-environment cultivation, forming a complete technical route and solution.

[0043] Constructed wetlands are artificial wastewater treatment systems built using the material cycling and energy flow of plants, microorganisms, and substrates within an ecosystem. They are characterized by simple management and low operating costs. Existing research indicates that nitrogen removal in wetland systems mainly involves processes such as ammonia volatilization, media sedimentation and adsorption, microbial nitrification / denitrification, and plant uptake. Microbial nitrification and denitrification are generally considered the most important pathways for nitrogen removal in wetlands, achieving removal rates as high as 60%-95%. Nitrification refers to the process under aerobic conditions where ammonia nitrogen is first converted to nitrite by nitrifying bacteria, and then nitrite is converted to nitrate by nitrifying bacteria. Denitrification occurs under anoxic conditions and refers to the process where nitrate is converted to nitrate through a series of intermediate products (NO2). - Nitrification (NOx, N2O) is reduced to nitrogen gas by denitrifying bacteria. In constructed wetlands, the rates of nitrification and denitrification need to be balanced to achieve maximum nitrogen removal efficiency.

[0044] Root exudates are a series of chemical substances released into the external environment during normal plant growth. High-molecular-weight compounds include mucilage and proteins, while low-molecular-weight organic compounds include amino acids, organic acids, sugars, phenols, and secondary metabolites. Amino acids and sugars secreted by roots can act as chemical attractants to attract plant growth-promoting bacteria (PGPRs) for colonization. Some species of these PGPRs can utilize tryptophan secreted by roots to synthesize plant growth hormones. Root exudates play an important role in promoting plant growth, nutrient activation, and mitigating environmental stress.

[0045] Wetland plants are plants that grow in areas where land and water meet, in moist soil or with shallow water accumulation. The term "wetland plants" broadly refers to plants that grow in transitionally moist environments. In a narrower sense, wetland plants specifically refer to those that grow in areas where land and water meet, in moist soil or with shallow water accumulation. Wetland plants are diverse, mainly including aquatic, marsh, halophytic, and some mesophytic herbaceous plants, and they possess unique ecological value in nature.

[0046] This invention provides a system and method for regulating nitrogen cycling in constructed wetlands using root exudates. The system includes a root exudate database, a machine learning prediction model, and a microcosm culture system. The root exudate database contains compound information on plant root exudates, including compound names, molecular formulas, structures, and biological activities, wherein the biological activities of the root exudates in the database are known. The machine learning prediction model screens compounds in the natural product database, predicts their denitrification potential, and selects several candidate substances. The microcosm culture system is made of plexiglass and designed according to the aspect ratio and surface loading rate of an actual constructed wetland. By adding candidate substances and detecting nitrogen removal and microbial responses, the denitrification effect and mechanism of action of the candidate substances are verified. This invention also provides a method for evaluating the impact of root exudates on nitrogen cycling in constructed wetlands. By calculating the rate of change of different forms of nitrogen concentration, the impact of candidate substances on nitrification and denitrification processes is assessed; and by analyzing microbial community structure and functional gene abundance, the microbial ecological mechanisms of the candidate substances are revealed. This system can be used to study the process and mechanism of nitrogen removal in constructed wetlands regulated by root exudates, providing new ideas and methods for enhanced nitrogen removal in constructed wetlands.

[0047] Specifically, the present invention constructs the following system and modules:

[0048] Root Exudate Database. This database compiles reported data on plant root exudates, including compound names, molecular formulas, structures, and physicochemical properties, providing data support for candidate screening.

[0049] Machine learning prediction unit. This unit uses the simplified molecular linear canonical string of a compound as the independent variable and its denitrification promotion ability as the dependent variable to predict the denitrification potential of root exudates and screen candidate substances.

[0050] Microcosm Culture System. This system simulates an artificial wetland environment under laboratory conditions. By adding candidate substances, it evaluates their impact on key processes of the wetland nitrogen cycle and functional microorganisms, revealing their mechanisms of action. The system consists of a culture device, a sample introduction device, and a monitoring device. The culture device includes: an acrylic glass apparatus, soil substrate, and wetland plants; the sample introduction device includes: a sample pump, a sample inlet, connecting tubing, and a culture medium storage tank; the monitoring device includes: a dissolved oxygen monitoring device, a temperature monitoring device, and a pH probe.

