A method and system for predicting microbial culture temperature based on domain features

Abstract

The present application belongs to the technical field of microbial culture method, and particularly relates to a microbial culture temperature prediction method and system based on domain characteristics, which extracts domain information by aligning protein sequences and forms a biological characteristic matrix according to a preset combination rule, and realizes division of temperature intervals and prediction of culture temperature based on the biological characteristic matrix through a machine learning model. Through the combination rule, the present application identifies the characteristics of a specific functional module and multiple key domains, improves the biological authenticity and prediction effect of the characteristics, so that the biological characteristic matrix constructed has high biological interpretability. Meanwhile, the machine learning model realizes division of temperature intervals and prediction of culture temperature, significantly reduces the error of temperature boundary samples, supports prediction of a larger temperature range, realizes smooth prediction of boundary samples, and significantly reduces the error caused by misclassification.

Description

Technical Field

[0001] This invention belongs to the technical field of microbial culture methods, specifically relating to a method and system for predicting microbial culture temperature based on domain characteristics. Background Technology

[0002] Culture temperature is a core environmental parameter for microbial culture. Temperature directly determines the metabolic rate and growth status of microorganisms by affecting enzyme activity. Below the optimal temperature, microbial growth is inhibited; above the optimal temperature, protein denaturation or even cell death may occur. Therefore, controlling the temperature within the optimal range is essential to ensure rapid and normal microbial growth and is the foundation for all subsequent applied research.

[0003] Therefore, finding the optimal culture temperature for microorganisms is crucial for their effective cultivation. Currently, existing methods in the field of microbial growth temperature prediction are based on the technical route of k-mer slices of the whole genome sequence and deep learning models. Chinese invention publication CN118155727 discloses a method for predicting the preferred temperature of microorganisms, a model training method, and related devices. This method involves segmenting the target bacterial genome into k-mer sequences of several base lengths to form features, constructing a one-dimensional convolutional neural network deep learning model for training, and then predicting the optimal growth temperature of the bacteria. This method achieves a certain degree of automated prediction of the optimal bacterial growth temperature, but it still has problems and shortcomings. Since k-mer frequency features themselves do not have clear biological meaning, they cannot effectively reveal the correlation mechanism between specific functional genes or structural domains and temperature adaptation, thus the prediction results lack biological interpretability. At the same time, a single model is easily affected by temperature classification boundary samples. When the actual temperature is close to the classification threshold, the prediction results are discontinuous, with large errors, causing instability in the prediction boundary samples. Furthermore, the k-mer frequency feature matrix has a large dimension and sparse information, resulting in high computational resource consumption for model training and inference. Summary of the Invention

[0004] The technical problem to be solved by the present invention is to overcome the shortcomings of existing methods for predicting the optimal culture temperature of microorganisms, such as lack of biological explanation, unstable prediction boundary samples, and high computational resource consumption. Therefore, the present invention provides a method and system for predicting microbial culture temperature based on domain characteristics.

[0005] A method for predicting microbial culture temperature based on domain characteristics includes the following steps:

[0006] Step S1: Obtain microbial information with culture temperature and obtain the corresponding protein sequence of the microorganism;

[0007] Step S2: Perform relevant domain alignment on the protein sequence and extract the domain information of the protein sequence corresponding to temperature adaptation-related genes; generate a biological feature matrix based on the domain information according to a preset combination rule;

[0008] Step S3: Divide the microorganisms into multiple temperature ranges, including low-temperature microorganisms, mesophilic microorganisms, and high-temperature microorganisms; based on the biological feature matrix, train a machine learning model for each temperature range of microorganisms;

[0009] Step S4: Obtain the protein sequence of the microorganism to be predicted, extract domain information based on the protein sequence, generate a biological feature matrix based on the domain information, input the biological feature matrix into the machine learning model, and predict the culture temperature of the microorganism.

[0010] Furthermore, the domain information includes the matching confidence and copy number of each relevant domain in the protein sequence.

