Multi-element machine learning model-based cross-species lung disease feature gene screening method and system, electronic system and storage device

Abstract

The invention provides a multi-element machine learning model-based cross-species lung disease characteristic gene screening method and system, an electronic system and a storage device. The method comprises the following steps of: acquiring single cell / transcriptome data related to mouse lung diseases from a public database and preprocessing the single cell / transcriptome data; training the model by adopting six machine learning algorithms and outputting a gene importance score; calculating the weight according to the model performance and normalizing the score; and integrating the cross-species scores through a weighted fusion formula, and outputting a feature gene list and a visual report. The system comprises a data acquisition and preprocessing module, a multi-element machine learning model training module, a weight calculation and normalization module, a cross-species comprehensive scoring module and a result output module. The screening accuracy, stability and generalization ability are improved through multi-algorithm integration and cross-species fusion, and the method can be widely applied to the fields of mechanism research of lung diseases, diagnosis marker development and drug target verification.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of bioinformatics and machine learning, and particularly provides a cross-species lung disease characteristic gene screening method, system, electronic system and storage device based on a multi-element machine learning model, which is used for accurately identifying characteristic genes related to lung diseases in humans and mice, and supporting disease mechanism research, development of diagnostic markers and verification of drug targets. BACKGROUND

[0002] Lung diseases, including silicosis, pulmonary fibrosis and chronic obstructive pulmonary disease, are major global health threats, and their pathological mechanisms involve multiple signaling pathways and cell type abnormalities. Accurate screening of characteristic genes closely related to disease occurrence and development is of great value for in-depth understanding of disease mechanisms, development of early diagnostic markers and discovery of treatment targets. However, the existing technology has two major defects: First, the stability of a single algorithm is insufficient. Traditional differential expression gene analysis or a single machine learning model is susceptible to data noise, sample size limitations and algorithm characteristics, resulting in poor robustness and weak generalization ability of the screening results, and inconsistent conclusions from different data sets, which seriously hinders clinical translation.

[0003] Second, the cross-species differences are not effectively integrated. Existing methods do not fully consider the physiological and genetic differences between humans and model animals (such as mice), lack a systematic cross-species data fusion framework, and result in a lack of universality of the screened genes and a reduction in translation efficiency.

[0004] Specifically, traditional differential expression analysis methods rely on statistical differences, ignore complex interaction networks between genes, and are sensitive to batch effects, which can easily produce false positive or false negative results. Single machine learning algorithms have obvious preference, and different algorithms are sensitive to linear / non-linear relationships, which can easily overfit in limited samples and have poor interpretability of complex models. In addition, existing technologies lack an effective cross-species information integration framework, making it difficult to use model animal data to improve the reliability of human disease research. Therefore, there is an urgent need to develop a high-precision gene screening scheme that integrates multiple algorithms and fuses cross-species data. SUMMARY

[0005] The present application provides a cross-species lung disease characteristic gene screening method, system, electronic system and storage device based on a multi-element machine learning model, which aims to provide a cross-species lung disease characteristic gene screening scheme, to solve the problems of poor stability of a single model and neglect of species differences by integrating multiple machine learning algorithms and fusing human and mouse data, to significantly improve the accuracy, generalization ability and interpretability of gene screening, and to provide a reliable tool for disease mechanism research and precision medicine.

[0006] To solve the above technical problems, the technical scheme provided by the present application is: The application provides a cross-species lung disease characteristic gene screening method based on a multi-element machine learning model, comprising the following steps: S1: data acquisition and preprocessing - batch acquisition of human and mouse lung disease single cell or transcriptome data from public databases, quality control, normalization, clustering and cell type annotation, and generation of standardized gene expression matrix; S2: multi-element machine learning model training - LASSO, random forest, decision tree, support vector machine, gradient boosting machine and XGBoost six algorithms are used to train the model, and the gene importance score of each algorithm is output; S3: weight calculation and normalization - based on the model performance of human and mouse data sets, the weight is calculated, and the Min-Max normalization processing is performed on the score of each algorithm; S4: cross-species comprehensive score and output - the human and mouse scores are integrated by a weighted fusion formula to obtain the comprehensive score of the target gene and output the characteristic gene list.