[0051] The technical process of this invention is as follows:

[0052] A root exudate database was constructed, and the data underwent SMILES normalization and activity binary labeling preprocessing. The specific steps of the activity binary labeling preprocessing are as follows: Based on literature reports and the results of biochemical experiments in PubChem's sub-database PubChemBioassays, root exudate characteristics were labeled: compounds promoting nitrification and denitrification were labeled as "1," and compounds without promoting nitrification and denitrification were labeled as "0." Furthermore, the SMILES expressions for these substances were obtained, and a training dataset was constructed to develop a machine learning model for screening highly reactive nitrogen removal regulators.

[0053] Using machine learning models, models for predicting nitrification and denitrification potential were constructed with different training set data volumes and different training parameter settings, and the optimal model was determined.

[0054] Based on the validated best model, machine learning was used to predict and screen natural product datasets to predict several candidate substances that can promote the denitrification capacity of constructed wetlands.

[0055] In the microcosm culture system, we studied the effects of candidate substances on key processes of nitrogen cycling and microorganisms in constructed wetlands, and revealed their mechanisms of action.

[0056] In the microcosm cultivation experiment, the present invention adopts the following technical approach:

[0057] In a simulated artificial wetland cultivation device, soil substrate is added to construct a stable nitrogen cycle system;

[0058] Candidate substances were added to the system at different concentration gradients, and a blank control group was set up.

[0059] Regularly monitor the concentrations of nitrogen forms such as ammonia nitrogen, nitrite nitrogen, and nitrate nitrogen in the system, and assess the contribution of nitrification and denitrification processes to total nitrogen removal;

[0060] Amplicon sequencing technology was used to analyze changes in the structure of the microbial community;

[0061] The abundance dynamics of key functional genes in the nitrogen cycle were analyzed using real-time quantitative PCR.

[0062] Through microcosm experiments, this invention can screen out key active substances that significantly promote nitrogen removal in constructed wetlands and preliminarily elucidate their mechanisms of action, providing a theoretical basis and technical guidance for subsequent applications. This invention utilizes cutting-edge technologies such as systems biology and artificial intelligence to accelerate the discovery of key active substances, providing new ideas and methods for improving the nitrogen removal efficiency of constructed wetlands. This technology has promising application prospects and significant potential for widespread application.

[0063] Specifically, the present invention provides:

[0064] A method for regulating nitrogen cycling in constructed wetlands using root exudates includes the following steps:

[0065] 1) Construct a database of plant root exudates, collecting information such as compound names, molecular formulas, structures, and biological activities;

[0066] 2) A machine learning algorithm is used, with simplified molecular linear canonical strings as input and the ability to promote nitrification and denitrification as output, to train a prediction model and optimize its parameters. The specific steps are as follows:

[0067] The collected SMILES structures were standardized, and duplicate data were merged to ensure uniqueness, ultimately constructing a standardized dataset containing 1000-4000 SMILES, of which 200-500 are substances that promote nitrification and denitrification. Based on the Chemprop deep learning model, the optimal model was determined by varying the amount of training data and different training parameter settings. To determine the optimal ratio of positive to negative samples, training was conducted using active to inactive substance ratios of 1:1, 4:6, 3:7, and 2:8. Key performance indicators included: average test score of the model training results, model prediction accuracy, and number of consecutive correct recognitions. The optimal model parameters included: average test score of the model training results > 0.8, and the preferred model average test score > 0.9.

[0068] 3) Use the trained machine learning model to screen compounds in the natural product database, predict their denitrification potential, and select a number of candidate substances;

[0069] 4) In the microcosm culture system, by adding candidate substances with different concentration gradients and setting up blank controls, nitrogen removal and microbial response were detected to verify the denitrification effect and mechanism of action of the candidate substances.

[0070] 5) Machine learning methods were used to integrate microcosm validation experimental data to reveal the nitrogen cycle process and microbial driving mechanism of constructed wetlands mediated by candidate substances, and to screen out key active substances and their optimal concentrations.

[0071] Optionally, the plant root exudate database includes compound information from literature reports and natural product databases, of which 200-500 are active substances.

[0072] Optionally, the machine learning prediction model is constructed using the Chemprop deep learning model, and the optimal model parameters are determined through validation using key metrics.