[0011] Furthermore, the preset combination rules include full hit rule, arbitrary hit rule, three-domain hit rule, minimum score rule, copy number rule, complementary hit rule, paired hit rule, and two-part hit rule;

[0012] In the full hit rule, when all structural domains within the rule are hit, the score is the sum of the matching confidence of all structural domains; otherwise, the score is zero.

[0013] In the arbitrary hit rule, when at least one structural domain within the rule is hit, the score is the highest matching confidence among the hit structural domains; otherwise, the score is zero.

[0014] In the three-domain hit rule, when all three structural domains within the rule are hit, the score is the sum of the matching confidence of the three structural domains; when two of the three structural domains within the rule are hit, the score is the product of the sum of the matching confidence of the two hit structural domains and a preset multiplier; otherwise, the score is zero.

[0015] In the minimum score rule, when both structural domains within the rule are hit, the score is the smaller of the matching confidence of the two structural domains; otherwise, the score is zero.

[0016] In the copy number rule, when one structure domain within the rule is hit, the score is the matching confidence of the hit structure domain; when two structure domains within the rule are hit, the score is the product of the matching confidence of the two structure domains and the preset multiplier; otherwise, the score is zero.

[0017] In the complementary hit rule, when both structural domains within the rule are hit, the score is the sum of the matching confidence of the two structural domains; when one structural domain within the rule is hit, the score is the product of the matching confidence of the hit structural domain and the preset multiplier; otherwise, the score is zero.

[0018] In the pairwise hit rule, when both structural domains within the rule are hit, the score is the sum of the matching confidence of the two structural domains; otherwise, the score is zero.

[0019] In the binary hit rule, when both the structural domains in the first component and the structural domains in the second component are hit, the score is the larger of the matching confidence scores of the two structural domains; otherwise, the score is zero.

[0020] Furthermore, the method also includes the following: calculating the score of each rule in the preset combination rules based on the protein sequence; the biological feature matrix includes the rule score of each rule in the preset combination rules; the rule score is the product of the score of each rule in the preset combination rules and the corresponding directional weight; the directional weight is obtained based on the relationship between the corresponding rule and the microbial culture temperature.

[0021] Furthermore, when performing relevant domain alignment on the protein sequence, the matching results with expected values ​​less than a preset threshold are filtered out.

[0022] Furthermore, training a machine learning model includes the following steps:

[0023] Using the XGBoost classification model with biological feature matrix as independent variable, microorganisms are classified into microorganisms in multiple temperature ranges, including low-temperature microorganisms, mesophilic microorganisms, and high-temperature microorganisms. The classification model is trained and outputs the probability distribution of each category.

[0024] For each temperature range, an independent XGBoost regression model is trained using samples of the corresponding category as training data to obtain a regressor for each temperature range.

[0025] Furthermore, the predictions made by the machine learning model include the following methodological steps:

[0026] The input biological feature matrix is ​​predicted by a classification model to obtain the probability that the input biological feature matrix belongs to each temperature range.

[0027] The biological feature matrix is ​​input into the regressor for each temperature range, and the corresponding predicted temperature value is output.

[0028] Based on the probability of each temperature range and the corresponding predicted temperature value, the final predicted culture temperature value is obtained through weighted calculation.

[0029] Furthermore, in the training of the machine learning model, the loss function of the classification model is multi-class log loss, and the evaluation method of the regressor includes mean absolute error and root mean square error.

[0030] A microbial culture temperature prediction system based on domain features, used to predict microbial culture temperature using the aforementioned method based on domain features, characterized in that it includes:

[0031] The rule score calculation module is used to calculate rule scores based on protein sequences using preset combination rules, and the rule scores are used to form a biological feature matrix.

[0032] A machine learning model is used as input to a biological feature matrix and outputs a predicted value for the final culture temperature.

[0033] Furthermore, the machine learning model includes:

[0034] A classification model is used to predict the probability that the input biological feature matrix belongs to each temperature range.

[0035] A regressor is used to take the biological feature matrix as input and output the corresponding predicted temperature value.

[0036] The weighted output unit is used to obtain the final predicted culture temperature value through weighted calculation based on the probability of each temperature range and the corresponding predicted temperature value.