[0007] Further, the S1 comprises: S11: database retrieval strategy formulation - formulating a retrieval formula covering silicosis, pulmonary fibrosis and chronic obstructive pulmonary disease, and the keywords including "lung disease", "pulmonary fibrosis", "COPD", "silicosis", "single cell RNA sequencing", "transcriptome", and ensuring that the target disease data is a subset of the retrieval results; S12: batch acquisition of cross-species data - batch acquisition of human (Homo sapiens) and mouse (Mus musculus) lung disease single cell transcriptome data from GEO and EMBL-EBI databases, and screening and downloading qualified data sets according to data quality, sample size and experimental design; S13: data quality control and merging, comprising: Quality evaluation: checking gene detection rate, cell survival rate and mitochondrial gene expression ratio; Eliminate low-quality samples and genes; Merging multiple data sets by inner join method; S14: batch effect removal and standardization, comprising: Dynamically select Harmony, SCTransform or scVI to remove batch effect; Standardizing the processed data to generate a standardized gene expression matrix; S15: cell clustering and type annotation, comprising: Unsupervised clustering analysis is performed by Leiden algorithm; By combining manual annotation with large language model-assisted annotation, the following two-level system is established: Level 1: Classified into major cell types, including epithelial cells, immune cells, and stromal cells; Level 2: Preserves cell subset information, including alveolar epithelial cell type I and II, and T cell subsets.

[0008] Furthermore, step S2 includes the following steps: S21: Construction of LASSO regression model: For binary classification problems: use a logistic regression model with L1 penalty, set solver='saga' and regularization strength C=1; For multi-class classification problems: the OneVsRestClassifier framework is used in combination with L1 regularized logistic regression, with C=0.1 and a maximum number of iterations of 10000. Gene importance scores are output based on the absolute value of the coefficients. S22: Random Forest Model Training: Construct a classifier containing 150 decision trees; Set the minimum number of split samples to 10 and the minimum number of leaf node samples to 10. Calculate the importance of features based on Gini impurity; S23: Support Vector Machine Model Implementation: A linear kernel function is used, with a regularization parameter C=0.1; Stratified sampling is used when the sample size is >5000. Feature importance is calculated using the absolute values ​​of linear SVM coefficients; S24: XGBoost gradient boosting model configuration: Parameter settings: Maximum depth = 5, Learning rate = 0.1, Number of estimators = 100; The objective function for binary classification is 'binary:logistic', and for multi-class classification it is 'multi:softprob'. Automatically switch to random forest when training fails; S25: Gradient Boosting Machine Model Training: We prioritize using LightGBM, with parameters of 100 estimators, a learning rate of 0.1, and a maximum depth of 3. Stratified sampling is enabled when the sample size is greater than 10,000. Based on the importance of gain calculation characteristics; S26: Decision Tree Model Construction: Set the maximum depth to 10, the minimum number of split samples to 5, and the minimum number of leaf node samples to 5. Calculate feature importance based on the reduction of Gini impurity.

[0009] Further, in S3: S31: Model performance evaluation index calculation Calculate two performance indicators for six machine learning algorithms on human and mouse datasets respectively: (a) Absolute residual error: obtained by calculating the absolute value of the difference between the model prediction probability and the true label one-hot encoding, reflecting the prediction accuracy of the model; (b) Area under ROC curve (AUC): calculated by the area under the receiver operating characteristic curve, where: (a1) Calculate the area under the ROC curve (AUC) directly for binary classification problems; (b1) Calculate the overall area under the ROC curve (AUC) using the macro-average strategy for multi-classification problems; S32: Weight distribution strategy design Design a weight distribution scheme based on the performance indicators in S31: (A) ROC curve area (AUC) weight distribution: proportional distribution, the algorithm with higher ROC curve area (AUC) value obtains larger weight; (B) Residual weight distribution: inverse distribution, the algorithm with smaller average absolute residual obtains larger weight; (C) Comprehensive weight calculation: determine the final weight by the arithmetic mean of the ROC curve area (AUC) weight and the residual weight; S33: Min-Max normalization implementation Standardize the gene importance scores output by each algorithm: The normalization formula is:

[0010] Where, : score of gene i for algorithm j; : normalized score of gene i for algorithm j; : minimum score in algorithm j; : maximum score in algorithm j; Standardize the score range to the [0, 1] interval to eliminate the dimensional differences between algorithms; S34: Cross-species weight consistency verification Calculate the weights of each algorithm on human and mouse datasets respectively; Verify the consistency of cross-species weight distribution through correlation analysis; Establish a weight correction mechanism: when the performance difference between two species exceeds the threshold value, adopt a conservative strategy to adjust the weight to avoid the influence of single species data deviation on the result.

[0011] Further, the S4 cross-species comprehensive scoring formula is:

[0012] wherein, : the performance weight of algorithm j on the mouse dataset; : the performance weight of algorithm j on the human dataset; : the number of algorithms selected for gene i in the mouse dataset; : the number of algorithms selected for gene i in the human dataset; : the normalized score of gene i of algorithm j; : the comprehensive score of cross-species gene i.