[0073] Optionally, the candidate substances include, but are not limited to, compounds such as fumaric acid, naringenin, genistein, and rutin secreted by plant roots.

[0074] Optionally, the microcosm cultivation system is made of plexiglass material and used to grow wetland plants (50 plants / m²). 2 ), lay soil substrate (10 cm-20 cm), and construct a pilot artificial wetland system.

[0075] A system for enhancing nitrogen removal in constructed wetlands using root exudates, comprising key active substances screened by the method and their prepared formulations.

[0076] Chemprop is a machine learning software package for predicting chemical properties. Its principle is based on graph neural networks, specifically the directed message passing neural network (D-MPNN) architecture. The following are its principles and specific processes:

[0077] 1. Model Structure

[0078] • Module 1: Local Feature Coding

[0079] The SMILES string is converted into a molecular graph using RDKit, with atoms as nodes and bonds as edges. Feature vectors are then constructed for atomic features (atomic number, number of bonds, charge, hybridization, etc.) and bond features (bond type, conjugation, 3D structure, etc.).

[0080] • Module 2: Directed Message Passing Neural Network (D-MPNN)

[0081] Messages are propagated along directed edges, and edge features are transformed through neural network layers and activation functions (ReLU by default). After T message propagation steps (3 by default), the features are iterated and updated, and finally aggregated into atomic embeddings.

[0082] • Module 3: Aggregate Functions

[0083] All atomic embeddings are aggregated into molecular embeddings, offering three methods: summation, scaled summation (divided by a user-specified scaling factor), and averaging. Additional molecular features, such as Morgan fingerprints, can be fused. • Module 4: Standard Feedforward Neural Network (FFN)

[0084] Information is passed unidirectionally from the input layer to the output layer, with no inter-layer feedback. As the final module of the Chemprop model, it maps molecular embeddings to the target variable space. Regression or classification prediction is performed through fully connected layers, and users can select the number of layers (default 2 layers) and the number of hidden neurons (default 300).

[0085] The activation function is the same as that of D-MPNN, with bias enabled by default. Prediction results are processed by the activation function: for binary classification tasks, the Sigmoid function is used to output the activity probability value, generating a prediction value between (0,1) for each compound; for multi-class classification tasks, the Softmax function is used to output the classification scores for each class, with a sum of 1.

[0086] 2. Training process

[0087] •Chemprop is fully end-to-end trainable, with the weights of D-MPNN and FFN updated simultaneously.

[0088] • By default, a single model is trained using random data splitting for a total of 30 epochs. However, smaller datasets may require more epochs. It is recommended to check the convergence of the learning curve.

[0089] • Using the Adam optimizer, the default learning rate increases linearly from 0.5% in the first two warm-up rounds and then decreases exponentially from 0.5% in the remaining rounds. Each optimization step uses a batch of 50 data points by default.

[0090] • Provides early stopping and dropout as regularization methods, and its PyTorch backend supports GPU-accelerated training and inference processes.

[0091] 3. Special Features

[0092] The Chemprop deep learning model supports various input features, multi-molecule model training, chemical reaction prediction, spectral data processing, transfer learning, hyperparameter optimization, and uncertainty estimation. In addition to SMILES strings, the model accepts additional features at the molecular, atomic, or bond level as input and can handle complex chemical systems such as solute / solvent combinations and atom-mapped reactions. It supports converting reactant and product SMILES pairs into pseudomolecules using specific keywords and provides multiple loss function options to adapt to different prediction tasks. The model incorporates various uncertainty estimation tools such as deep ensemble and dropout, and achieves automatic hyperparameter optimization through a tree-based Parzen estimator algorithm.

[0093] Example 1

[0094] According to embodiments of the present invention, a method and application for regulating nitrogen cycling in constructed wetlands using root exudates are provided. The method includes: a root exudate database for storing compound information of plant root exudates; a machine learning prediction model for screening candidate substances with the ability to promote nitrogen removal in wetlands; and a microcosmic culture system for verifying the nitrogen removal effect and mechanism of action of the candidate substances.

[0095] The root exudate database 1 stores information on the compound names, molecular formulas, structures, and biological activities of plant root exudates. Data is sourced from literature reviews and the natural product database (COCONUT), and structures are represented using the SMILES format for standardization.

