Nitrogenase activity prediction method and system based on machine learning

Abstract

The invention provides a nitrogenase activity prediction method and system based on machine learning. The method comprises the following steps: acquiring nitrogenase sample activity data; extracting multi-dimensional features from the nitrogenase sample; constructing a nitrogenase activity prediction model based on machine learning according to the fusion feature vector of the nitrogenase sample protein sequence, the expression level of nitrogenase, the codon preference of a nitrogenase sample target strain and the copy number of nitrogenase sample nitrogen fixation related genes; and predicting and obtaining the nitrogenase activity based on the constructed nitrogenase activity prediction model. The method comprises the following steps: acquiring nitrogenase sample activity data, and extracting multi-dimensional features including a fusion feature vector of a nitrogenase sample protein sequence, a nitrogenase expression level, codon preference of a nitrogenase sample target strain and a nitrogen fixation related gene copy number as model input; the nitrogenase activity prediction model is obtained based on machine learning, and the efficient and accurate nitrogenase activity prediction method is provided.

Description

Technical Field

[0001] The present application relates to the field of biotechnology, and in particular to a method and system for predicting nitrogenase activity based on machine learning. Background Art

[0002] Nitrogen-fixing microorganisms convert atmospheric nitrogen into ammonia through nitrogenase, a process that plays a vital role in the global nitrogen cycle. Nitrogen is the main component of the Earth's atmosphere, but most organisms in nature cannot directly use nitrogen because the triple bonds between its molecules are very stable. Nitrogen-fixing microorganisms can convert atmospheric nitrogen into ammonia through the nitrogenase system, thereby providing plants with an available nitrogen source, thereby supporting agricultural production and promoting the nitrogen cycle in the ecosystem. Nitrogenase itself is a highly complex metalloenzyme complex, which is mainly divided into three categories: molybdenum nitrogenase, vanadium nitrogenase and iron nitrogenase, among which molybdenum nitrogenase is the most common and most thoroughly studied. Different types of nitrogenase rely on different metal elements as their active centers, and metals such as molybdenum, vanadium and iron play a vital catalytic role in it.

[0003] The mechanism of action of nitrogenase is very complex, involving a series of electron transfer and chemical reaction processes. Although a large number of studies have been devoted to revealing the structural and functional relationship of nitrogenase, its complete mechanism of action has not yet been fully elucidated, especially the fine regulation and structural changes required for nitrogenase in the process of catalyzing nitrogen reduction are still hot research topics in the scientific community. Due to certain limitations in the activity and stability of nitrogenase, it faces many challenges in improving nitrogen fixation efficiency and applying it to agriculture and industry. At present, the improvement of nitrogenase activity still depends on the in-depth understanding and research of its molecular mechanism, as well as how to optimize its performance through genetic engineering or other technical means to meet the demand for nitrogen sources in agricultural production and promote its widespread application in environmental remediation, energy production and other fields.

[0004] In recent years, the rapid development of artificial intelligence technology, especially machine learning, has provided new tools for the study of complex biological systems. Elnaggar A. et al. published the pre-trained large model ProtT5 in IEEE transactions on pattern analysis and machine intelligence. These models can extract advanced features from large-scale protein sequences and perform well in biological function prediction. However, there is currently a lack of methods for comprehensively predicting and optimizing nitrogenase activity based on these new technologies. In addition, important factors that affect nitrogen fixation efficiency, such as nitrogenase expression level and codon preference, have not been fully integrated into the existing analysis framework.

[0005] Protein design is an extremely complex and challenging task. The difficulty lies in the fact that the function of proteins depends on complex structures, which are formed by amino acid sequences through highly nonlinear and dynamic interactions. Designing efficient nitrogen-fixing bacteria is particularly difficult. Nitrogenase is a highly complex and extremely oxygen-sensitive enzyme. Its activity not only depends on the specific amino acid composition, but is also significantly affected by factors such as the expression level of nitrogenase and codon preference. In addition, nitrogenase requires complex protection mechanisms from host cells, which makes engineering nitrogen-fixing bacteria to improve nitrogen fixation efficiency more cumbersome and unpredictable. Traditional experimental methods require a lot of time and resources to screen and optimize the design, while our model significantly improves the design efficiency. Our method can quickly predict nitrogenase activity, identify key genes and amino acid features, and provide optimization directions combined with explanatory analysis.

[0006] Therefore, it is necessary to provide an innovative method based on machine learning, combining the protein sequence characteristics of nitrogenase samples, the expression level of nitrogenase and codon preference, to comprehensively optimize the prediction method of nitrogenase activity and provide a theoretical basis and practical guidance for the genetic engineering of nitrogen-fixing microorganisms. Summary of the invention

[0007] The present application provides a method and system for predicting nitrogenase activity based on machine learning, which can solve the technical problem in the prior art that important factors affecting nitrogen fixation efficiency, such as nitrogenase expression level and codon preference, have not been fully integrated into the existing nitrogenase activity prediction and analysis framework.