[0013] On the other hand, the present application also claims to protect a cross-species lung disease feature gene screening system based on a multi-element machine learning model, comprising sequentially connected: a data retrieval module for batch obtaining human and mouse lung disease related single cell or transcriptome data from public databases; a data preprocessing module for quality control, normalization, clustering and cell type annotation of the original data, generating a standardized gene expression matrix; a candidate gene scoring module, which adopts LASSO, random forest, decision tree, support vector machine, gradient boosting machine and XGBoost six kinds of machine learning algorithms to train the model and output gene importance score; a weight calculation module for calculating the weight according to the model performance of human and mouse data sets, and Min-Max normalization of each algorithm score; a comprehensive scoring module for integrating human and mouse data through a weighted fusion formula to generate the final comprehensive score of the target gene and output visual evaluation charts; a result output module for outputting a high reliability lung disease feature gene list and a visual report according to the comprehensive score threshold.

[0014] Further, the data preprocessing module specifically includes: a data quality control unit for performing the following operations: The gene detection rate is greater than 200 and less than 5000, the mitochondrial gene expression ratio is less than 20%, and the genes expressed in at least 3 cells are retained; The batch effect removal unit removes systematic bias between technical replicates and ensures comparability between datasets using an adaptive batch effect correction strategy, dynamically selecting the Harmony, SCTransform or scVI algorithm according to the characteristics of the dataset and the degree of batch effect; The cell clustering unit implements unsupervised cell population identification based on the Leiden graph clustering algorithm, and obtains the cell clustering result at the optimal resolution by constructing a k-nearest neighbor graph and optimizing the modularity function. The cell type annotation unit establishes a double-level cell type annotation system: Level 1 - Coarse-grained annotation: Integrate mainstream annotation database resources such as scType and scCATCH, and combine cell type marker genes in published literature to achieve annotation of main cell types; Level 2 - Fine-grained annotation: Based on the Level 1 annotation results, further annotate cell subpopulations and functional states according to research objectives and disease-specific needs to form a high-resolution cell type atlas.

[0015] Further, in the candidate gene scoring module: The LASSO regression model uses L1 penalty term logistic regression, and sets solver='saga' and C=1 for binary classification, and uses OneVsRestClassifier framework and C=0.1 for multi-classification; The random forest model contains 150 decision trees, the minimum split sample size is 10, and the minimum leaf node sample size is 10; The support vector machine uses a linear kernel function, the regularization parameter C=0.1, and the maximum sample size limit is 5000; XGBoost sets the maximum depth to 5, the learning rate to 0.1, and the number of estimators to 100; Gradient boosting machine preferentially uses LightGBM, with parameters of estimator number 100, learning rate 0.1, and maximum depth 3; The decision tree sets the maximum depth to 10 and the minimum split sample size to 5.

[0016] In another aspect, the present application also claims an electronic system for implementing cross-species lung disease characteristic gene screening, comprising: A processor; A memory storing a computer program; When the processor executes the computer program, the steps of the method according to any one of the preceding embodiments are implemented.

[0017] In yet another aspect, the present application also claims to protect a storage device, which stores a computer program, When the computer program is executed by a processor, the steps of the method according to any one of the preceding aspects are implemented.

[0018] Compared with the prior art, the cross-species lung disease feature gene screening method, system, electronic system and storage device based on a multi-element machine learning model have the following beneficial technical effects: The present application significantly overcomes the single model bias by integrating six complementary machine learning algorithms, greatly improves the stability and anti-interference ability of feature gene screening, innovatively fuses human and mouse cross-species data, breaks through the limitation of traditional methods ignoring biological conservation, and ensures that the screening results have strong generalization and transformation value; At the same time, a visual evaluation system is constructed, which provides an efficient and reliable computing platform for lung disease mechanism analysis, diagnostic marker development and drug target discovery. BRIEF DESCRIPTION OF DRAWINGS

[0019] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the drawings needed to be used in the embodiments or prior art description will be briefly introduced below. Obviously, the drawings in the following description are some embodiments of the present application, and those skilled in the art can obtain other drawings according to these drawings without creative labor.

[0020] Figure 1 A flowchart of a cross-species lung disease feature gene screening method based on a multi-element machine learning model is provided for the embodiments of the present application.

[0021] Figure 2 A schematic diagram of a cross-species lung disease feature gene screening system based on a multi-element machine learning model is provided for the embodiments of the present application.

[0022] Figure 3 A structural block diagram of an electronic system is provided for the embodiments of the present application. DETAILED DESCRIPTION

[0023] In order to make the purpose, technical scheme and advantages of the embodiments of the present application more clear, the technical scheme in the embodiments of the present application will be described clearly and completely below in combination with the drawings in the embodiments of the present application. Obviously, the described embodiments are some embodiments of the present application, not all embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor are within the scope of protection of the present application.