[0037] Beneficial Effects: This invention discloses a method for predicting microbial culture temperature based on domain features. It extracts domain information by aligning protein sequences and forms a biological feature matrix according to preset combination rules. Based on this matrix, a machine learning model is used to divide temperature ranges and predict culture temperatures. This invention identifies the features of specific functional modules and multiple key domains through combination rules, improving the biological realism and predictive effectiveness of the features. The constructed biological feature matrix thus possesses high biological interpretability. Simultaneously, the machine learning model significantly reduces errors in temperature boundary samples, supports prediction over a wider temperature range, achieves smooth prediction of boundary samples, and significantly reduces errors caused by misclassification. Attached Figure Description

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

[0039] Figure 1 This diagram illustrates the main method flow steps of the present invention.

[0040] Figure 2 This is a schematic diagram of the process of the present invention. Detailed Implementation

[0041] To make the above-mentioned objectives, features, and advantages of this application more apparent and understandable, the specific embodiments of this application are described in detail below with reference to the accompanying drawings. Many specific details are set forth in the following description to provide a thorough understanding of this application. However, this application can be implemented in many other ways different from those described herein, and those skilled in the art can make similar modifications without departing from the spirit of this application. Therefore, this application is not limited to the specific embodiments disclosed below.

[0042] In the description of this application, it should be understood that the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of indicated technical features. Therefore, a feature defined as "first" or "second" may explicitly or implicitly include at least one of that feature. In the description of this application, "multiple" means at least two, such as two, three, etc., unless otherwise explicitly specified.

[0043] Example 1:

[0044] Reference Figure 1 and Figure 2 As shown, this embodiment discloses a method for predicting microbial culture temperature based on domain features, including the following method steps:

[0045] Step S1: Obtain microbial information with culture temperature and obtain the corresponding protein sequence of the microorganism;

[0046] Step S2: Perform relevant domain alignment on the protein sequence and extract the domain information of the protein sequence corresponding to temperature adaptation-related genes; generate a biological feature matrix based on the domain information according to a preset combination rule;

[0047] In this embodiment, the feature vector consists of three parts: the Pfam domain alignment score (bitscore) of the rule combination, the copy number of the rule combination in the genome, and the type weight. All rule combinations matched by each strain sample form a continuous feature vector matrix, which is input into the classification and regression model for temperature prediction. The copy number of the rule combination represents the number of times the matched gene appears in the bacterial genome, and is counted using a script based on the hmm alignment results. The type weight of the rule combination is a preset value. In this embodiment, the type weight is defined through literature research. For example, proteins related to heat shock that help with heat resistance are set to positive; those related to low temperature response and RNA unfolding mechanisms are set to negative; and genes that assist metabolism or are not temperature-specific are set to neutral.

[0048] Step S3: Divide the microorganisms into multiple temperature ranges, including low-temperature microorganisms, mesophilic microorganisms, and high-temperature microorganisms; based on the biological feature matrix, train a machine learning model for each temperature range of microorganisms;

[0049] Step S4: Obtain the protein sequence of the microorganism to be predicted, extract domain information based on the protein sequence, generate a biological feature matrix based on the domain information, input the biological feature matrix into the machine learning model, and predict the culture temperature of the microorganism.

[0050] In this embodiment, the microorganism is a strain.

[0051] Specifically, in step S1, microbial information including culture temperature is obtained, and the corresponding protein sequence of the microorganism is obtained.

[0052] In this embodiment, strain information with optimal culture temperature tags is obtained from a public database. Then, the protein sequences (.faa files) of the target strains in the NCBI database are automatically obtained in batches using Python and the ncbi-datasets tool, and the optimal growth temperature tag (OGT) of each strain is standardized.

[0053] As a preferred embodiment, the data acquisition selects high-quality genomes included in NCBI RefSeq, and the culture temperature information is preferably obtained from authoritative bacterial resource databases such as BacDive, thereby improving data reliability.

[0054] In step S2, the protein sequence is compared with relevant structural domains to extract the structural domain information of the protein sequence corresponding to temperature adaptation-related genes; a biological feature matrix is ​​generated based on the structural domain information according to a preset combination rule; the structural domain information includes the matching confidence and copy number of each relevant structural domain in the protein sequence.