[0008] In a first aspect, the present application provides a method for predicting nitrogenase activity based on machine learning, comprising the following steps:

[0009] Obtain nitrogenase sample activity data;

[0010] Extract fusion feature vectors, expression levels of nitrogenase, codon preferences of target strains of nitrogenase samples, and copy numbers of nitrogen fixation-related genes of nitrogenase samples from nitrogenase samples;

[0011] A nitrogenase activity prediction model was constructed based on machine learning according to the fusion feature vector of the nitrogenase sample protein sequence, the expression level of nitrogenase, the codon preference of the target strain of the nitrogenase sample, and the copy number of nitrogen fixation-related genes of the nitrogenase sample.

[0012] Based on the constructed nitrogenase activity prediction model, the nitrogenase activity is predicted.

[0013] In combination with the first aspect, in one embodiment, obtaining nitrogenase sample activity data specifically includes the following steps:

[0014] The activity values ​​of nitrogenase samples were measured by acetylene reduction method;

[0015] The nitrogenase samples are distinguished as positive samples and negative samples based on the activity values ​​of the nitrogenase samples obtained by measurement.

[0016] In combination with the first aspect, in one embodiment, extracting the fusion feature vector, the expression level of nitrogenase, the codon preference of the target strain of the nitrogenase sample, and the copy number of the nitrogen fixation-related gene of the nitrogenase sample from the nitrogenase sample specifically includes the following steps:

[0017] Obtaining a fusion feature vector of the protein sequence of the nitrogenase sample;

[0018] Obtain the expression level of nitrogenase;

[0019] Obtain the codon preference of the target strain for nitrogenase samples;

[0020] Obtain the copy number of nitrogen fixation-related genes in nitrogenase samples.

[0021] In combination with the first aspect, in one embodiment, the step of obtaining a fusion feature vector of a protein sequence of a nitrogenase sample specifically comprises the following steps:

[0022] The 1024-dimensional feature vectors of nifD, nifH, and nifK protein sequences were extracted from nitrogenase samples using the pre-trained model ProtT5;

[0023] The 343-dimensional feature vectors of nifD, nifH, and nifK protein sequences were extracted by calculating the triplet amino acid codes of nitrogenase samples;

[0024] The 400-dimensional feature vectors of nifD, nifH, and nifK protein sequences were extracted by calculating the dipeptide composition of the nitrogenase samples;

[0025] The 50-dimensional feature vectors of nifD, nifH, and nifK protein sequences were extracted by calculating the pseudo-amino acid composition of nitrogenase samples;

[0026] The 1024-dimensional feature vector, the 343-dimensional feature vector, the 400-dimensional feature vector and the 50-dimensional feature vector are fused to obtain a fused feature vector of the protein sequence of the nitrogenase sample.

[0027] In combination with the first aspect, in one embodiment, the nitrogenase activity prediction model is constructed based on machine learning according to the fusion feature vector of the nitrogenase sample protein sequence, the expression level of nitrogenase, the codon preference of the target strain of the nitrogenase sample, and the copy number of the nitrogen fixation-related gene of the nitrogenase sample, and specifically includes the following steps:

[0028] Divide the acquired nitrogenase activity data and the fused feature vector, nitrogenase expression level, gene copy number and codon preference into the first training set and the first test set;

[0029] The training set was used as the model input, and the classification model was constructed using the XGBoost algorithm, and the classification model was used as the nitrogenase activity prediction model.

[0030] In combination with the first aspect, in one embodiment, the nitrogenase activity prediction model is constructed based on machine learning according to the fusion feature vector of the nitrogenase sample protein sequence, the expression level of nitrogenase, the codon preference of the target strain of the nitrogenase sample, and the copy number of the nitrogen fixation-related gene of the nitrogenase sample, and specifically includes the following steps:

[0031] Divide the obtained nitrogenase activity data and fused feature vectors into a first-layer training set and a first-layer test set;

[0032] Taking the first-layer training set as input, the first-layer model is trained using the support vector regression method to obtain the predicted value of the first-layer model;

[0033] Divide the acquired nitrogenase activity data, the predicted value of the first-layer model, and the expression level of nitrogenase, gene copy number, and codon preference into a second-layer training set and a second-layer test set;

[0034] The second-layer training set was used as the model input, and the stacking optimization method was used to obtain the second-layer model, which was used as the nitrogenase activity prediction model.

[0035] In combination with the first aspect, in one embodiment, after constructing the nitrogenase activity prediction model based on machine learning according to the fusion feature vector of the nitrogenase sample protein sequence, the expression level of nitrogenase, the codon preference of the target strain of the nitrogenase sample and the copy number of the nitrogen fixation related gene of the nitrogenase sample, the method further includes the following steps:

[0036] The obtained nitrogenase activity prediction model was validated based on the second-layer test set to obtain the evaluation results of the model prediction performance.

[0037] In combination with the first aspect, in one embodiment, after constructing the nitrogenase activity prediction model based on machine learning according to the fusion feature vector of the nitrogenase sample protein sequence, the expression level of nitrogenase, the codon preference of the target strain of the nitrogenase sample and the copy number of the nitrogen fixation related gene of the nitrogenase sample, the method further includes the following steps:

[0038] The prediction model of nitrogenase activity was analyzed based on the Python package SHAP, and the characteristic contribution value of the nif gene was calculated;

[0039] Based on the characteristic contribution values ​​of nif genes, the smallest gene cluster was screened from nitrogenase samples.