[0024] As Figure 1As shown, the cross-species lung disease feature gene screening method based on a multi-element machine learning model of the present application comprises the following steps: S1: data acquisition and preprocessing - batch acquisition of human and mouse lung disease single cell or transcriptome data from public databases, quality control, normalization, clustering and cell type annotation, and generation of standardized gene expression matrix; In the embodiments of the present application, human and mouse lung disease related single cell / transcriptome data are batch acquired from public databases such as GEO, EMBL-EBI, etc. The original data is subjected to quality control, normalization, clustering and cell type annotation to form a standardized gene expression matrix.

[0025] S2: multi-element machine learning model training - six algorithms including LASSO, random forest, decision tree, support vector machine, gradient boosting machine and XGBoost are used to train the model, and the gene importance scores of each algorithm are outputted; S3: weight calculation and normalization - based on the model performance of human and mouse data sets, the weight is calculated, and the scores of each algorithm are subjected to Min-Max normalization processing; S4: cross-species comprehensive scoring and output - the human and mouse scores are integrated by a weighted fusion formula to obtain the comprehensive score of the target gene and output the feature gene list; In the embodiments of the present application, the human and mouse six model scores are integrated across species by a set weighted fusion formula to obtain the final comprehensive score of the target gene, and a high-confidence lung disease feature gene list and a visual report are outputted according to the comprehensive score threshold.

[0026] Further, in the embodiments of the present application, the specific steps of data acquisition and preprocessing S1 include: S11: database retrieval strategy formulation. According to the lung disease research target, a retrieval formula covering diseases such as silicosis, pulmonary fibrosis, chronic obstructive pulmonary disease, etc. is formulated, and the retrieval keywords include but are not limited to "lung disease", "pulmonary fibrosis", "COPD", "silicosis", "single cell RNA sequencing", "transcriptome", etc. The retrieval formula is ensured to be sufficiently extensive, and the target disease data is a subset of the retrieval results; S12: Batch acquisition of cross-species data. Batch acquire single-cell transcriptome data related to lung diseases of Homo sapiens and Mus musculus from public databases such as GEO (Gene Expression Omnibus), EMBL-EBI (European Molecular Biology Laboratory-European Bioinformatics Institute), etc. Filter the search results according to data quality, sample size, experimental design, etc. standards, and batch download single-cell transcriptome data sets that meet the research theme and are of qualified quality; S13: Data quality control and merging. Quality assessment of the downloaded raw data, including checking of gene detection rate, cell survival rate, mitochondrial gene expression ratio, etc. indicators, and eliminating low-quality samples and genes; using inner join method to merge multiple data sets, to ensure the consistency and integrity of the merged data; S14: Batch effect removal and standardization. According to the characteristics of the data and the severity of the batch effect, dynamically select and apply Harmony, SCTransform, scVI (single-cell Variational Inference) batch effect removal methods to eliminate technical differences caused by different experimental batches, sequencing platforms and experimental conditions; standardize the data after removing the batch effect to generate a standardized gene expression matrix; S15: Cell clustering and type annotation. Use Leiden algorithm for unsupervised clustering analysis of preprocessed single-cell data to identify different cell populations; combine manual annotation and large language model assisted annotation to identify cell types, establish a two-level annotation system: Level 1 annotation merges cell subpopulations into main cell types such as epithelial cells, immune cells, interstitial cells, etc.; Level 2 annotation retains cell subpopulation information such as alveolar epithelial cells type I, type II, T cell subpopulations, etc.; according to the accuracy requirements of subsequent analysis, select the corresponding annotation level for analysis.