[0055] The hmmscan algorithm was used to align each protein sequence with the Pfam-A database using HMM domains. Pfam domain hits and bit scores were extracted for genes related to temperature adaptation. Based on preset combination rules, copy number and type weight information were combined to generate a biologically interpretable feature matrix.

[0056] As a further improvement to this embodiment, when performing relevant domain alignment on the protein sequence, hit results with expected values ​​less than a preset threshold are filtered out. In this embodiment, the expected value (E-value) threshold for Pfam domain hit screening is 1e-5.

[0057] In this embodiment, the preset combination rules include All, Any, Triple, Min-Bitscore, Copy-Number, Reverse-Gyrase, Ahp-Pair, and Sensor-Response.

[0058] In the full hit rule, when all domains within the rule are hit, the score is the sum of the matching confidence of all domains; otherwise, the score is zero. In this embodiment, the full hit rule includes the GroEL+GroES dual subunit / chaperone system. Only when both GroEL+GroES domains are hit can the function of influencing the temperature adaptation of microorganisms be realized.

[0059] In the arbitrary hit rule, when at least one domain within the rule is hit, the score is the highest matching confidence among the hit domains; otherwise, the score is zero. In this embodiment, the arbitrary hit rule includes single-domain marker genes such as HtpG (Hsp90) and RNase R. When any of these domains is present, the function is considered to be present.

[0060] In the three-domain hit rule, when all three domains within the rule are hit, the score is the sum of the matching confidence of the three domains; when two of the three domains within the rule are hit, the score is the product of the sum of the matching confidence of the two hit domains and a preset multiplier; otherwise, the score is zero. In this embodiment, the three-domain hit rule includes the dnaK + dnaJ + grpE ternary folding chaperone. Since biological experiments show that the three subunits are synergistic, confirming one of them can partially compensate but the effect diminishes.

[0061] In the minimum score rule, when both domains within the rule are matched, the score is the smaller of the matching confidence of the two domains; otherwise, the score is zero. In this embodiment, the minimum score rule includes the FabA + FabB fatty acid synthase dual enzymes. When the activity of either enzyme is insufficient, it will become a bottleneck. Therefore, the weakest link represents the overall function.

[0062] In the copy number rule, when one domain within the rule is matched, the score is the matching confidence of the matched domain; when two domains within the rule are matched, the score is the product of the matching confidence of the two domains and a preset fold; otherwise, the score is zero. In this embodiment, the copy number rule includes superoxide dismutase (SOD) and catalase (Kat). In these domains, the higher the copy number, the higher the expression level and the stronger the antioxidant capacity. Therefore, an amplification factor is used for modeling.

[0063] In the complementary hit rule, when both domains within the rule are hit, the score is the sum of the matching confidence of the two domains; when one domain within the rule is hit, the score is the product of the matching confidence of the hit domain and a preset fold; otherwise, the score is zero. In this embodiment, the complementary hit rule includes the helicase domain of Reverse Gyrase + the Toprim domain. These domains are common in febrile bacteria, and the occasional absence of one domain can still weaken its function.

[0064] In the pairwise hit rule, when both domains within the rule are hit, the score is the sum of the matching confidence of the two domains; otherwise, the score is zero. In this embodiment, the pairwise hit rule includes mandatory paired subunits such as AhpC + AhpF. In these two types of domains, the two subunits form a reduction complex, and the absence of either one results in inactivation.

[0065] In the binary hit rule, when both the structural domains in the first component and the structural domains in the second component are hit, the score is the larger of the matching confidence scores of the two structural domains; otherwise, the score is zero. In this embodiment, the binary hit rule includes a DesK / DesR temperature sensing-signal binary system, which requires at least one sensor domain (PF00512 / PF02518) + response modulation domain (PF00072) to be hit simultaneously. The transduction chain needs to "sensing + transmitting" to complete the signal loop closure.