[0040] In a second aspect, the present application provides a nitrogenase activity prediction system based on machine learning, comprising:

[0041] A nitrogenase sample activity data acquisition module, used to acquire nitrogenase sample activity data;

[0042] The nitrogenase multidimensional feature acquisition module is used to extract fusion feature vectors, nitrogenase expression levels, codon preferences of target strains of nitrogenase samples, and the copy numbers of nitrogen fixation-related genes of nitrogenase samples from nitrogenase samples;

[0043] A nitrogenase activity prediction model construction module is communicated with the nitrogenase activity data and the nitrogenase multidimensional feature acquisition module, and is used to construct a nitrogenase activity prediction model based on machine learning according to the fusion feature vector of the nitrogenase sample protein sequence, the expression level of nitrogenase, the codon preference of the target strain of the nitrogenase sample, and the copy number of the nitrogen fixation-related gene of the nitrogenase sample;

[0044] The nitrogenase activity prediction model is communicatively connected to the nitrogenase activity prediction model construction module, and is used to predict and obtain the nitrogenase activity based on the constructed nitrogenase activity prediction model.

[0045] In conjunction with the second aspect, in one implementation, the multi-dimensional feature acquisition module includes:

[0046] A fusion feature vector acquisition unit, used to acquire a fusion feature vector of the protein sequence of the nitrogenase sample;

[0047] A nitrogenase expression level acquisition unit, used to acquire the expression level of nitrogenase;

[0048] A strain codon preference acquisition unit, used to obtain the codon preference of the target strain of the nitrogenase sample;

[0049] A gene copy number acquisition unit, used to obtain the copy number of nitrogen fixation-related genes of nitrogenase samples;

[0050] The feature vector fusion model is communicatively connected with the fused feature vector acquisition unit, the nitrogenase expression level acquisition unit, the strain codon preference acquisition unit and the gene copy number acquisition unit, and is used to fuse the 1024-dimensional feature vector, the 343-dimensional feature vector, the 400-dimensional feature vector and the 50-dimensional feature vector to obtain the fused feature vector of the protein sequence of the nitrogenase sample.

[0051] The beneficial effects brought by the technical solution provided by the embodiment of the present application include at least:

[0052] This application obtains the activity data of nitrogenase samples and extracts multidimensional features including the fusion feature vector of the nitrogenase sample protein sequence, the expression level of nitrogenase, the codon preference of the target strain of the nitrogenase sample, and the copy number of nitrogen fixation-related genes as model input, obtains the nitrogenase activity prediction model based on machine learning, incorporates the important factors affecting the efficiency of nitrogenase into the nitrogenase activity prediction analysis, and provides an efficient and accurate nitrogenase activity prediction method, which can provide support for the functional research of nitrogen fixation gene clusters. BRIEF DESCRIPTION OF THE DRAWINGS

[0053] Figure 1 A schematic diagram of a method for predicting nitrogenase activity based on machine learning provided in an embodiment of the present application;

[0054] Figure 2 The ROC curve diagram of the nitrogenase activity prediction model obtained by using the XGBoost machine learning algorithm provided in the embodiment of the present application;

[0055] Figure 3 A scatter plot of the predicted values ​​of the nitrogenase activity prediction model obtained using the stacked SVR machine learning algorithm provided in the embodiment of the present application;

[0056] Figure 4 A diagram illustrating the effect of the copy number of each gene on the prediction result in the stacked SVR machine learning model using SHAP analysis provided in the embodiments of the present application. DETAILED DESCRIPTION

[0057] In order to enable those skilled in the art to better understand the solution of the present application, the technical solution in the embodiments of the present application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this application.

[0058] The terms "including" and "having" and any variations thereof in the specification and claims of this application and the above-mentioned drawings are intended to cover non-exclusive inclusions. For example, a process, method, system, product or device that includes a series of steps or units is not limited to the listed steps or units, but optionally includes steps or units that are not listed, or optionally includes other steps or units inherent to these processes, methods, products or devices. The terms "first", "second" and "third" are used to distinguish different objects, etc., and do not represent a sequence, nor do they limit "first", "second" and "third" to different types.

[0059] In the description of the embodiments of the present application, "exemplary", "for example" or "for example" are used to indicate examples, illustrations or descriptions. Any embodiment or design described as "exemplary", "for example" or "for example" in the embodiments of the present application should not be interpreted as being more preferred or more advantageous than other embodiments or designs. Specifically, the use of words such as "exemplary", "for example" or "for example" is intended to present related concepts in a specific way.

[0060] In the description of the embodiments of the present application, unless otherwise specified, “ / ” means or, for example, A / B can mean A or B; the “and / or” in the text is merely a description of the association relationship of associated objects, indicating that three relationships may exist, for example, A and / or B can mean: A exists alone, A and B exist at the same time, and B exists alone. In addition, in the description of the embodiments of the present application, “multiple” refers to two or more than two.

[0061] In some processes described in the embodiments of the present application, multiple operations or steps that appear in a specific order are included, but it should be understood that these operations or steps may not be executed in the order in which they appear in the embodiments of the present application or in parallel, and the sequence number of the operation is only used to distinguish the different operations, and the sequence number itself does not represent any execution order. In addition, these processes may include more or fewer operations, and these operations or steps may be executed in sequence or in parallel, and these operations or steps may be combined.

