Plant cell type prediction method and device based on interpretable machine learning

Abstract

The invention provides a plant cell type prediction method and device based on interpretable machine learning, and the method comprises the steps: obtaining single cell data of a target organ of a target plant; preprocessing the single cell data to obtain standardized single cell data; performing feature screening on the standardized single cell data to obtain a hypervariable gene of the single cell data, the expression level of the hypervariable gene in a target cell being higher than a first expression threshold, and the expression level of the hypervariable gene in other cells being lower than the first expression threshold; performing normalization based on single cell data and hypervariant genes to obtain a cell gene matrix; determining a target multi-classification model corresponding to the target organ of the target plant; inputting the cell gene matrix into a target multi-classification model to obtain a predicted cell type output by the target multi-classification model; according to the invention, efficient and accurate prediction of cell types of various organs of various plants can be realized.

Description

Technical Field

[0001] The present invention relates to the field of artificial intelligence technology, and in particular to a method and device for predicting plant cell types based on interpretable machine learning. Background Art

[0002] Unlike higher animals, higher plant cells have high plasticity, and differentiated plants can acquire totipotency and regenerate complete fertile plants.

[0003] In single-cell data analysis, accurate classification of cells is the basis of data analysis. Specific marker genes of different cell populations are identified based on statistical methods. The final experimental method requires specific fluorescent labeling of the marker genes to express the cell type. The above method lacks accuracy and has great limitations.

[0004] It can be seen that the single-cell data analysis method for plants in the relevant technology has technical problems such as low accuracy and great limitations. Summary of the invention

[0005] The present invention provides a plant cell type prediction method and device based on interpretable machine learning, which is used to solve the technical problems of low accuracy and large limitations in the single-cell data analysis method for plants in the related art, and to achieve efficient and accurate prediction of cell types of multiple organs of multiple plants.

[0006] The present invention provides a method for predicting plant cell types based on interpretable machine learning, comprising the following steps.

[0007] Acquire single-cell data of a target organ of a target plant; preprocess the single-cell data to obtain standardized single-cell data; perform feature screening on the standardized single-cell data to obtain highly variable genes of the single-cell data, wherein the expression level of the highly variable genes in target cells is higher than a first expression threshold, and the expression level of the highly variable genes in other cells is lower than the first expression threshold; perform normalization based on the single-cell data and the highly variable genes to obtain a cell gene matrix; determine a target multi-classification model corresponding to the target organ of the target plant; input the cell gene matrix into the target multi-classification model to obtain a predicted cell type output by the target multi-classification model.

[0008] According to a plant cell type prediction method based on interpretable machine learning provided by the present invention, after inputting the cell gene matrix into the target multi-classification model to obtain the predicted cell type output by the target multi-classification model, the method also includes: performing SHAP value analysis on the target multi-classification model to obtain specific marker genes corresponding to the predicted cell type.

[0009] According to a plant cell type prediction method based on interpretable machine learning provided by the present invention, the single-cell data is preprocessed to obtain standardized single-cell data, including: determining the original single-cell RNA sequencing data of the single-cell data; performing quality control processing on the original single-cell RNA sequencing data to obtain filtered single-cell RNA sequencing data, wherein the quality control processing includes: removing low-expression genes and removing mitochondrial genes, wherein the expression level of the low-expression gene is lower than a second expression threshold; and standardizing the sequencing depth and gene expression amount of the filtered single-cell RNA sequencing data to obtain standardized single-cell data.

[0010] According to a plant cell type prediction method based on explainable machine learning provided by the present invention, before obtaining the single-cell data of the target organ of the target plant, the method also includes: obtaining single-cell data of different organs of multiple plant species; preprocessing and annotating the single-cell data to obtain a cell type label for each single-cell data; feature screening the single-cell data based on the cell type label to obtain a target number of highly variable genes for each single-cell data; normalizing each single-cell data and the highly variable genes of each single-cell data to obtain a cell gene matrix for each single cell; classifying the cell gene matrix of each single cell according to different organs of the multiple plant species to obtain a training data set for each organ of each plant species, wherein the training data set includes: the cell gene matrix and the cell type label corresponding to the cell gene matrix; based on the training data set of each organ of each plant species and the target machine learning algorithm, constructing a multi-classification model corresponding to each organ of each plant species.

[0011] According to a plant cell type prediction method based on explainable machine learning provided by the present invention, before constructing a multi-classification model corresponding to each organ of each plant species based on the training data set of each organ of each plant species and the target machine learning algorithm, the method also includes: obtaining a test data set for plant cell type prediction, wherein the test data set includes a test cell gene matrix and a cell type label corresponding to the test cell gene matrix; inputting the test cell gene matrix into a plurality of preset machine learning algorithms respectively to obtain the test predicted cell type and test time output by each of the plurality of machine learning algorithms; comparing the test predicted cell type output by each machine learning algorithm with the cell type label respectively to obtain the test accuracy of each machine learning algorithm; and determining the target machine learning algorithm based on the test time of each machine learning algorithm, the test accuracy of each machine learning algorithm and the model robustness of each machine learning algorithm.