[0066] In this embodiment, based on the protein sequence and a preset combination rule, the score of each rule in the preset combination rule is calculated; the biological feature matrix includes the rule score of each rule in the preset combination rule; the rule score is the product of the score of each rule in the preset combination rule and the corresponding directional weight; the directional weight is obtained based on the relationship between the corresponding rule and the microbial culture temperature.

[0067] The preset combination rules are shown in the table below:

[0068] Table 1: Preset Combination Rules

[0069] Rule Type Applicable scenarios (examples) Quantification formula Design Principles All GroEL+GroES and other indispensable dual subunit / companion systems Score = Sum of bit scores of the two fields only if all listed structure fields are hit; otherwise, score = 0. Only complete complexes possess functionality; fragment hits are considered invalid. Any Single-domain marker genes such as HtpG (Hsp90) and RNase R A hit in any structure field is sufficient → Score = Highest bit score in that field The presence of a flag field indicates that the function exists. Triple dnaK + dnaJ + grpE trivalent folding companion 3-domain hit: Score is the sum of the three domains; Missing 1 domain: 0.5x discount; Missing ≥2 domains: 0 Biological experiments show that the trisubunits work synergistically, providing partial compensation, but the effect diminishes over time. Min-Bitscore FabA + FabB fatty acid synthase dual enzyme If both fields are hit: the score is the bit score; otherwise, 0. Insufficient activity of any one enzyme can become a bottleneck; the weakest link represents the whole. Copy-Number Superoxide dismutase (SOD), catalase (Kat) 1 copy: original score; ≥2 copies: 1.5 × original score Increased copy number → increased expression level → enhanced antioxidant capacity, modeled using amplification factor. Reverse-Gyrase Reverse Gyrase helicase domain + Toprim domain Both fields present: Sum; Only one field present: 0.5 × bit score; None present: 0 Febrile bacteria often have a weakened effect even if a domain is missing. Ahp-Pair AhpC + AhpF and other pairs of subunits are required. The score is the sum of the two fields only if both fields co-occur; otherwise, it is 0. The two subunits form a reduction complex; the absence of either one results in inactivation. Sensor-Response DesK / DesR Temperature Sensing-Signal Binary System At least one receptor domain (PF00512 / PF02518) + response regulation domain (PF00072) is hit simultaneously; the score is the larger of the two. A transduction chain requires both "sensing" and "transmission" to complete a signal loop.

[0070] In step S3, the microorganisms are divided into multiple temperature ranges, including low-temperature microorganisms, mesophilic microorganisms, and high-temperature microorganisms; based on the biological feature matrix, a machine learning model is trained for each of the temperature ranges of microorganisms.

[0071] Specifically, training a machine learning model includes the following steps:

[0072] Using the XGBoost classification model with biological feature matrix as independent variable, microorganisms are classified into microorganisms in multiple temperature ranges, including low-temperature microorganisms, mesophilic microorganisms, and high-temperature microorganisms. The classification model is trained and outputs the probability distribution of each category.

[0073] In this embodiment, low-temperature microorganisms are <20℃, mesophilic microorganisms are 20–50℃, and high-temperature microorganisms are 50℃>℃.

[0074] For each temperature range, an independent XGBoost regression model is trained using samples of the corresponding category as training data to obtain a regressor for each temperature range.

[0075] In this embodiment, the main parameters of the XGBoost classification model include a maximum tree depth of 4, a learning rate of 0.05, a subsampling ratio of 0.7, and 100 early stopping rounds; the main parameters of the XGBoost regression model include a maximum tree depth of 4, a learning rate of 0.02, a subsampling ratio of 0.7, and 200 early stopping rounds. These parameters are determined through cross-validation and grid search. The classification model uses multi-class log loss (mlogloss) as its loss function, while the regression model uses mean absolute error (MAE) and root mean square error (RMSE) for evaluation.

[0076] The predictions made by the machine learning model include the following methodological steps:

[0077] The input biological feature matrix is ​​predicted by a classification model to obtain the probability that the input biological feature matrix belongs to each temperature range.

[0078] The biological feature matrix is ​​input into the regressor for each temperature range, and the corresponding predicted temperature value is output.

[0079] Based on the probability of each temperature range and the corresponding predicted temperature value, the final predicted culture temperature value is obtained through weighted calculation.

