A machine-learned genomic selection system and method
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- SHANDONG AGRICULTURAL UNIVERSITY
- Filing Date
- 2026-03-12
- Publication Date
- 2026-05-29
AI Technical Summary
Traditional genomic selection methods struggle to handle nonlinear relationships, have low efficiency in processing high-dimensional data, and are technically demanding, limiting their application in animal breeding.
This invention provides a machine learning-based genome selection system, including phenotypic processing, genotypic processing, machine learning methods, and data visualization modules. It supports multiple data formats and algorithms, provides automated hyperparameter optimization and interactive chart operations, and simplifies the operation process.
It significantly lowers the technical threshold, enabling breeding researchers without a programming background to efficiently analyze genomic data, improving the efficiency and accuracy of genomic selection, and promoting the widespread application of machine learning technology in animal breeding.
Smart Images
Figure CN122117009A_ABST
Abstract
Description
Technical Field
[0001] This invention belongs to the field of genome breeding technology, specifically relating to a machine learning-based genome selection system and method. Background Technology
[0002] In genomic data, nonlinear relationships may exist between genes, which traditional linear regression methods often struggle to handle. Machine learning methods can capture these complex interactions, thereby improving prediction accuracy. Genomic data in animal breeding often has extremely high dimensionality, and traditional statistical methods may encounter the "curse of dimensionality" with high-dimensional data. Machine learning methods can effectively process high-dimensional data and extract useful patterns from it in a data-driven manner.
[0003] Furthermore, existing genomic selection methods typically require complex operations and advanced programming skills, presenting a significant technical barrier for many breeding researchers. Traditional genomic selection tools are cumbersome to operate, and most only support basic linear models, failing to flexibly combine multiple algorithms for optimization, thus limiting their application in practical breeding work. Summary of the Invention
[0004] To address the aforementioned technical problems, this invention provides a machine learning-based genome selection system and method to solve the problems in the prior art. The technical solution adopted by this invention is as follows: A machine learning-based genome selection system includes: The phenotypic processing module is used for importing phenotypic data, filling missing values, detecting outliers, and standardizing and normalizing. The genotype processing module is used to import genotype data and perform encoding conversion, missing value handling, and data filtering. The Machine Learning Methods module is used to train and evaluate models using machine learning algorithms and provides hyperparameter optimization capabilities. The data visualization module is used to display the distribution of genotype-phenotype data and model evaluation results.
[0005] Furthermore, the phenotyping module provides multiple missing value imputation methods and allows users to select the imputation method according to their needs; including: The data import unit supports importing tabular data files in CSV, Excel, and TXT formats. The missing value processing unit is used to automatically identify missing values in the data and provide at least two missing value imputation methods, including mean imputation, median imputation and interpolation. The outlier detection unit is used to detect outliers using box plots, Z-scores, or IQR methods, and decides whether to delete or correct outliers based on user selection. Standardized units are used to standardize phenotypic data so that their mean is zero and their standard deviation is one.
[0006] Furthermore, the genotype processing module includes: The genotype data import unit is used to import genotype data in VCF and PLINK formats and supports automatic sample and label identification. It supports multiple genotype data formats, is used to remove SNP markers with missing values exceeding a predetermined threshold, and supports minimum allele frequency filtering and Hardy-Weinberg balance detection.
[0007] Furthermore, the machine learning methods module includes: The model training unit is used to select one or more machine learning algorithms for modeling based on the training data. The algorithms include kernel ridge regression, support vector regression, random forest, and K-nearest neighbor. Cross-validation unit, used to perform K-fold cross-validation to evaluate the accuracy and stability of the model; The hyperparameter optimization unit is used to optimize the hyperparameters of the model through grid search or random search, and the hyperparameters include learning rate, tree depth, and regularization parameter; The model evaluation unit is used to calculate evaluation metrics based on the prediction results, including accuracy, precision, MSE, and RMSE.