[0027] Further, in the embodiments of the present application, the specific steps of the multi-element machine learning model training S2 include: S21: LASSO regression model construction. According to the number of categories of the classification problem, select the corresponding model architecture: for binary classification problems, use a logistic regression model with an L1 penalty term, set the solver to'saga' solver, and the regularization strength C = 1; for multi-classification problems, use the OneVsRestClassifier framework combined with L1 regularization logistic regression, the regularization strength C = 0.1, and the maximum number of iterations is set to 10000 times to ensure model convergence; automatically perform feature selection through the sparsity characteristics of LASSO regression, and extract the absolute value of the coefficient as the gene importance score; S22: Random forest model training. Construct a random forest classifier containing 150 decision trees, set the minimum split sample size to 10 and the minimum leaf node sample size to 10, and use parallel computing to speed up the training process; evaluate the model performance through Out-of-Bag (OOB) error rate, and draw a curve of OOB error rate changing with the number of trees; calculate the feature importance score of each gene based on Gini impurity; S23: Support vector machine model implementation. To improve computational efficiency, stratified sampling is performed on large-scale data sets with a maximum sample size limit of 5000; use the SVM model with a linear kernel function, the regularization parameter C = 0.1, and enable class weight balancing and probability estimation functions; use the OneVsRestClassifier framework to handle multi-classification problems, and calculate feature importance through the absolute value of the linear SVM coefficient; S24: XGBoost gradient boosting model configuration. Set the model parameters: maximum depth 5, learning rate 0.1, estimator number 100, sub-sample ratio 0.8, and feature subsampling ratio 0.8; automatically select the objective function according to the number of classification categories: 'binary:logistic' for binary classification and'multi:softprob' for multi-classification; when XGBoost training fails, automatically switch to random forest as an alternative model to ensure the stability of the analysis process; S25: Gradient boosting machine model training. Preferentially try to use LightGBM implementation, parameter settings include 100 estimators, learning rate 0.1, and maximum depth 3; if LightGBM is not available, use sklearn's GradientBoostingClassifier; for large-scale data sets (more than 10000 samples), use stratified sampling strategy to reduce computational complexity; extract feature importance scores based on gain; S26: Decision tree model construction. Set the maximum depth of the decision tree to 10, the minimum split sample size to 5, and the minimum leaf node sample size to 5 to prevent model overfitting; use 5-fold cross-validation to evaluate the model's generalization performance; calculate the feature importance score based on the Gini impurity reduction, and generate a strong gene importance ranking with high interpretability.

[0028] Further, in the embodiments of the present application, the weight calculation and normalization S3 comprises the following steps: S31: Model performance evaluation index calculation. Calculate the absolute inverse residual and the area under the ROC curve (AUC) as performance evaluation indexes for six machine learning algorithms on human and mouse data sets respectively; the absolute inverse residual is obtained by calculating the absolute value of the difference between the model prediction probability and the true label one-hot encoding, reflecting the accuracy of the model prediction; the ROC-AUC value is calculated by the area under the receiver operating characteristic curve, and for binary classification problem, directly calculate AUC, for multi-classification problem, use macro-average strategy to calculate the overall AUC; S32: Weight distribution strategy design. Based on the model performance index, design the weight distribution scheme: AUC weight is distributed in proportion, the algorithm with higher AUC value obtains larger weight; residual weight adopts inverse distribution, the algorithm with smaller average absolute residual obtains larger weight; the comprehensive weight is calculated by the arithmetic mean of AUC weight and residual weight, to ensure the balance of two evaluation dimensions; S33: Min-Max normalization implementation. Perform Min-Max standardization processing on the gene importance score output by each algorithm, and unify the score range to the [0, 1] interval; the normalization formula is:

[0029] : the score of gene i of algorithm j; : the score of gene i of algorithm j after normalization; : the minimum score in algorithm j; : the maximum score in algorithm j; The difference in dimension and scale between different algorithms is eliminated through normalization processing, ensuring the effectiveness of subsequent weight fusion; S34: Cross-species weight consistency verification. Calculate the weight of each algorithm on the human and mouse data sets respectively, and verify the consistency of cross-species weight distribution through correlation analysis; establish a weight correction mechanism, when a certain algorithm shows too large difference between the two species, use the conservative strategy to adjust the weight, to avoid the deviation of single species data affecting the final result.

[0030] Further, in the embodiments of the present application, the cross-species comprehensive score and output S4 comprises: S41: Homologous gene mapping is established. Based on the NCBI HomoloGene database and the Ensembl Compara database, a human-mouse gene homologous relationship mapping table is established, and a one-to-one homologous gene matching strategy is adopted to ensure the biological significance of cross-species comparison. For the case where multiple homologous genes exist, the gene pair with the highest sequence similarity and consistent functional annotation is preferentially selected; S42: Cross-species comprehensive score calculation. The weighted fusion strategy is adopted to integrate the gene importance scores of human and mouse data, and the comprehensive score calculation formula is:

[0031] : the performance weight of algorithm j on the mouse dataset; : the performance weight of algorithm j on the human dataset; : the number of algorithms selected for gene i in the mouse dataset; : the number of algorithms selected for gene i in the human dataset; : the score of gene i of algorithm j after normalization; : the comprehensive score of cross-species gene i.

[0032] S43: Score effectiveness verification. The biological relevance of high-score genes is verified through gene function enrichment analysis; the correlation coefficient of human and mouse scores is calculated to evaluate the cross-species consistency; a score threshold dynamic adjustment mechanism is established to adaptively determine the feature gene screening threshold according to the disease type and data quality.

[0033] The present scheme significantly improves the accuracy and stability of feature gene screening through multi-algorithm integration; the generalization ability of the model is enhanced through cross-species information fusion; a complete visual evaluation system is provided to enhance the interpretability of the model; and an efficient and reliable computing tool is provided for precise diagnosis and treatment of lung diseases and new drug research and development.