[0062] In order to make the objectives, technical solutions and advantages of the present application clearer, the implementation methods of the present application will be further described in detail below with reference to the accompanying drawings.

[0063] First, please refer to Figure 1 The present invention provides a method for predicting nitrogenase activity based on machine learning, comprising the following steps:

[0064] Step S1: obtaining nitrogenase sample activity data;

[0065] Step S2: extracting fusion feature vectors, expression levels of nitrogenase, codon preferences of target strains of nitrogenase samples, and copy numbers of nitrogen fixation-related genes of nitrogenase samples from nitrogenase samples;

[0066] Step S3: constructing a nitrogenase activity prediction model based on machine learning according to the fusion feature vector of the nitrogenase sample protein sequence, the expression level of nitrogenase, the codon preference of the target strain of the nitrogenase sample, and the copy number of the nitrogen fixation-related gene of the nitrogenase sample;

[0067] Step S5: Based on the constructed nitrogenase activity prediction model, the nitrogenase activity is predicted.

[0068] This application obtains the activity data of nitrogenase samples and extracts multidimensional features including the fusion feature vector of the nitrogenase sample protein sequence, the expression level of nitrogenase, the codon preference of the target strain of the nitrogenase sample, and the copy number of nitrogen fixation-related genes as model input, obtains the nitrogenase activity prediction model based on machine learning, incorporates the important factors affecting the efficiency of nitrogenase into the nitrogenase activity prediction analysis, and provides an efficient and accurate nitrogenase activity prediction method, which can provide support for the functional research of nitrogen fixation gene clusters.

[0069] In one embodiment, the step S1 of obtaining nitrogenase sample activity data specifically includes the following steps:

[0070] Step S11: measuring the activity value of the nitrogenase sample by acetylene reduction method, and the activity value unit of the nitrogenase sample obtained by standardized measurement is nmolC2H2 / mg protein / hour;

[0071] Step S12: Distinguish the nitrogenase samples into positive samples and negative samples based on the activity values ​​of the nitrogenase samples obtained by measurement. Specifically, the nitrogenase samples with an activity higher than 50 nmol C2H2 / mg protein / hour are positive, and the nitrogenase samples with an activity lower than this value are negative.

[0072] In one embodiment, the step S2: extracting the fusion feature vector, the expression level of nitrogenase, the codon preference of the target strain of the nitrogenase sample, and the copy number of the nitrogen fixation-related gene of the nitrogenase sample from the nitrogenase sample, specifically comprises the following steps:

[0073] Step S21: obtaining a fusion feature vector of the protein sequence of the nitrogenase sample;

[0074] Step S22: obtaining the expression level of nitrogenase;

[0075] Step S23: obtaining the codon preference of the target strain of the nitrogenase sample;

[0076] Step S24: Obtain the copy number of nitrogen fixation related genes in the nitrogenase sample.

[0077] In one embodiment, the step S21: obtaining the fusion feature vector of the protein sequence of the nitrogenase sample specifically includes the following steps:

[0078] Step S211: Use the pre-trained model ProtT5 (ProtT5-XL-UniRef50) downloaded from the website (https: / / github.com / agemagician / ProtTrans) to extract nifD, nifH, and nifK protein sequences from the nitrogenase sample, generate a 1024-dimensional feature matrix for each, and take the average value to generate a fixed-dimensional feature vector;

[0079] Step S212: Calculate the triple amino acid code (CT) of the nitrogenase sample using the "Conjoint Triad (CT) vector size = 343" module from the website (https: / / bioinfo.usu.edu / profeatx / submit / ) and extract the 343-dimensional feature vector of the nifD, nifH, and nifK protein sequences;

[0080] Step S213: Calculate the dipeptide composition (DPC) using the "Dipeptide Composition (DPC), size = 400" module in the website (https: / / bioinfo.usu.edu / profeatx / submit / ) and extract the 400-dimensional feature vector of the nifD, nifH, and nifK nitrogenase sequences;

[0081] Step S214: Calculate the pseudo-amino acid composition (PAAC) using “Pseudo-Amino Acid Composition (PAAC), vector size = 50” in the website (https: / / bioinfo.usu.edu / profeatx / submit / ) and extract the 50-dimensional feature vectors of the nifD, nifH, and nifK nitrogenase sequences;

[0082] Step S215: Fusing the 1024-dimensional feature vector, the 343-dimensional feature vector, the 400-dimensional feature vector and the 50-dimensional feature vector to obtain a fused feature vector of the protein sequence of the nitrogenase sample, which is specifically implemented as follows:

[0083] Ensure that the acquired 1024-dimensional, 343-dimensional, 400-dimensional and 50-dimensional feature vectors are one-to-one corresponding at the sample level, that is, each nitrogenase sample protein sequence has corresponding feature vectors of these four different dimensions. Then these feature vectors are spliced ​​and integrated in the dimensional direction in order. For example, for a certain sequence, its 1024-dimensional feature vector, 343-dimensional feature vector, 400-dimensional feature vector and 50-dimensional feature vector are connected end to end in sequence to form a new fused feature vector with a dimension of 1024+343+400+50=1817. The fused feature vector acquisition work of the protein sequences of all nitrogenase samples is completed and applied to the prediction task of sample enzyme activity.