[0008] Furthermore, the data visualization module includes: The data distribution visualization unit is used to visualize the distribution of phenotypic and genotypic data through histograms, star plots, or scatter plots. Feature importance visualization unit, used to show the importance of each genotype marker in the machine learning model, including bar charts and scatter plots; The model evaluation visualization unit is used to generate and display the evaluation results of the model's confusion matrix and ROC curve; Interactive chart units are used to support user interaction with charts, including selecting displayed features, adjusting parameters, and zooming in and out.
[0009] A machine learning-based genome selection method includes: Step 1, Phenotypic Processing: This step involves importing phenotypic data, imputing missing values, detecting outliers, and standardizing and normalizing the data. Step 2, Genotype Processing: This step involves importing genotype data and performing encoding conversion, missing value handling, and data filtering. Step 3, Machine Learning Methods: Used to train and evaluate models using machine learning algorithms, and provides hyperparameter optimization capabilities; Step 4, data visualization, is used to display the distribution of genotype-phenotype data and model evaluation results.
[0010] The present invention has the following beneficial effects: This invention features high stability, ease of operation, and high integration, significantly reducing the technical threshold for genome selection. It enables breeding researchers without a programming background to efficiently perform genome data analysis, promoting the widespread application of machine learning technology in animal breeding. Attached Figure Description
[0011] Figure 1 This is a flowchart. Detailed Implementation
[0012] The following will be described in conjunction with embodiments of the present invention. Figure 1 The technical solutions in the embodiments of the present invention will be clearly and completely described. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Unless otherwise specified, the technical means used in the embodiments are conventional means well known to those skilled in the art.
[0013] This invention provides a machine learning-based genome selection system and method, enabling breeding researchers to complete the entire process from data preprocessing to genome selection model training, evaluation, and prediction through simple selection operations. The platform supports multiple machine learning algorithms and provides automated hyperparameter tuning, model selection, and result visualization functions, significantly improving the efficiency and accuracy of genome selection.
[0014] A machine learning-based genome selection system includes: The phenotypic processing module is used to import phenotypic data, fill in missing values, detect outliers, and standardize and normalize it. The phenotypic processing module can be a built-in component of the software. It first receives phenotypic data files in CSV, Excel or TXT format uploaded by the user through the data import unit.
[0015] The phenotypic data processing module is responsible for collecting, organizing, and preprocessing phenotypic data related to genotype data. To improve data quality and prediction accuracy, a linear model is used to correct the phenotypic data, eliminating the influence of environment-related fixed and random effects.
[0016] Building upon this foundation, the system automatically detects, labels, or removes outlier data points through model fitting and data analysis, while simultaneously standardizing the data. By controlling the influence of these environmental factors, the accuracy of subsequent genotype and phenotype data is ensured, thereby improving the reliability of model training. The module provides data visualization support based on user needs, such as histograms to display data distribution.
[0017] The genotype processing module is used to import genotype data and perform encoding conversion, missing value handling, and data filtering. The genotype data processing module is responsible for processing and analyzing genotype data, including importing genotype files (supporting multiple formats such as VCF, BAM, and PLINK), data cleaning, and quality control (e.g., filtering low-frequency SNPs, removing Hardy-Weinberg imbalance markers). This module integrates the functionality of the PLINK software, allowing users to process and quality control genomic data by calling PLINK in the background as needed. PLINK functionality is seamlessly integrated through subprocess calls and a command builder. The system builds a PLINK command template library covering all commonly used quality control and analysis functions. Command execution uses an asynchronous mode, supporting real-time progress monitoring and output capture. The genotype processing module calls external software; it first receives user-uploaded VCF or PLINK format genotype data files through the genotype data import unit and automatically identifies samples and markers.
[0018] The Machine Learning Methods module is used to train and evaluate models using machine learning algorithms and provides hyperparameter optimization capabilities. The machine learning module integrates various machine learning algorithms for training and optimizing genome selection models. It supports model selection, hyperparameter tuning, training process monitoring, and model evaluation. This module offers multiple machine learning algorithm options, such as kernel ridge regression, random forest, K-nearest neighbor, support vector regression, and neural networks. It supports hyperparameter optimization methods such as grid search, random search, and Bayesian optimization, and automatically records and compares model performance under different hyperparameter combinations. It supports K-fold cross-validation to improve model robustness. It provides various performance metrics, such as MSE, RMSE, and R-squared for regression tasks. 2 To evaluate model performance; the machine learning module can be a comprehensive software framework that receives preprocessed phenotypic and genotypic data through the model training unit. Users can select appropriate machine learning algorithms for model training based on their specific needs.