[0034] According to another aspect of the embodiments of the present application, a multi-element machine learning model-based cross-species lung disease feature gene screening system for implementing the above-mentioned multi-element machine learning model-based cross-species lung disease feature gene screening method is also provided. Figure 2 is a schematic diagram of an optional multi-element machine learning model-based cross-species lung disease feature gene screening system according to the embodiments of the present application, as shown in Figure 2 The system can include: a data retrieval module for batch obtaining human and mouse lung disease related single cell or transcriptome data from public databases; a data preprocessing module for quality control, normalization, clustering and cell type annotation of raw data, and generating a standardized gene expression matrix; a candidate gene scoring module for training models using LASSO, random forest, decision tree, support vector machine, gradient boosting machine and XGBoost machine learning algorithms and outputting gene importance scores; a weight calculation module for calculating weights according to the model performance of human and mouse data sets, and Min-Max normalization of algorithm scores; a comprehensive scoring module for integrating human and mouse data through a weighted fusion formula to generate the final comprehensive score of the target gene, and outputting a visual evaluation chart; a result output module for outputting a high-confidence lung disease feature gene list and a visual report according to the comprehensive score threshold.

[0035] According to another aspect of the embodiments of the present application, a server or terminal for screening lung disease feature genes across species based on a multi-element machine learning model is also provided.

[0036] Figure 3 is a structural block diagram of a terminal according to an embodiment of the present application, as shown in Figure 3 the terminal can include one or more (only one is shown in the figure) processors 401, a memory 403, and a transmission device 405, as shown in Figure 3 the terminal can further include an input / output device 407.

[0037] The memory 403 can be used to store software programs and modules, such as the program instructions / modules corresponding to the method and system for screening lung disease feature genes across species based on a multi-element machine learning model according to the embodiments of the present application. The processor 401 executes various functional applications and data processing by running the software programs and modules stored in the memory 403, i.e. implements the method for screening lung disease feature genes across species based on a multi-element machine learning model. The memory 403 can include a high-speed random access memory, and can also include a non-volatile memory, such as one or more magnetic storage devices, flash memories, or other non-volatile solid-state memories. In some examples, the memory 403 can further include a memory remotely arranged with respect to the processor 401, which can be connected to the terminal through a network. Examples of the above-mentioned network include but are not limited to the Internet, an intranet, a local area network, a mobile communication network, and a combination thereof.

[0038] The aforementioned transmission device 405 is used to receive or send data via a network, and can also be used for data transfer between the processor and memory. Specific examples of the network described above may include wired networks and wireless networks. In one example, the transmission device 405 includes a Network Interface Controller (NIC), which can be connected to other network devices and a router via a network cable to communicate with the Internet or a local area network. In another example, the transmission device 405 is a radio frequency (RF) module used for wireless communication with the Internet.

[0039] Specifically, memory 403 is used to store application programs.

[0040] The processor 401 can invoke the application program stored in the memory 403 via the transmission device 405 to perform the following steps: (a) Obtain single-cell data of human / mouse lung diseases from public databases and generate a standardized gene expression matrix; (b) Gene importance scores are output using six machine learning algorithms; (c) Calculate the cross-species comprehensive score based on model performance weights; (d) Output a list of characteristic genes and a visualization report.

[0041] Optionally, specific examples in this embodiment can refer to the examples described in the above embodiments, and will not be repeated here.

[0042] Those skilled in the art will understand that Figure 3 The structure shown is for illustrative purposes only. The terminal can be a smartphone (such as an Android phone, an iOS phone, etc.), a tablet computer, a PDA, a mobile internet device (MID), a PAD, or other terminal devices. Figure 3 This does not limit the structure of the aforementioned electronic device. For example, the terminal may also include components that are more... Figure 3 The more or fewer components shown (such as network interfaces, display devices, etc.), or having the same Figure 3 The different configurations shown.

[0043] Those skilled in the art will understand that all or part of the steps in the various methods of the above embodiments can be implemented by a program instructing the hardware related to the terminal device. The program can be stored in a computer-readable storage medium, which may include: flash drive, read-only memory (ROM), random access memory (RAM), disk or optical disk, etc.

[0044] The embodiment of the present application further provides a storage medium. Optionally, in the embodiment, the storage medium can be used to execute program codes of the method for screening characteristic genes of lung diseases among different species based on a multi-element machine learning model.

[0045] Optionally, in the embodiment, the storage medium can be located on at least one of the plurality of network devices in the network shown in the foregoing embodiment.