[0012] According to a plant cell type prediction method based on explainable machine learning provided by the present invention, the target machine learning algorithm is the LightGBM algorithm.

[0013] The present invention also provides a plant cell type prediction device based on interpretable machine learning, comprising the following modules: an acquisition module, used to acquire single-cell data of a target organ of a target plant; a preprocessing module, used to preprocess the single-cell data to obtain standardized single-cell data; a screening module, used to perform feature screening on the standardized single-cell data to obtain highly variable genes of the single-cell data, wherein the expression level of the highly variable genes in the target cells is higher than a first expression threshold, and the expression level of the highly variable genes in other cells is lower than the first expression threshold; a normalization module, used to perform normalization based on the single-cell data and the highly variable genes to obtain a cell gene matrix; a determination module, used to determine a target multi-classification model corresponding to the target organ of the target plant; and a prediction module, used to input the cell gene matrix into the target multi-classification model to obtain a predicted cell type output by the target multi-classification model.

[0014] The present invention also provides an electronic device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the program, a plant cell type prediction method based on interpretable machine learning as described above is implemented.

[0015] The present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements any of the plant cell type prediction methods based on interpretable machine learning as described above.

[0016] The present invention also provides a computer program product, comprising a computer program, which, when executed by a processor, implements any of the above-described methods for predicting plant cell types based on interpretable machine learning.

[0017] The plant cell type prediction method and device based on interpretable machine learning provided by the present invention can eliminate noise and deviation in the data and improve the accuracy and reliability of the data by preprocessing single-cell data; through feature screening, highly variable genes with significant differences in expression levels in single-cell data can be identified, which is of great significance for distinguishing different cell types. At the same time, by setting a first expression threshold, genes that are highly expressed only in target cells can be further screened out, thereby improving the accuracy and pertinence of feature selection; based on normalization processing of single-cell data and highly variable genes, the expression amount differences between different samples can be eliminated, making the data more comparable and easier to analyze; by determining a target multi-classification model corresponding to the target organ of the target plant, the cell gene matrix can be input into the model in a targeted manner, so as to classify cell types more efficiently and accurately and identify the predicted cell types in the target organ of the target plant; thereby solving the technical problems of low accuracy and large limitations in the single-cell data analysis method for plants in the related art. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] In order to more clearly illustrate the technical solutions in the present invention or the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced one by one below. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0019] Figure 1 It is a schematic diagram of the process of the plant cell type prediction method based on interpretable machine learning provided by the present invention.

[0020] Figure 2 It is a schematic diagram of the overall process of the plant cell type prediction method based on interpretable machine learning provided by the present invention.

[0021] Figure 3 It is a schematic diagram for comparing the prediction effects of the model based on the LightGBM algorithm provided by the present invention.

[0022] Figure 4 It is a schematic diagram of the structure of a plant cell type prediction device based on interpretable machine learning provided by the present invention.

[0023] Figure 5 It is a schematic diagram of the physical structure of the electronic device provided by the present invention. DETAILED DESCRIPTION

[0024] In order to make the purpose, technical solution and advantages of the present invention clearer, the technical solution of the present invention will be clearly and completely described below in conjunction with the drawings of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0025] Single-cell sequencing technology is developing rapidly and has become a powerful method for studying gene expression in complex multicellular organisms. Compared with bulk sequencing, single-cell sequencing technology can identify rare cell populations and reveal cell state transitions at different developmental stages, which are difficult to capture using traditional methods. Therefore, solving scientific problems at the single-cell level is the focus of current research.

[0026] Unlike higher animals, higher plants have high plasticity in cell fate, and differentiated plants can acquire totipotency and regenerate complete fertile plants. Nowadays, many single-cell sequencing technologies have been applied to model organisms such as Arabidopsis, rice, and corn, and these studies are particularly important for plant research. In single-cell data analysis, accurate classification of cells is the basis of data analysis. Cells are classified using unsupervised clustering algorithms, and specific marker genes of different cell populations are identified based on statistical methods. The final experimental method requires the marker genes to be fluorescently labeled with specific expression cell types. These methods lack accuracy and have great limitations.

[0027] Therefore, starting from a variety of single-cell omics and machine learning methods, standardizing and uniformly processing the data, and modeling and analyzing the multi-omics data of different single-cell populations are of great significance for analyzing cell heterogeneity and identifying new cell subpopulations and marker genes.

[0028] The main technical problem to be solved by the present invention is to perform intelligent prediction on plant single cell data and to mine new plant cell type-specific marker genes.

[0029] In order to solve the above technical problems, the present invention integrates the existing plant single-cell data, re-integrates and analyzes the data, and builds a machine learning model for each tissue of each species to accurately predict cell types. In the model, cell type-specific marker genes are screened out according to the order of feature importance for subsequent experimental verification.