[0019] The data visualization module is used to display the distribution of genotype-phenotype data and model evaluation results.
[0020] The data visualization module is used to display and analyze data, model results, and predictions. It provides interactive charts and custom plotting capabilities to help users gain a deeper understanding of data distribution, model performance, and prediction results. The module uses a plotting engine based on Matplotlib and Seaborn, combined with Plotly to provide basic interactive functionality. The module is designed with a three-layer structure: a data adaptation layer, a plotting engine layer, and an output processing layer. The data adaptation layer is responsible for format conversion and standardization of genotype, phenotype, and model output data. The plotting engine layer supports various visualization formats such as heatmaps, box plots, and histograms. The module allows users to customize chart styles and parameter configurations, improving the flexibility of data analysis and result interpretation.
[0021] Furthermore, the phenotyping module provides multiple missing value imputation methods and allows users to select the imputation method according to their needs; including: The data import unit supports importing tabular data files in CSV, Excel, and TXT formats. It employs a multi-mode parser design, supporting adaptive parsing of various file formats. The system has a built-in format detector that automatically recognizes CSV, TSV, Excel, and plain text formats. Encoding detection integrates the chardet library, automatically recognizing multiple encoding formats such as UTF-8, GBK, and ASCII. Delimiter recognition uses a regular expression engine, supporting mixed delimiter scenarios. Users can also input column names for fixed effects and covariates according to specific needs for data correction. The missing value processing unit is used to automatically identify missing values in the data and provide at least two missing value imputation methods, including mean imputation, median imputation and interpolation. The outlier detection unit is used to detect outliers using box plots, Z-scores, or IQR methods, and decides whether to delete or correct outliers based on user selection. Standardized units are used to standardize phenotypic data so that their mean is zero and their standard deviation is one.
[0022] Furthermore, the genotype processing module includes: The genotype data import unit is used to import genotype data in VCF and PLINK formats and supports automatic sample and label identification. It supports multiple genotype data formats for deleting SNP markers with missing values exceeding a predetermined threshold, and supports minimum allele frequency filtering and Hardy-Weinberg balance detection. Each quality control step is configurable, and the system provides recommended values. The quality control process generates detailed reports, including statistical information such as the number of markers before and after filtering, sample size, and missing value distribution. Reordering and selective execution of quality control steps are supported.
[0023] Furthermore, the machine learning methods module includes: The model training unit is used to select one or more machine learning algorithms for modeling based on the training data. These algorithms include kernel ridge regression, support vector regression, random forest, and K-nearest neighbors. The algorithm layer is a deeply customized development based on scikit-learn version 1.3+, encapsulating all algorithms through the adapter pattern and providing a unified fit / predict interface. Specific algorithm implementations include: kernel ridge regression using sklearn.kernel_ridge.KernelRidge, supporting kernel functions such as RBF and multinomial; random forest based on sklearn.ensemble.Random Forest Regressor for feature importance evaluation; and neural networks integrated with the PyTorch framework, providing fully connected and convolutional network options. The cross-validation unit is used to perform K-fold cross-validation to evaluate the accuracy and stability of the model; it is implemented based on sklearn.model_selection.KFold to implement standard K-fold cross-validation. The hyperparameter optimization unit is used to optimize the model's hyperparameters, including learning rate, tree depth, and regularization parameters, through grid search or random search. The hyperparameter optimization module adopts a modular design and incorporates the following technologies: grid search based on sklearn.model_selection.GridSearchCV, supporting exhaustive multi-parameter combinations; random search using the Hyperopt library's Random Search for efficient parameter space sampling; and Bayesian optimization integrating the Optuna framework for intelligent parameter search based on the TPE algorithm. The optimization process supports an early stopping mechanism, dynamically terminating undesirable parameter combinations based on validation set performance. A result caching system avoids redundant computation and supports pausing and resuming the optimization process. Distributed optimization is implemented through the Dask task scheduler, supporting parallel computation across multiple nodes. The model evaluation unit is used to calculate evaluation metrics based on the prediction results, including accuracy, precision, MSE, and RMSE.