[0046] Optionally, in the embodiment, the storage medium can include, but is not limited to, a U disk, a read-only memory (ROM), a random access memory (RAM), a mobile hard disk, a magnetic disk or an optical disk, and various media capable of storing program codes.

[0047] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present application, rather than limit the same; even though the present application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that the technical solutions recorded in the foregoing embodiments can still be modified, or some technical features can be replaced equivalently; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the embodiments of the present application.

Claims

1. A method for screening characteristic genes of lung diseases across species based on a multi-element machine learning model, characterized in that, Comprising the following steps: S1: Data acquisition and preprocessing - batch acquisition of human and mouse lung disease single cell or transcriptome data from public databases, quality control, normalization, clustering and cell type annotation, generating standardized gene expression matrix; S2: Multivariate machine learning model training - LASSO, random forest, decision tree, support vector machine, gradient boosting machine and XGBoost six algorithms are used to train the model, and the gene importance score of each algorithm is output; S3: Weight calculation and normalization - based on the model performance of human and mouse data sets, the weight is calculated, and the Min-Max normalization processing is performed on the score of each algorithm; S4: Cross-species comprehensive scoring and output - integrate human and mouse scores through a weighted fusion formula to obtain the comprehensive score of the target gene and output the feature gene list.

2. The method of claim 1, wherein, The S1 comprises: S11: Database retrieval strategy - develop a search formula covering silicosis, pulmonary fibrosis, chronic obstructive pulmonary disease, and the keywords include "lung disease", "pulmonary fibrosis", "COPD", "silicosis", "single cell RNA sequencing", "transcriptome", and ensure that the target disease data is a subset of the search results; S12: Batch acquisition of cross-species data - batch acquisition of human (Homo sapiens) and mouse (Mus musculus) lung disease single cell transcriptome data from GEO and EMBL-EBI databases, and filter and download qualified data sets according to data quality, sample size, and experimental design; S13: Data quality control and merging, including: Quality assessment: check gene detection rate, cell survival rate, and mitochondrial gene expression ratio; Remove low-quality samples and genes; Use inner join to merge multiple data sets; S14: Batch effect removal and standardization, including: Dynamically select Harmony, SCTransform or scVI to remove batch effects; Standardize the processed data to generate a standardized gene expression matrix; S15: Cell clustering and type annotation, including: Use Leiden algorithm for unsupervised clustering analysis; Combine manual annotation and large language model assisted annotation to establish the following two-level system: Level 1: merged into main cell types, including epithelial cells, immune cells, and interstitial cells; Level 2: retain cell subpopulation information, including alveolar epithelial cells type I, type II, and T cell subgroups.

3. The method of claim 1, wherein, In S2, the following steps are included: S21: LASSO regression model construction: For binary classification problem: use L1 penalty term logistic regression model, set solver='saga', regularization strength C=1; For multi-classification problem: use OneVsRestClassifier framework combined with L1 regularization logistic regression, set C=0.1, maximum iteration number 10000; The gene importance score is output by the absolute value of the coefficient; S22: Random Forest Model Training: Build a classifier containing 150 decision trees; Set the minimum split sample size = 10, the minimum leaf node sample size = 10; Calculate feature importance based on Gini impurity; S23: Support Vector Machine Model Implementation: Use linear kernel function, regularization parameter C = 0.1; Enable stratified sampling when sample size > 5000; Calculate feature importance by the absolute value of the linear SVM coefficient; S24: XGBoost Gradient Boosting Model Configuration: Set parameters: maximum depth = 5, learning rate = 0.1, estimator number = 100; Binary classification objective function is 'binary:logistic', and multi-classification is'multi:softprob'; Switch to random forest automatically when training fails; S25: Gradient Boosting Machine Model Training: Prioritize using LightGBM implementation, parameters: estimator number 100, learning rate 0.1, maximum depth 3; Enable stratified sampling when sample size > 10000; Calculate feature importance based on gain; S26: Decision Tree Model Construction: Set maximum depth = 10, minimum split sample size = 5, minimum leaf node sample size = 5; Calculate feature importance based on Gini impurity reduction.