[0030] Optionally, the plant cell type prediction method based on interpretable machine learning in the embodiment of the present application can be executed by a server, or by a terminal device, or jointly by a server and a terminal device, taking the plant cell type prediction method based on interpretable machine learning in this embodiment executed by a server as an example.

[0031] Figure 1 is a schematic diagram of a process of predicting plant cell types based on interpretable machine learning provided by the present invention, such as Figure 1 As shown, the method includes the following: Step 101, obtaining single cell data of a target organ of a target plant.

[0032] It should be noted that the single-cell data is extracted from the target organ of the target plant. It is understandable that the target plant and the target organ are predetermined.

[0033] In an embodiment of the present invention, single-cell RNA sequencing (scRNA-seq) data from specific organs of a target plant are collected; single-cell sequencing technologies, such as 10x Genomics, SMART-seq2, etc., are used to sequence cells of the target organ to obtain single-cell data containing a large amount of cellular gene expression information.

[0034] Step 102: preprocess the single-cell data to obtain standardized single-cell data.

[0035] The main purpose of standardizing single-cell data is to eliminate technical bias, batch effects, and improve data consistency and comparability.

[0036] In an embodiment of the present invention, based on indicators such as sequencing depth and the number of genes detected, cells with lower quality are identified and removed; genes with extremely low expression levels in all cells are removed, which may be caused by sequencing noise.

[0037] Since the sequencing depth of different cells may be different, it is necessary to divide the gene expression of each cell by the sequencing depth of the cell (i.e., total reads or UMI number) to eliminate the impact of sequencing depth on expression; perform a logarithmic transformation (such as log2 or log10) on the normalized expression to reduce the impact of extreme values ​​and make the data closer to a normal distribution.

[0038] If the data come from different experimental batches, specific algorithms are used to correct for batch effects.

[0039] In some embodiments, the expression level of each gene in the single-cell data is converted into a Z score, that is, the average value of the gene in all cells is subtracted and then divided by the standard deviation to eliminate the expression level differences between genes.

[0040] Step 103, feature screening is performed on the standardized single-cell data to obtain highly variable genes in the single-cell data, wherein the expression level of the highly variable genes in the target cells is higher than the first expression threshold, and the expression level of the highly variable genes in other cells is lower than the first expression threshold.

[0041] Perform feature screening on normalized single-cell data to identify hypervariable genes (i.e., genes whose expression in a specific cell type is significantly different from that in other cell types).

[0042] Based on the overall distribution of the data or the expression characteristics of a specific cell type, a first expression threshold is set. The first expression threshold is used to distinguish between highly expressed and lowly expressed genes.

[0043] For each gene in the normalized single-cell data, its expression level in the target cell and other cells is compared, and genes whose expression level in the target cell is higher than the first expression threshold and whose expression level in other cells is lower than the first expression threshold are selected as highly variable genes.

[0044] Step 104, normalizing the single cell data and the highly variable genes to obtain a cell gene matrix.

[0045] In the embodiment of the present invention, the expression data of these highly variable genes in all cells are extracted to obtain the highly variable gene expression data; The highly variable gene expression data are normalized and organized into a cell gene matrix; wherein the rows of the cell gene matrix represent genes, the columns represent cells, and the values ​​in the matrix represent the expression levels of the genes in the cells.

[0046] Here, the purpose of normalization is to eliminate the differences in expression levels between different genes so that the expression data of different genes can be compared on the same scale.

[0047] For example, commonly used normalization methods include Z-score normalization, min-max normalization, or quantilenormalization.

[0048] Step 105, determining a target multi-classification model corresponding to a target organ of a target plant.

[0049] In order to perform cell type classification more efficiently and accurately, a target multi-classification model corresponding to a target organ of a target plant is determined from a plurality of preset multi-classification models for different plant organs of different plant species.

[0050] Step 106, inputting the cell gene matrix into the target multi-classification model to obtain the predicted cell type output by the target multi-classification model.

[0051] In an embodiment of the present invention, the cell gene matrix is ​​passed as input data to the target multi-classification model, for example, the prediction function of the multi-classification model is called, and the cell gene matrix is ​​used as a parameter of the function; the target multi-classification model uses the received cell gene matrix as input, performs internal calculations, including feature extraction, classification decision and other steps, and outputs prediction results, which include: a vector or list of predicted cell types corresponding to the single-cell data.

[0052] Through the embodiments of the present invention, single-cell data of a target organ of a target plant are obtained; the single-cell data are preprocessed to obtain standardized single-cell data; feature screening is performed on the standardized single-cell data to obtain highly variable genes of the single-cell data, wherein the expression level of the highly variable genes in the target cells is higher than a first expression threshold, and the expression level of the highly variable genes in other cells is lower than the first expression threshold; normalization is performed based on the single-cell data and the highly variable genes to obtain a cell gene matrix; a target multi-classification model corresponding to the target organ of the target plant is determined; the cell gene matrix is ​​input into the target multi-classification model to obtain a predicted cell type output by the target multi-classification model; the cell gene matrix can be input into the model in a targeted manner to perform cell type classification more efficiently and accurately, and identify the predicted cell type in the target organ of the target plant; thereby solving the technical problem that the single-cell data analysis method for plants in the related art has a low accuracy rate and large limitations.