[0024] Furthermore, the data visualization module includes: The data distribution visualization unit is used to visualize the distribution of phenotypic and genotypic data through histograms, star plots, or scatter plots. Feature importance visualization unit, used to show the importance of each genotype marker in the machine learning model, including bar charts and scatter plots; The model evaluation visualization unit is used to generate and display the evaluation results of the model's confusion matrix and ROC curve; Interactive chart units, built on Tkinter or PyQt, feature simple parameter adjustment panels that allow users to adjust parameters in real time using sliders, drop-down menus, and color pickers. This supports interactive operations on the charts, including selecting displayed features, adjusting parameters, and zooming in and out.
[0025] A machine learning-based genome selection method includes: Step 1, Phenotypic Processing: This step involves importing phenotypic data, imputing missing values, detecting outliers, and standardizing and normalizing the data. Step 2, Genotype Processing: This step involves importing genotype data and performing encoding conversion, missing value handling, and data filtering. Step 3, Machine Learning Methods: Used to train and evaluate models using machine learning algorithms, and provides hyperparameter optimization capabilities; Step 4, data visualization, is used to display the distribution of genotype-phenotype data and model evaluation results.
[0026] This invention takes the prediction of growth traits in Large White pigs as an example and provides the following specific implementation methods: This embodiment demonstrates how to use a machine learning-based genomic selection platform to predict growth traits in Large White pigs, and how to use the support vector regression (SVR) algorithm for model training, optimization, and evaluation.
[0027] In a certain region, growth trait data were collected from 3288 pigs of 1644 Large White pigs, including data on age at 100 kg and backfat thickness at 100 kg. Genotypic data involved 1644 individuals.
[0028] Quality control was performed on the genotype data, with the specific standards being: a deletion rate of less than 10% in the sample, a minimum allele frequency greater than 0.05, and a Hardy-Weinberg equilibrium (hWE) value greater than 10. ^-6 These standards ensure that the SNP data used is sufficiently representative of the population and provides accurate genotype data for model training.
[0029] Quality control was performed on the phenotypic data, and a linear model was used to correct for the effects of environmental factors.
[0030] Support Vector Regression (SVR) is a regression method based on Support Vector Machines (SVM) that can handle both linear and nonlinear regression problems. In this embodiment, the SVR model uses the RBF kernel function, which effectively captures the nonlinear relationship between genotype and phenotype. By selecting an appropriate kernel function, SVR can handle complex nonlinear regression tasks, and is particularly suitable for the complex interactions between genotype and phenotype in genome selection tasks.
[0031] During SVR model training, Bayesian optimization is used to tune the model's hyperparameters. Bayesian optimization constructs a Gaussian process regression (GPR) as a surrogate model to find the optimal hyperparameters with a smaller number of experiments. The main hyperparameters include: C: a penalty parameter that controls the balance between training error and model complexity; ε: defining the tolerance for regression error; and gamma: a parameter of the kernel function that controls the width of the Gaussian kernel. Through Bayesian optimization, we can find an optimal combination of hyperparameters that achieves the best balance between prediction accuracy and computational efficiency in the SVR model.
[0032] Ten-fold cross-validation is used to evaluate the performance of the SVR model. In each cross-validation, the dataset is split into 10 subsets: 9 subsets for training and 1 subset for validation. The final model performance is evaluated by combining the results from different folds. Ten-fold cross-validation assesses the model's stability and generalization ability by splitting the training and validation sets multiple times, avoiding errors caused by a single validation set. Pearson correlation coefficient (PCC): measures the linear correlation between the model's predicted values and the actual values. The closer the PCC value is to 1, the more accurate the model's prediction. Mean squared error (MSE) and root mean square error (RMSE): measure the error between the predicted results and the actual values; smaller values indicate a better model fit.
[0033] When using SVR for genome-wide selection, the table shows that SVR performs well in predicting both growth traits.