[0084] In one embodiment, the step S22: obtaining the expression level of nitrogenase is specifically implemented as follows:

[0085] Step S221: Use the R package coRdon (https: / / bioconductor.org / packages / release / bioc / html / coRdon.html) to calculate the E value and Fop value of the lnA, nifB, nifD, nifH, nifK, nifE, nifN, and nifX genes;

[0086] Step S222: Calculate the CAI values ​​of genes lnA, nifB, nifD, nifH, nifK, nifE, nifN, and nifX using the Python package CAI (https: / / cai.readthedocs.io / en / latest / );

[0087] The E value, Fop value and CAI value obtained by calculation reflect the gene expression level of the nitrogenase sample.

[0088] In one embodiment, the step S23: obtaining the codon preference of the target strain of the nitrogenase sample is specifically implemented as follows:

[0089] The Python package CAI (https: / / cai.readthedocs.io / en / latest / ) was used to calculate RSCU (relative synonymous codon usage) of each species genome;

[0090] The ECD (Euclidean distance) of each target strain genome relative to the codon-free genome was calculated using the following formula:

[0091]

[0092] Where n is the number of codon preferences, x is the codon preference of different amino acids in the target strain, n ≥ i ≥ 1;

[0093] Codon bias of target strains of nitrogenase samples by RSCU and ECD reactions.

[0094] In one embodiment, the step S24: obtaining the copy number of nitrogen fixation-related genes of the nitrogenase sample is specifically implemented as follows:

[0095] The copy numbers of 34 genes related to the nitrogen fixation process of nitrogenase (nifD, nifH, nifK, amtB, fixA, fixB, fixC, fixX, glnK, lnA, nifA, nifB, nifE, nifF, nifJ, nifL, nifM, nifN, nifP, nifQ, nifS, nifT, nifU, nifV, nifW, nifX, nifY, nifZ, rnfA, rnfB, rnfC, rnfD, rnfE, rnfG) were counted.

[0096] In one embodiment, step S3: constructing a nitrogenase activity prediction model based on machine learning according to the fusion feature vector of the nitrogenase sample protein sequence, the expression level of nitrogenase, the codon preference of the target strain of the nitrogenase sample, and the copy number of the nitrogen fixation-related gene of the nitrogenase sample, specifically includes the following steps:

[0097] Step S31A: Divide the obtained nitrogenase activity data and the fused feature vector, the expression level of nitrogenase, the gene copy number and the codon preference nitrogen fixation into a first training set and a first test set;

[0098] Step S32 A: Using the training set as model input, using the XGBoost algorithm to build a classification model, and using the classification model as a nitrogenase activity prediction model.

[0099] In one embodiment, the step S3: after constructing a nitrogenase activity prediction model based on machine learning according to the fusion feature vector of the nitrogenase sample protein sequence, the expression level of nitrogenase, the codon preference of the target strain of the nitrogenase sample and the copy number of the nitrogen fixation-related gene of the nitrogenase sample, further comprises the following steps:

[0100] Step S4A: The prediction performance of the classification model was evaluated using five-fold cross validation based on the first test set, with an average AUC of 0.9365 and an F1 value of 0.85.

[0101] Through steps S31A-S32A and step S4A, the fusion feature vector of the nitrogenase sample protein sequence, the expression level of nitrogenase, the codon preference of the target strain of the nitrogenase sample and the copy number of the nitrogen fixation-related gene of the nitrogenase sample were used as the model training set. The best classification model based on XGBoost achieved an average AUC of 0.9365 and an F1 score of 0.85 in five-fold cross validation, and a nitrogenase activity prediction model with good prediction performance was obtained. The ROC (Receiver Operating Characteristic) curve of the nitrogenase activity prediction model obtained by the XGBoost machine learning algorithm is shown in the figure. Figure 2 shown.

[0102] In one embodiment, step S3: constructing a nitrogenase activity prediction model based on machine learning according to the fusion feature vector of the nitrogenase sample protein sequence, the expression level of nitrogenase, the codon preference of the target strain of the nitrogenase sample, and the copy number of the nitrogen fixation-related gene of the nitrogenase sample, specifically includes the following steps:

[0103] Step S31B: dividing the obtained nitrogenase activity data and the fused feature vector into a first-layer training set and a first-layer test set;

[0104] Step S32B: using the first layer training set as input, using the support vector regression method to train a first layer model, and obtaining the predicted value of the first layer model;

[0105] Step S33B: dividing the obtained nitrogenase activity data, the predicted value of the first-layer model, and the expression level of nitrogenase, the gene copy number, and the codon preference into a second-layer training set and a second-layer test set;

[0106] Step S34B: Using the second-layer training set as the model input, a stacking optimization method is used to obtain the second-layer model (i.e., a stacked SVR machine learning model), and the second-layer model is used as a nitrogenase activity prediction model.