[0053] According to a plant cell type prediction method based on interpretable machine learning provided by the present invention, after inputting the cell gene matrix into the target multi-classification model to obtain the predicted cell type output by the target multi-classification model, the method further includes: The SHAP value analysis was performed on the target multi-classification model to obtain specific marker genes corresponding to the predicted cell types.

[0054] In the embodiment of the present invention, based on the feature importance of the target multi-classification model and the analysis of SHAP importance, marker genes specific to each cell type are screened out.

[0055] In some embodiments, SHAP value calculation analysis is performed on each constructed model to identify genes with high importance in distinguishing cell types, and the contribution of each gene in the prediction of a specific cell type is quantified according to the SHAP value corresponding to each cell type. The final result is output in the form of a table.

[0056] SHAP (SHapley Additive exPlanations) value analysis is a method for explaining the prediction results of machine learning models, which is applicable to classification and regression tasks. Through SHAP value analysis, we can understand the contribution of each feature (gene in this case) to the model prediction results, thereby identifying specific marker genes corresponding to the predicted cell type.

[0057] For example, according to the type of target multi-classification model (such as tree-based model, linear model, etc.), select a suitable SHAP interpreter for initialization; use the initialized SHAP interpreter to calculate the SHAP value of the cell gene matrix. This will generate a SHAP value for each gene in each cell sample, indicating the contribution of the gene to the model prediction result.

[0058] By analyzing the size and positive and negative of the SHAP value, specific marker genes corresponding to the predicted cell type can be identified. Generally, genes with larger SHAP values ​​and clear positive and negative values ​​are more likely to be specific marker genes.

[0059] Through the embodiment of the present invention, the target multi-classification model is interpreted using SHAP value analysis, and specific marker genes corresponding to the predicted cell type are identified. The specific marker genes are used to understand the biological characteristics of the cell type.

[0060] According to a plant cell type prediction method based on interpretable machine learning provided by the present invention, single cell data is preprocessed to obtain standardized single cell data, including: Determine the raw single-cell RNA sequencing data of single-cell data; Performing quality control processing on the original single-cell RNA sequencing data to obtain filtered single-cell RNA sequencing data, wherein the quality control processing includes: removing low-expression genes and removing mitochondrial genes, wherein the expression level of the low-expression gene is lower than a second expression threshold; The sequencing depth and gene expression of the filtered single-cell RNA sequencing data were standardized to obtain standardized single-cell data.

[0061] The second expression threshold is pre-set and used to distinguish low-expression genes from high-expression genes. For example, the gene expression level is compared with the second expression threshold, genes with expression levels higher than the threshold are retained, and low-expression genes with expression levels lower than the threshold are removed.

[0062] In an embodiment of the present invention, data collection and preprocessing includes: collecting single cell data and marker gene data of multiple plant species.

[0063] The collected raw single-cell RNA sequencing data was quality controlled to remove low-quality cells, low-expression genes and mitochondrial genes, ensuring the reliability of the subsequent analysis data. For the filtered data, the data was standardized to eliminate technical errors, so that the gene expression of cells with different sequencing depths and library sizes can be comparable, thereby better presenting the differences at the gene level.

[0064] Screen for highly variable genes that are highly expressed in some cells but lowly expressed in other cells. These genes are usually the most important for distinguishing cell types and are therefore used in subsequent analysis. The highly variable genes are further used for data normalization analysis, and the resulting matrix is ​​used for subsequent modeling analysis.

[0065] At the same time, we integrate the single-cell data of different species at the organ level. After merging the single-cell data of different species from the same organ but different sources, we use Harmony to perform batch correction on the data with the data source as the variable.

[0066] Finally, the integrated data were re-annotated for cell types. Based on known marker genes and cell characteristics from the literature, appropriate cell type labels were assigned to each cell. This process may include manual review and manual correction to ensure the accuracy of the annotations.

[0067] According to a plant cell type prediction method based on interpretable machine learning provided by the present invention, before obtaining single cell data of a target organ of a target plant, the method further includes: Obtain single-cell data from different organs of multiple plant species; Preprocess and annotate single-cell data to obtain the cell type label of each single-cell data; Perform feature screening on single-cell data based on cell type labels to obtain the target number of highly variable genes for each single-cell data; Based on the normalization of each single cell data and the highly variable genes of each single cell data, the cell gene matrix of each single cell is obtained; Classifying the cell gene matrix of each single cell according to different organs of multiple plant species to obtain a training data set for each organ of each plant species, wherein the training data set includes: a cell gene matrix and a cell type label corresponding to the cell gene matrix; Based on the training data set of each organ of each plant species and the target machine learning algorithm, a multi-classification model corresponding to each organ of each plant species is constructed.