[0096] Machine learning prediction unit 2 is constructed using Chemprop model 3, with simplified molecular linear canonical strings as input features and the ability to promote nitrification and denitrification as output labels. Optimal model parameters are determined through key indicators. This model is used to screen compounds in a natural product database to predict their denitrification potential. The active substance screening unit 4 selects several candidate substances with the highest scores. Specifically, the training set contains 395 active substances labeled "1". To determine the optimal ratio of positive to negative samples, training is performed with active to inactive substance ratios of 1:1, 4:6, 3:7, and 2:8, respectively. The AUC values ​​on the test set are 0.8203, 0.7895, 0.8312, and 0.8638, respectively. The results show that the model performs best when the active:inactive ratio is 2:8, achieving the highest goodness of fit (AUC = 0.8638) and continuously and accurately identifying 114 positive samples with predicted activity values ​​> 0.99. The model also demonstrates good recognition ability on an independent validation set of 60 compounds. Therefore, this model was chosen for subsequent prediction of highly active nitrogen removal regulators. The COCONUT database contains 412,000 natural products, covering diverse chemical structures and biological activities. Using the SMILES expressions of substances in the database as input, the constructed optimal prediction model was used for large-scale prediction of nitrogen removal regulatory activities.

[0097] The microcosmic cultivation system 5 is made of plexiglass, with a substrate 8 such as soil laid at the bottom to construct an artificial wetland system. Within this microcosmic cultivation system, nitrogen removal and microbial responses are detected by adding candidate substances at different concentration gradients, verifying the nitrogen removal effect of the candidate substances. The concentrations of ammonia nitrogen, nitrite nitrogen, and nitrate nitrogen in the system are monitored periodically, and the nitrification and denitrification rates and their nitrogen removal contribution rates are calculated.

[0098] We used amplicon sequencing and real-time quantitative PCR to analyze the effects of candidate substances on key microbial groups and functional genes in the nitrogen cycle.

[0099] The specific steps are as follows: Quantification of the 16S rRNA gene, bacterial ammonia oxidation gene AOB (amoA), archaea ammonia oxidation gene AOA (amoA), and denitrification genes nirK, nirS, and nosZ was performed on a real-time PCR system using SYBR green as the fluorescent dye. The reaction volume was 25 μL, and the premixed reagent was Takara Premix Ex Taq. TM Gene copy number in unknown samples was determined based on a standard curve. Quantitative results with a correlation coefficient greater than 0.98 and an amplification efficiency greater than 98% were used. The specificity of amplified products was confirmed by melting curve analysis for all samples, and the size of amplicon fragments was determined by agarose gel electrophoresis.

[0100] Amplicon sequencing was performed using primers 515F and 806R to amplify the V4 region of the bacterial 16S rRNA gene. PCR reactions were conducted in 50 μL systems containing 20–30 ng template DNA, 0.3 μM primers, and 25 μL Premix Ex Taq. Amplification conditions included 94°C pre-denaturation for 1 min, followed by 30 cycles (94°C denaturation for 20 s, 57°C annealing for 25 s, 72°C extension for 30 s), and a final extension at 72°C for 10 min. PCR products were purified using a gel extraction kit and quantified using the Qubit dsDNA HS kit. Libraries were constructed according to the Illumina MiSeq platform guidelines, and high-throughput sequencing was performed on the Illumina MiSeq platform.

[0101] Experimental results:

[0102] The results showed that the predicted activity scores of 1537 natural products were >0.90. Through toxicological assessment, natural source verification and literature review, four candidate substances with high activity, clear source and no research gap were finally screened out: fumaric acid (predicted activity 0.95049), naringenin (0.85517), genistein (0.87742) and rutin (0.92458).

[0103] Example 2

[0104] According to an embodiment of the present invention, a method for screening candidate substances in root exudates using machine learning methods is provided. The method includes the following steps:

[0105] Step 1: Based on literature review and experimental analysis, collect chemical information on plant root exudates, including compound names, molecular formulas, SMILES codes, etc., and construct a root exudate database 1.