[0107] In one embodiment, the step S3: after constructing a nitrogenase activity prediction model based on machine learning according to the fusion feature vector of the nitrogenase sample protein sequence, the expression level of nitrogenase, the codon preference of the target strain of the nitrogenase sample and the copy number of the nitrogen fixation-related gene of the nitrogenase sample, further comprises the following steps:

[0108] Step S4B: Based on the second-layer test set, the prediction performance of the second-layer model is evaluated by a five-fold cross-validation method to achieve an average R 2 The predicted value scatter plot of the nitrogenase activity prediction model obtained by using the stacked SVR machine learning algorithm is as follows Figure 3 shown.

[0109] Through steps S31B-S34B and step S4B, the fusion feature vector of the nitrogenase sample protein sequence, the expression level of nitrogenase, the codon preference of the target strain of the nitrogenase sample, and the copy number of the nitrogen fixation-related gene of the nitrogenase sample were used as the model training set. The regression model with the best performance based on the SVR stacking method was used as the nitrogenase activity prediction model, achieving an average R 2 Excellent results of 0.5572 and MAE of 0.2530.

[0110] In one embodiment, the step S5: after predicting and obtaining the nitrogenase activity based on the constructed nitrogenase activity prediction model, further comprises the following steps:

[0111] Step S61: Analyze the nitrogenase activity prediction model based on the Python package SHAP (https: / / pypi.org / project / shap / ) and calculate the characteristic contribution value of the nif gene; use SHAP to analyze the influence of the copy number of each gene on the prediction result in the stacked SVR machine learning model. Figure 4 As shown;

[0112] Figure 4 (a) is the relationship between the nifA gene copy number and the SHAP value; Figure 4 Middle (b) is the relationship between the nifB gene copy number and SHAP value; Figure 4 Middle (c) is the relationship between the nifD gene copy number and SHAP value; Figure 4 Middle (d) is the relationship between the nifE gene copy number and SHAP value; Figure 4 Middle (e) is the relationship between the nifH gene copy number and SHAP value; Figure 4 Middle (f) is the relationship between the nifK gene copy number and SHAP value; Figure 4 Middle (g) is the relationship between the nifN gene copy number and SHAP value; Figure 4 Middle (h) is the relationship between the nifQ gene copy number and SHAP value; Figure 4 (i) is the relationship between the nifS gene copy number and SHAP value; Figure 4 Middle (j) is the relationship between nifI gene copy number and SHAP value; Figure 4 Middle (k) is the relationship between the nifW gene copy number and SHAP value; Figure 4 Middle (l) is the relationship between the nifX gene copy number and SHAP value; Figure 4 Middle (m) is the relationship between the nifU gene copy number and SHAP value; Figure 4 Middle (n) is the relationship between the nifV gene copy number and SHAP value; Figure 4 Middle (o) is the relationship between the nifZ gene copy number and SHAP value;

[0113] Step S62: According to the characteristic contribution value of the nif gene, the minimum gene cluster is selected from the nitrogenase sample; specifically, the nif gene whose characteristic contribution value is higher than the characteristic contribution value threshold is selected as the minimum gene cluster of the nitrogenase sample.

[0114] In a second aspect, the present application provides a nitrogenase activity prediction system based on machine learning, comprising:

[0115] A nitrogenase sample activity data acquisition module, used to acquire nitrogenase sample activity data;

[0116] The nitrogenase multidimensional feature acquisition module is used to extract fusion feature vectors, nitrogenase expression levels, codon preferences of target strains of nitrogenase samples, and the copy numbers of nitrogen fixation-related genes of nitrogenase samples from nitrogenase samples;

[0117] A nitrogenase activity prediction model construction module is communicated with the nitrogenase activity data and the nitrogenase multidimensional feature acquisition module, and is used to construct a nitrogenase activity prediction model based on machine learning according to the fusion feature vector of the nitrogenase sample protein sequence, the expression level of nitrogenase, the codon preference of the target strain of the nitrogenase sample, and the copy number of the nitrogen fixation-related gene of the nitrogenase sample;

[0118] The nitrogenase activity prediction model is communicatively connected to the nitrogenase activity prediction model construction module, and is used to predict and obtain the nitrogenase activity based on the constructed nitrogenase activity prediction model.

[0119] In one embodiment, the multi-dimensional feature acquisition module includes:

[0120] A fusion feature vector acquisition unit, used to acquire a fusion feature vector of the protein sequence of the nitrogenase sample;

[0121] A nitrogenase expression level acquisition unit, used to acquire the expression level of nitrogenase;

[0122] A strain codon preference acquisition unit, used to obtain the codon preference of the target strain of the nitrogenase sample;

[0123] A gene copy number acquisition unit, used to obtain the copy number of nitrogen fixation-related genes of nitrogenase samples;

[0124] The feature vector fusion model is communicatively connected with the fused feature vector acquisition unit, the nitrogenase expression level acquisition unit, the strain codon preference acquisition unit and the gene copy number acquisition unit, and is used to fuse the 1024-dimensional feature vector, the 343-dimensional feature vector, the 400-dimensional feature vector and the 50-dimensional feature vector to obtain the fused feature vector of the protein sequence of the nitrogenase sample.