[0080] Specifically, the predicted final culture temperature is expressed as follows:

[0081] ;

[0082] Where P(c|x) is the probability of the class predicted by the classifier. This is the output of the regression model for this category, where c is the total number of categories in the interval.

[0083] In the training of the machine learning model, the loss function of the classification model is multi-class log loss, and the evaluation method of the regressor includes mean absolute error and root mean square error.

[0084] In step S4, the protein sequence of the microorganism to be predicted is obtained, domain information is extracted based on the protein sequence, a biological feature matrix is ​​generated based on the domain information, and the biological feature matrix is ​​input into the machine learning model to predict the culture temperature of the microorganism.

[0085] Example 2:

[0086] This embodiment provides a microbial culture temperature prediction system based on domain features, used to predict microbial culture temperature using the aforementioned microbial culture temperature prediction method based on domain features. Its features include:

[0087] The rule score calculation module is used to calculate rule scores based on protein sequences using preset combination rules, and the rule scores are used to form a biological feature matrix.

[0088] A machine learning model is used as input to a biological feature matrix and outputs a predicted value for the final culture temperature.

[0089] The machine learning model includes:

[0090] A classification model is used to predict the probability that the input biological feature matrix belongs to each temperature range.

[0091] A regressor is used to take the biological feature matrix as input and output the corresponding predicted temperature value.

[0092] The weighted output unit is used to obtain the final predicted culture temperature value through weighted calculation based on the probability of each temperature range and the corresponding predicted temperature value.

[0093] This embodiment identifies the features of specific functional modules and multiple key structural domains through combination rules, improving the biological authenticity and predictive performance of the features, thereby constructing a biological feature matrix with high biological interpretability. Simultaneously, a machine learning model is used to divide temperature ranges and predict culture temperatures, significantly reducing errors for temperature boundary samples, supporting predictions over a wider temperature range, achieving smooth predictions for boundary samples, and significantly reducing errors caused by misclassification.

[0094] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0095] The embodiments described above are merely illustrative of several implementation methods of this application, and while the descriptions are relatively specific and detailed, they should not be construed as limiting the scope of the patent application. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of this application, and these all fall within the protection scope of this application. Therefore, the protection scope of this patent application should be determined by the appended claims.

Claims

1. A method for predicting microbial culture temperature based on domain characteristics, characterized in that, The method includes the following steps: obtaining microbial information with culture temperature, obtaining the protein sequence corresponding to the microorganism; performing relevant structural domain alignment on the protein sequence, and extracting the structural domain information of the protein sequence corresponding to the temperature adaptation-related genes; A biological feature matrix is ​​generated based on the structural domain information according to a preset combination rule. Based on the protein sequence, and according to a preset combination rule, the score of each rule in the preset combination rule is calculated; the biological feature matrix includes the rule score of each rule within the preset combination rule; the rule score is the product of the score of each rule in the preset combination rule and the corresponding directional weight; the directional weight is obtained based on the relationship between the corresponding rule and the microbial culture temperature; microorganisms are divided into multiple temperature ranges including low-temperature microorganisms, mesophilic microorganisms, and high-temperature microorganisms; low-temperature microorganisms are microorganisms with a temperature range of less than 20°C, mesophilic microorganisms are microorganisms with a temperature range of 20–50°C, and high-temperature microorganisms are microorganisms with a temperature range of greater than 50°C; based on the biological feature matrix, a machine learning model is trained for each of the microorganisms in the temperature range; the protein sequence of the microorganism to be predicted is obtained, domain information is extracted based on the protein sequence, a biological feature matrix is ​​generated based on the domain information, and the biological feature matrix is ​​input into the machine learning model to predict the culture temperature of the microorganism; The preset combination rules include a full hit rule, an arbitrary hit rule, and a three-domain hit rule. In the full hit rule, when all domains within the rule are hit, the score is the sum of the matching confidence scores of all domains; otherwise, the score is zero. The full hit rule includes the GroEL and GroES dual-subunit chaperone system. In the arbitrary hit rule, when at least one domain within the rule is hit, the score is the highest matching confidence score among the hit domains; otherwise, the score is zero. The arbitrary hit rule includes the HtpG and RNase R single-domain marker genes. In the three-domain hit rule, when all three domains within the rule are hit, the score is the sum of the matching confidence scores of the three domains; when two of the three domains within the rule are hit, the score is the product of the sum of the matching confidence scores of the two hit domains and a preset folding ratio; otherwise, the score is zero. The three-domain hit rule includes the ternary folding chaperone system of dnaK, dnaJ, and grpE.