[0106] Step 2: Using Chemprop Algorithm 3, a binary classification model was trained with simplified molecular linear canonical strings as input and the presence or absence of substances promoting nitrification and denitrification as output. The model performance was then evaluated using key performance indicators. Specifically, the following steps were taken: Literature related to nitrogen removal regulation was retrieved from the Web of Science database, and candidate compounds with nitrification and denitrification-promoting activities were systematically screened. Binary classification labels were established using information from the PubChem database: active substances promoting nitrification and denitrification were labeled "1," and inactive substances were labeled "0." The SMILES expressions for the compounds were obtained, and a training dataset was constructed to develop a machine learning model for screening nitrogen cycle regulatory substances. The training set contained 395 active substances labeled "1." To determine the optimal ratio of positive to negative samples, active to inactive substance ratios of 1:1, 4:6, 3:7, and 2:8 were used for training, with AUC values ​​of 0.8203, 0.7895, 0.8312, and 0.8638 for the test set, respectively. The results showed that the model performed best with an activity:inactivity ratio of 2:8, achieving the highest good fit (AUC = 0.8638) and consistently and accurately identifying 114 positive samples with predicted activities > 0.99. The model performed well on an independent validation set of 60 compounds, therefore it was selected for subsequent prediction tasks.

[0107] Step 3: Using the trained machine learning model, the active substance screening unit 4 predicts the nitrogen-promoting probability scores of all compounds in the COCONUT natural product database. In other words, SMILES are compiled from the COCONUT database to construct a massive natural product library. The optimal training model is used to screen the natural product library, identifying natural products with the potential to promote nitrification and denitrification processes. The specific steps and relevant parameters for the optimal training model to screen the natural product library are as follows: The COCONUT database contains 412,000 natural products, covering diverse chemical structures and biological activities. Using the SMILES expressions of substances in the database as input, the constructed prediction model is used to perform large-scale prediction of nitrogen cycle regulatory activities.

[0108] Step 4: Based on the prediction score (predicted activity value greater than 0.85), select the top 10% of compounds as candidate substances, and comprehensively consider factors such as their source, price, and safety to finally determine 3-10 candidate substances for validation of the microcosm culture system.

[0109] Experimental results:

[0110] In step 2, the optimal model training set contains 1667 data points, of which 395 are active substances. The average test score of the model training results is 0.9, and the number of consecutive correct identifications is 114 (114 / 395). The training set predicts that 117 substances have an activity >0.90.

[0111] In step 3, the database contained a total of 412,000 substances, of which 1,537 substances with an activity greater than 0.90 were screened out.

[0112] In step 4, fumaric acid, naringenin, genistein, and rutin were identified as candidate substances.

[0113] Table 1. Activity prediction of candidate natural products

[0114]

[0115] Example 3

[0116] According to an embodiment of the present invention, a microcosmic cultivation system 5 and its method for verifying the denitrification effect of candidate substances are provided. The system is made of plexiglass and has dimensions of 100 cm in length, 40 cm in width, and 50 cm in height. A 20 cm thick soil substrate 8 is laid at the bottom of the system, and wetland plants 6 such as reeds (50 plants / m²) are planted. 2 Simulated wastewater 5 was continuously introduced to construct a stable artificial wetland system.

[0117] Candidate substances were prepared into solutions with concentrations of 0.1, 1, 10, 100, and 1000 μmol / L, with three replicates for each concentration, and a blank control group was also included. The experimental groups were continuously supplemented with the candidate substance solutions fumaric acid, naringenin, genistein, and rutin once daily for 30 consecutive days. Overlying water samples were collected every two days to determine the concentrations of ammonia nitrogen, nitrite nitrogen, nitrate nitrogen, and total nitrogen. The procedures for determining the concentrations of ammonia nitrogen, nitrite nitrogen, nitrate nitrogen, and total nitrogen are detailed in the following reference: Jiang, C., Zhang, L., Chi, Y., Xu, S., Xie, Y., Yang, D., ... & Zhuang, X. (2024). Rapid start-up of an innovative pilot-scale staged PN / A continuous process for enhanced nitrogen removal from mature landfill leachate via robust NOBelimination and efficient biomass retention. Water Research, 249, 120949.

[0118] Simultaneously, matrix samples were collected at the beginning and end of the experiment, DNA was extracted, amplicon sequencing was used to analyze the microbial community composition, and real-time quantitative PCR was used to analyze the abundance of functional genes.