4. The method of claim 1, wherein, In S3: S31: Model Performance Evaluation Index Calculation Calculate two performance indicators for six machine learning algorithms on human and mouse data sets respectively: (a) Absolute inverse residual: obtained by calculating the absolute value of the difference between the model prediction probability and the true label one-hot encoding, reflecting the prediction accuracy of the model; (b) Area under ROC curve (AUC): calculated by the area under the receiver operating characteristic curve, where: (a1) Calculate the area under the ROC curve (AUC) directly for binary classification problems; (b1) Calculate the overall area under the ROC curve (AUC) using the macro-average strategy for multi-classification problems; S32: Weight Distribution Strategy Design Design a weight distribution scheme based on the performance indicators in S31: (A) ROC curve area (AUC) weight distribution: proportional distribution, the algorithm with higher ROC curve area (AUC) value gets larger weight; (B) Residual weight distribution: inverse distribution, the algorithm with smaller average absolute residual gets larger weight; (C) Comprehensive weight calculation: determine the final weight by the arithmetic mean of ROC curve area (AUC) weight and residual weight; S33: Min-Max Normalization Implementation Standardize the gene importance scores output by each algorithm: The normalization formula is: Where, : score of gene i of algorithm j; : score of gene i for algorithm j after normalization; : minimum score in algorithm j; : maximum score in algorithm j; Standardize the score range to the [0, 1] interval to eliminate the dimensional differences between algorithms; S34: Cross-species weight consistency verification Calculate the weight of each algorithm on human and mouse data sets respectively; Verify the consistency of cross-species weight distribution through correlation analysis; Establish a weight correction mechanism: when the performance difference between two species for an algorithm exceeds the threshold, use a conservative strategy to adjust the weight to avoid the influence of single-species data bias on the result.

5. The method of claim 1, wherein, The cross-species comprehensive scoring formula in S4 is: Where, : performance weight on mouse dataset in algorithm j; : performance weight on human dataset in algorithm j; : Number of algorithms selected for gene i in the mouse dataset; : Number of algorithms selected in human dataset for gene i; : score of gene i for algorithm j after normalization; : Integrated score of cross-species gene i. 6.A system for screening cross-species lung disease characteristic genes based on a multi-element machine learning model, characterized in that, It includes the following sequentially connected: a data retrieval module, configured to batch acquire single-cell or transcriptome data related to human and mouse lung diseases from public databases; a data preprocessing module, configured to perform quality control, normalization, clustering, and cell type annotation on the raw data, and generate a standardized gene expression matrix; a candidate gene scoring module, configured to train models by using six machine learning algorithms, including LASSO, random forest, decision tree, support vector machine, gradient boosting machine, and XGBoost, and output gene importance scores; a weight calculation module, configured to calculate weights according to the model performance of human and mouse data sets, and perform Min-Max normalization on the scores of the algorithms; a comprehensive scoring module, configured to integrate human and mouse data by using a weighted fusion formula, generate a final comprehensive score of a target gene, and output a visual evaluation chart; a result output module, configured to output a high-confidence lung disease feature gene list and a visual report according to a comprehensive score threshold.

7. The system of claim 6, wherein, The data preprocessing module specifically comprises: a data quality control unit, configured to perform the following operations: genes with a detection rate of >200 and <5,000 and mitochondrial gene expression ratio of <20% are reserved, and genes expressed in at least 3 cells are reserved; a batch effect removal unit, configured to dynamically select Harmony, SCTransform, or scVI algorithm according to the characteristics and batch effect degree of the data set by using an adaptive batch effect correction strategy, eliminate systematic bias between technical repeats, and ensure comparability between data sets; a cell clustering unit, configured to realize unsupervised cell population recognition based on a Leiden graph clustering algorithm, obtain cell clustering results at an optimal resolution by constructing a k-nearest neighbor graph and optimizing a modularity function; a cell type annotation unit, configured to establish a double-level cell type annotation system: Level 1: coarse-grained annotation: integrate mainstream annotation database resources such as scType and scCATCH, combine cell type marker genes in published literature, and realize annotation of main cell types; Level 2: fine-grained annotation: based on the annotation results of Level 1, further annotate cell subpopulations and functional states according to research targets and disease-specific requirements, and form a high-resolution cell type atlas.

8. The system of claim 6, wherein, In the candidate gene scoring module: the LASSO regression model adopts L1 penalty term logistic regression, sets solver='saga' and C=1 in binary classification, and adopts OneVsRestClassifier framework and C=0.1 in multi-classification; the random forest model contains 150 decision trees, the minimum split sample number is 10, and the minimum leaf node sample number is 10; the support vector machine adopts a linear kernel function, a regularization parameter C=0.1, and a maximum sample size limit of 5000; XGBoost is set to a maximum depth of 5, a learning rate of 0.1, and an estimator number of 100; the gradient boosting machine preferentially uses LightGBM, and the parameters are an estimator number of 100, a learning rate of 0.1, and a maximum depth of 3; the decision tree is set to a maximum depth of 10 and a minimum split sample number of 5.

9. An electronic system for implementing a cross-species lung disease signature gene screening, comprising, comprise: a processor; a memory storing a computer program; The steps of the method according to any one of claims 1-5 are implemented when the processor executes the computer program.

10. A storage device storing a computer program, characterized by The steps of the method according to any one of claims 1-5 are implemented when the computer program is executed by a processor.

Images