2. The method for predicting microbial culture temperature based on domain features according to claim 1, characterized in that, The domain information includes the matching confidence and copy number of each relevant domain in the protein sequence.

3. The method for predicting microbial culture temperature based on domain features according to claim 1, characterized in that, The preset combination rules also include at least one of the following: minimum score rule, copy number rule, complementary hit rule, pair hit rule, and binary hit rule; In the minimum score rule, when both structural domains within the rule are hit, the score is the smaller of the matching confidence of the two structural domains; otherwise, the score is zero. In the copy number rule, when one structure domain within the rule is hit, the score is the matching confidence of the hit structure domain; when two structure domains within the rule are hit, the score is the product of the matching confidence of the two structure domains and the preset multiplier; otherwise, the score is zero. In the complementary hit rule, when both structural domains within the rule are hit, the score is the sum of the matching confidence of the two structural domains; when one structural domain within the rule is hit, the score is the product of the matching confidence of the hit structural domain and the preset multiplier; otherwise, the score is zero. In the pairwise hit rule, when both structural domains within the rule are hit, the score is the sum of the matching confidence of the two structural domains; otherwise, the score is zero. In the binary hit rule, when both the structural domains in the first component and the structural domains in the second component are hit, the score is the larger of the matching confidence scores of the two structural domains; otherwise, the score is zero.

4. The method for predicting microbial culture temperature based on domain features according to claim 1, characterized in that, When performing relevant domain alignment on the protein sequence, the matching results with expected values ​​less than a preset threshold are filtered out.

5. The method for predicting microbial culture temperature based on domain features according to claim 1, characterized in that, Training a machine learning model involves the following steps: Using the XGBoost classification model with biological feature matrix as independent variable, microorganisms are classified into microorganisms in multiple temperature ranges, including low-temperature microorganisms, mesophilic microorganisms, and high-temperature microorganisms. The classification model is trained and outputs the probability distribution of each category. For each temperature range, an independent XGBoost regression model is trained using samples of the corresponding category as training data to obtain a regressor for each temperature range.

6. The method for predicting microbial culture temperature based on domain features according to claim 5, characterized in that, The predictions made by the machine learning model include the following methodological steps: The input biological feature matrix is ​​predicted by a classification model to obtain the probability that the input biological feature matrix belongs to each temperature range. The biological feature matrix is ​​input into the regressor for each temperature range, and the corresponding predicted temperature value is obtained from the output. Based on the probability of each temperature range and the corresponding predicted temperature value, the final predicted culture temperature is obtained through weighted calculation.

7. The method for predicting microbial culture temperature based on domain features according to claim 5, characterized in that, In the training of the machine learning model, the loss function of the classification model is multi-class log loss, and the evaluation method of the regressor includes mean absolute error and root mean square error.

8. A microbial culture temperature prediction system based on domain features, used to predict microbial culture temperature using the microbial culture temperature prediction method based on domain features according to any one of claims 1-7, characterized in that, include: The rule score calculation module is used to calculate rule scores based on protein sequences using preset combination rules, and the rule scores are used to form a biological feature matrix. A machine learning model is used as input to a biological feature matrix and outputs a predicted value for the final culture temperature.

9. A microbial culture temperature prediction system based on domain features according to claim 8, characterized in that, The machine learning model includes: A classification model is used to predict the probability that the input biological feature matrix belongs to each temperature range. A regressor is used to take the biological feature matrix as input and output the corresponding predicted temperature value. The weighted output unit is used to obtain the final predicted culture temperature value through weighted calculation based on the probability of each temperature range and the corresponding predicted temperature value.

Images