[0125] Among them, the functional implementation of each module in the above-mentioned nitrogenase activity prediction system based on machine learning corresponds to the various steps in the above-mentioned nitrogenase activity prediction method embodiment based on machine learning, and its functions and implementation processes will not be repeated here one by one.

[0126] In a third aspect, an embodiment of the present application provides a nitrogenase activity prediction device based on machine learning. The nitrogenase activity prediction device based on machine learning can be a personal computer (PC), a laptop computer, a server, or other device with data processing capabilities.

[0127] The communication interface includes an input / output (I / O) interface, a physical interface, and a logical interface, etc., which are used to interconnect devices within the machine learning-based nitrogenase activity prediction device, and an interface for interconnecting the machine learning-based nitrogenase activity prediction device with other devices (such as other computing devices or user devices). The physical interface can be an Ethernet interface, an optical fiber interface, an ATM interface, etc.; the user device can be a display, a keyboard, etc.

[0128] The memory can be various types of storage media, such as random access memory (RAM), read-only memory (ROM), non-volatile RAM (NVRAM), flash memory, optical storage, hard disk, programmable ROM (PROM), erasable PROM (EPROM), electrically erasable PROM (EEPROM), etc.

[0129] The processor may be a general-purpose processor, which may call the nitrogenase activity prediction program based on machine learning stored in the memory, and execute the nitrogenase activity prediction method based on machine learning provided in the embodiment of the present application. For example, the general-purpose processor may be a central processing unit (CPU). The method executed when the nitrogenase activity prediction program based on machine learning is called may refer to the various embodiments of the nitrogenase activity prediction method based on machine learning in the present application, which will not be repeated here.

[0130] In a fourth aspect, an embodiment of the present application also provides a readable storage medium.

[0131] The readable storage medium of the present application stores a nitrogenase activity prediction program based on machine learning, wherein when the nitrogenase activity prediction program based on machine learning is executed by a processor, the steps of the nitrogenase activity prediction method based on machine learning as described above are implemented.

[0132] Among them, the method implemented when the nitrogenase activity prediction program based on machine learning is executed can refer to the various embodiments of the nitrogenase activity prediction method based on machine learning in this application, and will not be repeated here.

[0133] It should be noted that the serial numbers of the above-mentioned embodiments of the present application are only for description and do not represent the advantages or disadvantages of the embodiments.

[0134] Through the description of the above implementation methods, those skilled in the art can clearly understand that the above-mentioned embodiment methods can be implemented by means of software plus a necessary general hardware platform, and of course by hardware, but in many cases the former is a better implementation method. Based on such an understanding, the technical solution of the present application is essentially or the part that contributes to the prior art can be embodied in the form of a software product, which is stored in a storage medium (such as ROM / RAM, disk, CD) as described above, and includes a number of instructions for a terminal device to execute the methods described in each embodiment of the present application.

[0135] The above are only preferred embodiments of the present application, and are not intended to limit the patent scope of the present application. Any equivalent structure or equivalent process transformation made using the contents of the present application specification and drawings, or directly or indirectly applied in other related technical fields, are also included in the patent protection scope of the present application.

Claims

1. A method for predicting nitrogenase activity based on machine learning, characterized in that: The following steps are involved: Obtain nitrogenase sample activity data; Extract fusion feature vectors, expression levels of nitrogenase, codon preferences of target strains of nitrogenase samples, and copy numbers of nitrogen fixation-related genes of nitrogenase samples from nitrogenase samples; A nitrogenase activity prediction model was constructed based on machine learning according to the fusion feature vector of the nitrogenase sample protein sequence, the expression level of nitrogenase, the codon preference of the target strain of the nitrogenase sample, and the copy number of nitrogen fixation-related genes of the nitrogenase sample. Based on the constructed nitrogenase activity prediction model, the nitrogenase activity is predicted.

2. The method for predicting nitrogenase activity based on machine learning according to claim 1, characterized in that: The method of obtaining nitrogenase sample activity data specifically comprises the following steps: The activity values ​​of nitrogenase samples were measured by acetylene reduction method; The nitrogenase samples are distinguished as positive samples and negative samples based on the activity values ​​of the nitrogenase samples obtained by measurement.

3. The method for predicting nitrogenase activity based on machine learning according to claim 1, characterized in that: The step of extracting the fusion feature vector, the expression level of nitrogenase, the codon preference of the target strain of the nitrogenase sample, and the copy number of the nitrogen fixation-related gene of the nitrogenase sample from the nitrogenase sample specifically comprises the following steps: Obtaining a fusion feature vector of the protein sequence of the nitrogenase sample; Obtain the expression level of nitrogenase; Obtain the codon preference of the target strain for nitrogenase samples; Obtain the copy number of nitrogen fixation-related genes in nitrogenase samples.

4. The method for predicting nitrogenase activity based on machine learning according to claim 1, characterized in that: The step of obtaining the fusion feature vector of the protein sequence of the nitrogenase sample specifically comprises the following steps: The 1024-dimensional feature vectors of nifD, nifH, and nifK protein sequences were extracted from nitrogenase samples using the pre-trained model ProtT5; The 343-dimensional feature vectors of nifD, nifH, and nifK protein sequences were extracted by calculating the triplet amino acid codes of nitrogenase samples; The 400-dimensional feature vectors of nifD, nifH, and nifK protein sequences were extracted by calculating the dipeptide composition of the nitrogenase samples; The 50-dimensional feature vectors of nifD, nifH, and nifK protein sequences were extracted by calculating the pseudo-amino acid composition of nitrogenase samples; The 1024-dimensional feature vector, the 343-dimensional feature vector, the 400-dimensional feature vector and the 50-dimensional feature vector are fused to obtain a fused feature vector of the protein sequence of the nitrogenase sample.