[0068] In an embodiment of the present invention, single-cell RNA sequencing data are collected from different organs of multiple plant species. These data are usually obtained by high-throughput sequencing technology and contain gene expression information for each cell.

[0069] The raw single-cell data are preprocessed, including removing noise, correcting batch effects, removing low-quality cells, etc., and using known gene markers or databases, the cell type of each single-cell data is annotated to obtain the cell type label of each cell. For example, this is achieved by comparing the gene expression profile with the characteristic gene expression profile of a known cell type.

[0070] Based on the cell type labels, feature screening is performed on the single-cell data of each organ of each plant species to identify highly variable genes in each organ (i.e., genes with significant expression differences between different cell types). This can be achieved, for example, through differential expression analysis, variance analysis, and other methods.

[0071] According to the analysis requirements, determine the number of highly variable genes that need to be retained for each single-cell data. The target number can be determined based on the characteristics of the data, the purpose of the analysis, and the limitations of computing resources.

[0072] Based on each single cell data and its highly variable genes, normalization is performed to eliminate the differences between different gene expression levels, making the data more suitable for subsequent machine learning algorithms. Normalization methods include but are not limited to z-score normalization, min-max normalization, etc.

[0073] The normalized data are organized into a cell gene matrix, where each row represents a gene, each column represents a cell, and the values ​​in the matrix represent the expression levels of the genes in the cells.

[0074] According to the different organs of multiple plant species, the cell gene matrix is ​​classified to obtain a training data set for each organ of each plant species. The data set includes the cell gene matrix and the corresponding cell type label.

[0075] Choose a suitable machine learning algorithm such as support vector machine (SVM), random forest, gradient boosted tree (GBDT), or deep learning algorithm.

[0076] Using the training data set of each organ of each plant species, a multi-classification model corresponding to each organ is trained so that the multi-classification model can predict the type of cell based on the cell gene matrix.

[0077] The performance of the model is evaluated through indicators such as cross-validation, accuracy, recall, and F1 score, and the model is optimized based on the evaluation results to improve the prediction accuracy of the model.

[0078] Through the embodiments of the present invention, a multi-classification model can be constructed for each organ of each plant species, thereby achieving accurate classification of single-cell data and accurate prediction of cell types.

[0079] According to a plant cell type prediction method based on explainable machine learning provided by the present invention, before constructing a multi-classification model corresponding to each organ of each plant species based on a training data set of each organ of each plant species and a target machine learning algorithm, the method further includes: Acquire a test data set for plant cell type prediction, wherein the test data set includes a test cell gene matrix and a cell type label corresponding to the test cell gene matrix; Inputting the test cell gene matrix into a plurality of preset machine learning algorithms respectively, and obtaining the test predicted cell type and test time output by each of the plurality of machine learning algorithms; The test predicted cell types output by each machine learning algorithm are compared with the cell type labels to obtain the test accuracy of each machine learning algorithm; The target machine learning algorithm is determined based on the test time of each machine learning algorithm, the test accuracy of each machine learning algorithm, and the model robustness of each machine learning algorithm.

[0080] In an embodiment of the present invention, a test data set for plant cell type prediction is collected. The test data set includes a test cell gene matrix, i.e., gene expression data of each test cell, and a cell type label corresponding to these gene matrices, i.e., the actual cell type of each test cell.

[0081] Choose from a range of preset machine learning algorithms, such as support vector machines (SVM), random forests, gradient boosted trees (such as XGBoost, LightGBM), neural networks (such as deep learning models), etc. The above algorithms have been preliminarily trained or tuned.

[0082] The test cell gene matrix is ​​input into these preset machine learning algorithms respectively. Each algorithm will make predictions based on the input data and output the test predicted cell type.

[0083] While inputting the test data, the test time required for each algorithm to complete the prediction is recorded. The test time is used to evaluate the algorithm's computational efficiency and feasibility in practical applications.

[0084] The test predicted cell types output by each machine learning algorithm are compared with the cell type labels in the test data set, for example, by calculating evaluation indicators such as accuracy, recall, and F1 score; based on the above comparison results, the test accuracy of each machine learning algorithm is calculated.

[0085] It should be noted that in addition to test accuracy, it is also necessary to consider the test time (i.e., computational efficiency) and model robustness (i.e., the performance stability of the model under different data sets and conditions) of each machine learning algorithm. The optimal target machine learning algorithm is determined based on the comprehensive evaluation results of test accuracy, test time, and model robustness.

[0086] Through the embodiments of the present invention, the performance of multiple machine learning algorithms on the plant cell type prediction task can be systematically evaluated, and the target machine learning algorithm can be determined by combining a comprehensive evaluation of accuracy, test time, and model robustness.

[0087] According to a plant cell type prediction method based on explainable machine learning provided by the present invention, the target machine learning algorithm is the LightGBM algorithm.