[0119] The steps for amplicon sequencing analysis of microbial community composition are as follows: Amplicon sequencing was performed using primers 515F and 806R to amplify the V4 region of the bacterial 16S rRNA gene. The PCR reaction was carried out in a 50 μL system containing 20-30 ng template DNA, 0.3 μM primers, and 25 μL Premix Ex Taq. Amplification conditions were: 94℃ pre-denaturation for 1 min, followed by 30 cycles (94℃ denaturation for 20 s, 57℃ annealing for 25 s, 72℃ extension for 30 s), and a final extension at 72℃ for 10 min. The PCR products were purified using a gel extraction kit and quantified using the Qubit dsDNA HS kit. Libraries were constructed according to the Illumina MiSeq platform guidelines, and high-throughput sequencing was performed on the Illumina MiSeq platform.

[0120] The steps of real-time quantitative PCR are as follows: Quantification of 16S rRNA genes, bacterial ammonia oxidation genes AOB (amoA) and AOA (amoA) archaea ammonia oxidation genes, and denitrification genes nirK, nirS, and nosZ are performed on a real-time PCR system using SYBR green as the fluorescent dye. The reaction volume is 25 μL, and the premixed reagent is Takara Premix Ex Taq. TM Gene copy number in unknown samples was determined based on a standard curve. Quantitative results with a correlation coefficient greater than 0.98 and an amplification efficiency greater than 98% were used. The specificity of amplified products was confirmed by melting curve analysis for all samples, and the size of amplicon fragments was determined by agarose gel electrophoresis.

[0121] The impact of candidate substances on nitrification and denitrification processes was evaluated by calculating the rate of change of nitrogen concentration in different forms.

[0122] The influence of candidate substances on microbial community structure was revealed by methods such as principal coordinate analysis and nonmetric multidimensional scaling analysis; the relationship between microbial community structure and nitrogen cycle process was studied by correlation analysis and regression analysis.

[0123] Machine learning methods were used to integrate and validate the experimental data, revealing the nitrogen cycle process and microbial driving mechanism of constructed wetlands mediated by the candidate substances, and screening out key active substances and their optimal concentrations.

[0124] By comprehensively evaluating the denitrification effect and microbial ecological mechanism of candidate substances, key active substances and their optimal concentration range suitable for enhanced denitrification in constructed wetlands were screened out.

[0125] Experimental results:

[0126] Naringenin and genistein were dissolved in methanol and added to the soil matrix 8 (20 g) at a concentration of 100 μg / g dry soil. The control group was added with the same volume of sterile deionized water and methanol. After 14 days of incubation in the dark at room temperature, the denitrification enzyme activity data are shown in Table 2.

[0127] Table 2. Denitrifying enzyme activity in microcosm culture

[0128]

[0129] The results above show that the efficient root exudates obtained by the method of regulating nitrogen cycle in constructed wetlands using root exudates in this invention can significantly promote the denitrification process.

[0130] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in the present invention should be included within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. A method for regulating nitrogen cycle in constructed wetlands using root exudates, characterized in that, The method includes the following steps: 1) Construct a plant root exudate database, which collects the following information about plant root exudates: compound name, molecular formula, structure, physicochemical properties and biological activity, and performs SMILES normalization and activity binary labeling preprocessing on the data; 2) A machine learning algorithm is used, with simplified molecular linear canonical strings of compounds as input and the ability to promote nitrification and denitrification as output. The data from the plant root exudate database is used to train a machine learning prediction model and optimize the parameters. 3) Use the machine learning prediction model described in step 2) to screen compounds in the natural product database, predict their denitrification potential, and select candidate substances; 4) Construct a microcosm culture system. In the microcosm culture system, by adding the candidate substances from step 3) at different concentration gradients, nitrogen removal and microbial response are detected to verify the denitrification effect of the candidate substances. The microcosm culture system includes: a culture device, a sample introduction device, and a monitoring device. The culture device includes: an plexiglass device, a soil substrate, and wetland plants. The sample introduction device includes: a sample pump, a sample inlet, a connecting tube, and a simulated wastewater storage tank. The monitoring device includes: a dissolved oxygen monitoring device, a temperature monitoring device, and a pH probe.