[0034] In this embodiment, for genome-wide selection of growth traits in Large White pigs, the nonlinear relationship between phenotypic and genotypic data was successfully captured using the Support Vector Regression (SVR) algorithm. Bayesian optimization was employed to optimize the model's hyperparameters, further improving the model's prediction accuracy and computational efficiency. Evaluation methods such as 10-fold cross-validation ensured the model's robustness and accuracy.
[0035] The genomic selection platform of this invention is not only applicable to the prediction of growth traits, but can also be extended to the application of genomic selection for other animal traits, showing broad practical application prospects. Its innovation and efficiency will provide strong technical support for precision breeding in modern animal husbandry, promoting the popularization and development of genomic breeding technology.
[0036] The above embodiments are merely descriptions of preferred embodiments of the present invention and are not intended to limit the scope of the present invention. Any modifications, alterations, alterations, or substitutions made by those skilled in the art to the technical solutions of the present invention without departing from the spirit of the present invention should fall within the protection scope defined by the claims of the present invention.
Claims
1. A machine learning-based genome selection system, characterized in that, include: The phenotypic processing module is used for importing phenotypic data, filling missing values, detecting outliers, and standardizing and normalizing. The genotype processing module is used to import genotype data and perform encoding conversion, missing value handling, and data filtering. The Machine Learning Methods module is used to train and evaluate models using machine learning algorithms and provides hyperparameter optimization capabilities. The data visualization module is used to display the distribution of genotype-phenotype data and model evaluation results.
2. The machine learning genome selection system according to claim 1, characterized in that, The phenotyping module provides multiple missing value imputation methods and allows users to select the imputation method according to their needs; including: The data import unit supports importing tabular data files in CSV, Excel, and TXT formats. The missing value processing unit is used to automatically identify missing values in the data and provide at least two missing value imputation methods, including mean imputation, median imputation and interpolation. The outlier detection unit is used to detect outliers using box plots, Z-scores, or IQR methods, and decides whether to delete or correct outliers based on user selection. Standardized units are used to standardize phenotypic data so that their mean is zero and their standard deviation is one.
3. The machine learning genome selection system according to claim 1, characterized in that, The genotype processing module includes: The genotype data import unit is used to import genotype data in VCF and PLINK formats and supports automatic sample and label identification. It supports multiple genotype data formats, is used to remove SNP markers with missing values exceeding a predetermined threshold, and supports minimum allele frequency filtering and Hardy-Weinberg balance detection.
4. The machine learning genome selection system according to claim 1, characterized in that, The machine learning methods module includes: The model training unit is used to select one or more machine learning algorithms for modeling based on the training data. The algorithms include kernel ridge regression, support vector regression, random forest, and K-nearest neighbor. Cross-validation unit, used to perform K-fold cross-validation to evaluate the accuracy and stability of the model; The hyperparameter optimization unit is used to optimize the hyperparameters of the model through grid search or random search, and the hyperparameters include learning rate, tree depth, and regularization parameter; The model evaluation unit is used to calculate evaluation metrics based on the prediction results, including accuracy, precision, MSE, and RMSE.
5. A machine learning-based genome selection system according to claim 1, characterized in that, The data visualization module includes: The data distribution visualization unit is used to visualize the distribution of phenotypic and genotypic data through histograms, star plots, or scatter plots. Feature importance visualization unit, used to show the importance of each genotype marker in the machine learning model, including bar charts and scatter plots; The model evaluation visualization unit is used to generate and display the evaluation results of the model's confusion matrix and ROC curve; Interactive chart units are used to support user interaction with charts, including selecting displayed features, adjusting parameters, and zooming in and out.
6. A machine learning-based genome selection method, based on the machine learning-based genome selection system of claim 1, characterized in that, include: Step 1, Phenotypic Processing: This step involves importing phenotypic data, imputing missing values, detecting outliers, and standardizing and normalizing the data. Step 2, Genotype Processing: This step involves importing genotype data and performing encoding conversion, missing value handling, and data filtering. Step 3, Machine Learning Methods: Used to train and evaluate models using machine learning algorithms, and provides hyperparameter optimization capabilities; Step 4, data visualization, is used to display the distribution of genotype-phenotype data and model evaluation results.