[0088] The following describes the plant cell type prediction method based on interpretable machine learning provided by the present invention, with an example in a practical application scenario, which specifically includes the following steps.

[0089] refer to Figure 2 , Figure 2 It is a schematic diagram of the overall process of the plant cell type prediction method based on interpretable machine learning provided by the present invention, which includes: data processing (Data Processing), model training (Train Model) and downstream analysis (Downstream Analysis).

[0090] like Figure 2 As shown in the figure, data processing includes inputting single-cell RNA sequencing, namely scRNA (Genes), followed by feature selection of the input data to identify top genes with highly variable expression, and constructing a cell gene matrix based on the top genes and cells, where the rows of the cell gene matrix represent genes, the columns represent cells, and the values ​​in the matrix represent the expression levels of the genes in the cells.

[0091] Model training specifically includes the following modules: Logistic Regression, Support Vector Machine, CatBoost, Neural Networks, and LightGBM.

[0092] Downstream analysis specifically includes: predicting cell type (Cell Type) and determining marker genes (MarkerGenes).

[0093] Step 1, data collection and preprocessing: Collect single-cell data and marker gene data of multiple plant species.

[0094] The collected raw single-cell RNA sequencing data were quality controlled to remove low-quality cells, low-expressed genes, and mitochondrial genes, ensuring the reliability of the data for subsequent analysis.

[0095] For the filtered data, the data were standardized to eliminate technical errors and make the gene expression levels of cells with different sequencing depths and library sizes comparable, thereby better presenting the differences at the gene level.

[0096] Screen for highly variable genes that are highly expressed in some cells but lowly expressed in others. These genes are often the most important for distinguishing cell types and are therefore used in subsequent analysis.

[0097] The highly variable genes were further used for data normalization analysis, and the obtained matrix was used for subsequent modeling analysis.

[0098] At the same time, single-cell data from different species are integrated at the organ level. Single-cell data from different species and the same organ from different sources are merged and batch corrected using Harmony with the data source as the variable.

[0099] Finally, we re-annotated the cell types of the integrated data. Based on known marker genes and cell characteristics in the literature, we assigned appropriate cell type labels to each cell. This process may include manual review and manual correction to ensure the accuracy of the annotation.

[0100] Step 2: Based on the upstream standard process, 5000 highly variable genes for each data were calculated as the input features of the model.

[0101] The upstream standard process here is the data collection and preprocessing in step 1 above.

[0102] Step 3, multi-classification learning model construction: By comparing the classification effects of multiple machine learning models, the optimal model was selected based on calculation accuracy, calculation time cost and model robustness. Finally, the LightGBM algorithm was used, which can effectively reduce overfitting and improve the accuracy and generalization ability of the model. A model for cell type prediction was constructed for each organ of each species.

[0103] refer to Figure 3 , Figure 3It is a schematic diagram of the comparison of the prediction effects of the model based on the LightGBM algorithm provided by the present invention, including comparison of the area under the curve (AUC) of the model between different cell types (Model AUC Comparison in each Cell Type), and the horizontal axis includes: Seurat prediction (prediction_Seurat), singleR prediction (prediction_singleR), scPred prediction (prediction_scPred) and scMarker prediction (prediction_scMarker).

[0104] like Figure 3 As shown, the cell types include: Root endodermis, Sclerenchyma, Pericycle, root cortex, meristematic cell, Exodermis, Root hair, Root epidermis, Root cap, Root stele, Protoxylem, phloem and Metaxylem.

[0105] Meristem cells, outer cortex, root hairs, epidermal cells, root cap, spindle, and secondary xylem and phloem Modeling is performed on the single-cell data (cell × gene matrix) of each organ of each species, and the cell type of each cell is classified (for example, epidermis, pith, outer epidermis, etc.). The marker genes of each cell type in different organs are obtained based on the importance of feature ranking of the multi-classification model. The obtained genes can be used for downstream experimental verification and other work.

[0106] It should be noted that single-cell technology is not used in many plant species, about 30 species, and only a few species (rice, corn, Arabidopsis) have multiple organs studied, while most other species have only one organ. In the embodiment of the present invention, 25 species and about 2.5 million cells are involved. The "each species" stated here refers to the species data that have been collected (i.e., a total of 25).

[0107] Step 4, marker gene mining: Based on the feature importance of the model and the analysis of SHAP importance, the marker genes specific to each cell type are screened out.

[0108] SHAP value calculation and analysis are performed on each constructed model to identify genes with high importance in cell type differentiation, and the contribution of each gene in the prediction of a specific cell type is quantified according to the SHAP value corresponding to each cell type. The final results are output to the local environment in the form of a table.

[0109] Through the above methods, the present invention can realize the prediction of cell types of various organs of various plants, and has the mining of cell type-specific marker genes and good cross-species applicability, providing a new technical means for plant single-cell research.