2. The method according to claim 1, characterized in that, The method further includes: 5) The experimental data obtained in step 4) are integrated using the machine learning algorithm to reveal the nitrogen cycle process and microbial driving mechanism of the constructed wetland mediated by the candidate substances, and to screen out key active substances and their optimal concentrations.

3. The method according to claim 1, characterized in that, In step 1), the plant root exudate database includes compound information from literature reports and natural product databases, of which the number of active substances is 200-500. And / or, in step 2), the machine learning prediction model is constructed using the Chemprop deep learning model, and the optimal model parameters are determined by verification through key indicators, wherein the key indicators include: the average test score of the model training results, the model prediction and recognition accuracy, and the number of consecutive correct recognitions, and the optimal model parameters include: the average test score of the model training results > 0.8, and the preferred model training results average test score > 0.

9.

4. The method according to claim 1, characterized in that, In step 3), the candidate substances include: fumaric acid, naringenin, genistein and rosin, secreted by plant roots.

5. The method according to claim 1, characterized in that, In step 4), the micro-universe cultivation system includes plexiglass material, on which a soil matrix is ​​laid, simulated sewage is prepared, wetland plants are planted, and an artificial wetland pilot system is constructed. And / or, periodically monitor the concentrations of ammonia nitrogen, nitrite nitrogen, nitrate nitrogen and / or total nitrogen in the system, and assess the contribution of nitrification and denitrification processes to total nitrogen removal; And / or, use amplicon sequencing technology to analyze changes in microbial community structure; And / or, using real-time quantitative PCR, analyze the effects of candidate substances on key functional genes of the nitrogen cycle.

6. A system for enhancing nitrogen removal in constructed wetlands using root exudates, characterized in that, The system includes key active substances screened by the method according to any one of claims 1-5 and / or formulations prepared from said key active substances.

7. A system for regulating nitrogen cycling in constructed wetlands using root exudates, characterized in that, The system includes: (1) Plant root exudate database: The database collects data on plant root exudates, including the following information: compound name, molecular formula, structure, physicochemical properties and biological activity, and performs SMILES normalization and activity binary labeling preprocessing on the data to provide data support for screening candidate substances; (2) Machine learning prediction unit: The machine learning prediction unit includes a machine learning prediction model, which is trained by using data from the plant root exudate database, with the simplified molecular linear canonical string of the compound as the independent variable and the ability to promote nitrification and denitrification as the dependent variable, and is used to screen compounds in the natural product database, predict their denitrification potential, and screen candidate substances. (3) Microcosm culture system: The microcosm culture system is used to simulate the artificial wetland environment. By adding the candidate substances, its impact on key processes of wetland nitrogen cycle and functional microorganisms is evaluated. The microcosm culture system includes: a culture device, a sample injection device, and a monitoring device. The culture device includes: an plexiglass device, a soil substrate, and wetland plants. The sample injection device includes: a sample injection pump, a sample inlet, a connecting pipe, and a simulated wastewater storage tank. The monitoring device includes: a dissolved oxygen monitoring device, a temperature monitoring device, and a pH probe.

8. The system according to claim 7, characterized in that, The machine learning prediction model is constructed using the Chemprop deep learning model. The optimal model parameters are determined through validation using key indicators, including: the average test score of the model training results, the model prediction accuracy, and the number of consecutive correct recognitions. The optimal model parameters include: the average test score of the model training results > 0.8, and the preferred model training results average test score > 0.

9.

9. The system according to claim 7, characterized in that, The candidate substances include: fumaric acid, naringenin, genistein, and rosin, which are secreted by plant roots.

10. The system according to claim 7, characterized in that, The micro-universe cultivation system includes plexiglass material, with soil substrate laid at the bottom of the plexiglass device, wetland plants planted, simulated wastewater prepared, and an artificial wetland pilot system constructed. And / or, periodically monitor the concentrations of ammonia nitrogen, nitrite nitrogen, nitrate nitrogen and / or total nitrogen in the system, and assess the contribution of nitrification and denitrification processes to total nitrogen removal; And / or, use amplicon sequencing technology to analyze changes in microbial community structure; And / or, using real-time quantitative PCR, analyze the effects of candidate substances on key functional genes of the nitrogen cycle.

Images