5. The method for predicting nitrogenase activity based on machine learning according to claim 1, characterized in that: The method constructs a nitrogenase activity prediction model based on machine learning according to the fusion feature vector of the nitrogenase sample protein sequence, the expression level of nitrogenase, the codon preference of the target strain of the nitrogenase sample, and the copy number of the nitrogen fixation-related gene of the nitrogenase sample, and specifically includes the following steps: Divide the acquired nitrogenase activity data and the fused feature vector, nitrogenase expression level, gene copy number and codon preference into the first training set and the first test set; The training set was used as the model input, and the classification model was constructed using the XGBoost algorithm, and the classification model was used as the nitrogenase activity prediction model.

6. The method for predicting nitrogenase activity based on machine learning according to claim 1, characterized in that: The method constructs a nitrogenase activity prediction model based on machine learning according to the fusion feature vector of the nitrogenase sample protein sequence, the expression level of nitrogenase, the codon preference of the target strain of the nitrogenase sample, and the copy number of the nitrogen fixation-related gene of the nitrogenase sample, and specifically includes the following steps: Divide the obtained nitrogenase activity data and fused feature vectors into a first-layer training set and a first-layer test set; Taking the first-layer training set as input, the first-layer model is trained using the support vector regression method to obtain the predicted value of the first-layer model; Divide the acquired nitrogenase activity data, the predicted value of the first-layer model, and the expression level of nitrogenase, gene copy number, and codon preference into a second-layer training set and a second-layer test set; The second-layer training set was used as the model input, and the stacking optimization method was used to obtain the second-layer model, which was used as the nitrogenase activity prediction model.

7. The method for predicting nitrogenase activity based on machine learning according to claim 6, characterized in that: After constructing the nitrogenase activity prediction model based on machine learning according to the fusion feature vector of the nitrogenase sample protein sequence, the expression level of nitrogenase, the codon preference of the target strain of the nitrogenase sample and the copy number of the nitrogen fixation related gene of the nitrogenase sample, the following steps are also included: The obtained nitrogenase activity prediction model was validated based on the second-layer test set to obtain the evaluation results of the model prediction performance.

8. The method for predicting nitrogenase activity based on machine learning according to claim 1, characterized in that: After constructing the nitrogenase activity prediction model based on machine learning according to the fusion feature vector of the nitrogenase sample protein sequence, the expression level of nitrogenase, the codon preference of the target strain of the nitrogenase sample and the copy number of the nitrogen fixation related gene of the nitrogenase sample, the following steps are also included: The prediction model of nitrogenase activity was analyzed based on the Python package SHAP, and the characteristic contribution value of the nif gene was calculated; Based on the characteristic contribution values ​​of nif genes, the smallest gene cluster was screened from nitrogenase samples.

9. A nitrogenase activity prediction system based on machine learning, characterized in that: include: A nitrogenase sample activity data acquisition module, used to acquire nitrogenase sample activity data; The nitrogenase multidimensional feature acquisition module is used to extract fusion feature vectors, nitrogenase expression levels, codon preferences of target strains of nitrogenase samples, and the copy numbers of nitrogen fixation-related genes of nitrogenase samples from nitrogenase samples; A nitrogenase activity prediction model construction module is communicated with the nitrogenase activity data and the nitrogenase multidimensional feature acquisition module, and is used to construct a nitrogenase activity prediction model based on machine learning according to the fusion feature vector of the nitrogenase sample protein sequence, the expression level of nitrogenase, the codon preference of the target strain of the nitrogenase sample, and the copy number of the nitrogen fixation-related gene of the nitrogenase sample; The nitrogenase activity prediction model is communicatively connected to the nitrogenase activity prediction model construction module, and is used to predict and obtain the nitrogenase activity based on the constructed nitrogenase activity prediction model.

10. The nitrogenase activity prediction system based on machine learning according to claim 9, characterized in that: The multi-dimensional feature acquisition module comprises: A fusion feature vector acquisition unit, used to acquire a fusion feature vector of the protein sequence of the nitrogenase sample; A nitrogenase expression level acquisition unit, used to acquire the expression level of nitrogenase; A strain codon preference acquisition unit, used to obtain the codon preference of the target strain of the nitrogenase sample; A gene copy number acquisition unit, used to obtain the copy number of nitrogen fixation-related genes of nitrogenase samples; The feature vector fusion model is communicatively connected with the fused feature vector acquisition unit, the nitrogenase expression level acquisition unit, the strain codon preference acquisition unit and the gene copy number acquisition unit, and is used to fuse the 1024-dimensional feature vector, the 343-dimensional feature vector, the 400-dimensional feature vector and the 50-dimensional feature vector to obtain the fused feature vector of the protein sequence of the nitrogenase sample.

Images