[0110] The plant cell type prediction device based on explainable machine learning provided by the present invention is described below. The plant cell type prediction device based on explainable machine learning described below and the plant cell type prediction method based on explainable machine learning described above can be referenced to each other.

[0111] refer to Figure 4 , Figure 4 It is a schematic diagram of the structure of a plant cell type prediction device based on interpretable machine learning provided by the present invention.

[0112] An acquisition module 401 is used to acquire single cell data of a target organ of a target plant; A preprocessing module 402 is used to preprocess the single cell data to obtain standardized single cell data; A screening module 403 is used to perform feature screening on the standardized single-cell data to obtain highly variable genes in the single-cell data, wherein the expression level of the highly variable genes in the target cells is higher than the first expression threshold, and the expression level of the highly variable genes in other cells is lower than the first expression threshold; A normalization module 404 is used to perform normalization based on single cell data and highly variable genes to obtain a cell gene matrix; A determination module 405 is used to determine a target multi-classification model corresponding to a target organ of a target plant; The prediction module 406 is used to input the cell gene matrix into the target multi-classification model to obtain the predicted cell type output by the target multi-classification model.

[0113] Specifically, the plant cell type prediction device based on explainable machine learning provided by the present invention can implement all the method steps implemented in the above-mentioned plant cell type prediction method embodiment based on explainable machine learning, and can achieve the same technical effect. The parts and beneficial effects of this embodiment that are the same as the method embodiment will not be described in detail here.

[0114] Figure 5 is a schematic diagram of the physical structure of the electronic device provided by the present invention, such as Figure 5As shown, the electronic device may include: a processor 510, a communication interface 520, a memory 530 and a communication bus 540, wherein the processor 510, the communication interface 520 and the memory 530 communicate with each other through the communication bus 540. The processor 510 may call the logic instructions in the memory 530 to execute a plant cell type prediction method based on interpretable machine learning, the method comprising: obtaining single cell data of a target organ of a target plant; preprocessing the single cell data to obtain standardized single cell data; feature screening the standardized single cell data to obtain highly variable genes of the single cell data, wherein the expression level of the highly variable genes in the target cells is higher than the first expression threshold, and the expression level of the highly variable genes in other cells is lower than the first expression threshold; normalizing based on the single cell data and the highly variable genes to obtain a cell gene matrix; determining a target multi-classification model corresponding to the target organ of the target plant; inputting the cell gene matrix into the target multi-classification model to obtain a predicted cell type output by the target multi-classification model.

[0115] In addition, the logic instructions in the above-mentioned memory 530 can be implemented in the form of a software functional unit and can be stored in a computer-readable storage medium when it is sold or used as an independent product. Based on such an understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art or the part of the technical solution, can be embodied in the form of a software product, and the computer software product is stored in a storage medium, including a number of instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), disk or optical disk, etc. Various media that can store program codes.

[0116] On the other hand, the present invention also provides a computer program product, which includes a computer program, which can be stored on a non-transitory computer-readable storage medium. When the computer program is executed by a processor, the computer can execute the plant cell type prediction method based on explainable machine learning provided by the above-mentioned methods, and the method includes: obtaining single-cell data of the target organ of the target plant; preprocessing the single-cell data to obtain standardized single-cell data; performing feature screening on the standardized single-cell data to obtain highly variable genes in the single-cell data, wherein the expression level of the highly variable genes in the target cells is higher than the first expression threshold, and the expression level of the highly variable genes in other cells is lower than the first expression threshold; normalizing based on the single-cell data and the highly variable genes to obtain a cell gene matrix; determining a target multi-classification model corresponding to the target organ of the target plant; inputting the cell gene matrix into the target multi-classification model to obtain the predicted cell type output by the target multi-classification model.

[0117] On the other hand, the present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, is implemented to execute the plant cell type prediction method based on explainable machine learning provided by the above-mentioned methods, the method comprising: obtaining single-cell data of a target organ of a target plant; preprocessing the single-cell data to obtain standardized single-cell data; performing feature screening on the standardized single-cell data to obtain highly variable genes in the single-cell data, wherein the expression level of the highly variable genes in the target cells is higher than a first expression threshold, and the expression level of the highly variable genes in other cells is lower than the first expression threshold; normalizing based on the single-cell data and the highly variable genes to obtain a cell gene matrix; determining a target multi-classification model corresponding to the target organ of the target plant; inputting the cell gene matrix into the target multi-classification model to obtain a predicted cell type output by the target multi-classification model.

[0118] The device embodiments described above are merely illustrative, wherein the units described as separate components may or may not be physically separated, and the components displayed as units may or may not be physical units, that is, they may be located in one place, or they may be distributed on multiple network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the scheme of this embodiment. Ordinary technicians in this field can understand and implement it without paying creative labor.

[0119] Through the description of the above implementation methods, those skilled in the art can clearly understand that each implementation method can be implemented by means of software plus a necessary general hardware platform, and of course, can also be implemented by hardware. Based on this understanding, the above technical solution is essentially or the part that contributes to the prior art can be embodied in the form of a software product, and the computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, a disk, an optical disk, etc., including a number of instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the methods described in each embodiment or some parts of the embodiments.

[0120] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A plant cell type prediction method based on interpretable machine learning, characterized in that: include: Obtain single-cell data of target organs of target plants; Preprocessing the single cell data to obtain standardized single cell data; Performing feature screening on the standardized single-cell data to obtain highly variable genes in the single-cell data, wherein the expression level of the highly variable genes in the target cells is higher than a first expression threshold, and the expression level of the highly variable genes in other cells is lower than the first expression threshold; Normalizing the single cell data and the highly variable genes to obtain a cell gene matrix; determining a target multi-classification model corresponding to a target organ of the target plant; The cell gene matrix is ​​input into the target multi-classification model to obtain the predicted cell type output by the target multi-classification model.

2. The plant cell type prediction method based on interpretable machine learning according to claim 1, characterized in that: After inputting the cell gene matrix into the target multi-classification model to obtain the predicted cell type output by the target multi-classification model, the method further includes: The target multi-classification model is subjected to SHAP value analysis to obtain specific marker genes corresponding to the predicted cell types.

3. The plant cell type prediction method based on interpretable machine learning according to claim 1, characterized in that: The preprocessing of the single cell data to obtain standardized single cell data includes: Determine the original single-cell RNA sequencing data of the single-cell data; Performing quality control processing on the original single-cell RNA sequencing data to obtain filtered single-cell RNA sequencing data, wherein the quality control processing includes: removing low-expression genes and removing mitochondrial genes, wherein the expression level of the low-expression gene is lower than a second expression threshold; The sequencing depth and gene expression level of the filtered single-cell RNA sequencing data are standardized to obtain standardized single-cell data.

4. The plant cell type prediction method based on interpretable machine learning according to claim 1, characterized in that: Before acquiring the single cell data of the target organ of the target plant, the method further includes: Obtain single-cell data from different organs of multiple plant species; Preprocessing and annotating the single-cell data to obtain a cell type label for each single-cell data; Performing feature screening on the single cell data based on the cell type label to obtain a target number of highly variable genes for each single cell data; Normalizing each single cell data and the highly variable genes of each single cell data to obtain a cell gene matrix of each single cell; Classifying the cell gene matrix of each single cell according to different organs of the multiple plant species to obtain a training data set for each organ of each plant species, wherein the training data set includes: the cell gene matrix and a cell type label corresponding to the cell gene matrix; Based on the training data set of each organ of each plant species and the target machine learning algorithm, a multi-classification model corresponding to each organ of each plant species is constructed.

5. The method for predicting plant cell types based on interpretable machine learning according to claim 4, characterized in that: Before constructing a multi-classification model corresponding to each organ of each plant species based on the training data set of each organ of each plant species and the target machine learning algorithm, the method further includes: Acquire a test data set for plant cell type prediction, wherein the test data set includes a test cell gene matrix and a cell type label corresponding to the test cell gene matrix; Inputting the test cell gene matrix into a plurality of preset machine learning algorithms respectively, and obtaining the test predicted cell type and test time output by each of the plurality of machine learning algorithms; Comparing the test predicted cell types output by each machine learning algorithm with the cell type labels respectively to obtain the test accuracy of each machine learning algorithm; A target machine learning algorithm is determined based on the test time of each machine learning algorithm, the test accuracy of each machine learning algorithm, and the model robustness of each machine learning algorithm.

6. The method for predicting plant cell types based on interpretable machine learning according to claim 4, characterized in that: The target machine learning algorithm is the LightGBM algorithm.

7. A plant cell type prediction device based on interpretable machine learning, characterized in that: include: An acquisition module, used for acquiring single-cell data of a target organ of a target plant; A preprocessing module, used for preprocessing the single cell data to obtain standardized single cell data; A screening module, used for performing feature screening on the standardized single-cell data to obtain highly variable genes in the single-cell data, wherein the expression level of the highly variable genes in the target cells is higher than a first expression threshold, and the expression level of the highly variable genes in other cells is lower than the first expression threshold; A normalization module, used for normalizing the single cell data and the highly variable genes to obtain a cell gene matrix; A determination module, used to determine a target multi-classification model corresponding to a target organ of the target plant; A prediction module is used to input the cell gene matrix into the target multi-classification model to obtain the predicted cell type output by the target multi-classification model.

8. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein: When the processor executes the computer program, the plant cell type prediction method based on interpretable machine learning as described in any one of claims 1 to 6 is implemented.

9. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the plant cell type prediction method based on interpretable machine learning as described in any one of claims 1 to 6 is implemented.

10. A computer program product, comprising a computer program, characterized in that When the computer program is executed by a processor, the plant cell type prediction method based on interpretable machine learning as described in any one of claims 1 to 6 is implemented.

Images