Beef cattle breeding genome selection auxiliary decision-making system

Abstract

The invention discloses an auxiliary decision-making system for beef cattle breeding genome selection, and relates to the technical field of beef cattle breeding screening. The system comprises a data acquisition and quality control module, a function annotation and feature construction module, an interpretable multi-character prediction and gene-environment interaction modeling module, an economic weight mapping and multi-target selection optimization module, a pairing simulation and long-term evaluation module and an interaction decision and closed-loop learning module. According to the method, auditing effect attribution of gene-environment interaction is incorporated into an economic weight mapping and pairing optimization process, and long-term heredity and economic consequences of candidate schemes are evaluated through intergenerational simulation. The implementation result and the actual representative form are periodically written back and used for model updating to form a self-adaptive closed loop, so that the field environment difference, seasonal fluctuation and long-term genetic sustainability are clearly quantified and managed in decision making, and the scene suitability and the sustainability of long-term breeding income are fundamentally improved.

Description

Technical Field

[0001] This invention relates to the field of beef cattle breeding screening technology, specifically to a beef cattle breeding genome selection auxiliary decision-making system. Background Technology

[0002] As the beef cattle industry transforms and upgrades towards high quality and efficiency, genomic selection technology has become a core supporting means to improve beef cattle breeding efficiency and optimize the genetic quality of the population. By mining the correlation between genomic data and economic traits, early selection and precision breeding can be achieved, which is of great significance for reducing fattening costs and improving market competitiveness. However, existing beef cattle genomic selection-related technologies and auxiliary tools have a core shortcoming: the lack of an end-to-end closed-loop decision-making system that comprehensively considers gene-environment interaction effects, long-term genetic sustainability, and scenario-based economic value. This makes it difficult to meet the actual needs of large-scale, refined breeding, specifically as follows:

[0003] Existing technologies mostly focus on point estimation of genome-based breeding values ​​(GEBV), neglecting the impact of gene (G) and environment (E) interactions on breeding outcomes. They also lack auditable attribution mechanisms for key gene loci of target traits, environmental factors, and the intensity of their interactions. Furthermore, the economic weights are fixed and not dynamically adjusted in conjunction with scenario-based cost-benefit parameters. There is insufficient coordination of constraints such as inbreeding control and population diversity preservation. Moreover, there is a lack of pedigree-level generational simulation mechanisms to predict long-term breeding outcomes, and no adaptive mechanism has been established for writing back actual breeding results and periodically updating the model. This results in poor scenario adaptability, insufficient sustainability of long-term breeding benefits, and an inability to fully realize the core value of genomic data.

[0004] In response to this, this application proposes a genomic selection-assisted decision-making system for beef cattle breeding to address the aforementioned issues. Summary of the Invention

[0005] The purpose of this invention is to provide a genomic selection-assisted decision-making system for beef cattle breeding, in order to solve the problems of poor scenario adaptability, insufficient sustainability of long-term breeding benefits, and inability to fully realize the core value of genomic data in existing technologies.

[0006] To achieve the above objectives, the present invention provides the following technical solution:

[0007] A genomic selection-assisted decision-making system for beef cattle breeding includes:

[0008] The data acquisition and quality control module is used to acquire genomic data, preprocess the genomic data and calculate the confidence level to obtain integrated data with confidence level.

[0009] The functional annotation and feature construction module is used to perform gene functional annotation on the integrated data with confidence, construct linkage fragments and temporal phenotypic features, and output an interpretable feature set;

[0010] An interpretable multi-trait prediction and gene-environment interaction modeling module is used to output the genomic estimated breeding value and its confidence interval and G×E sensitivity score for each individual based on the interpretable feature set;

[0011] The economic weight mapping and multi-objective selection optimization module is used to map the breeding value into a scenario-based economic value and generate selection priorities and candidate pairings under inbreeding and diversity constraints.

[0012] The pairing simulation and long-term evaluation module is used to perform intergenerational simulation of the candidate pairings and output long-term genetic and economic trajectories.

[0013] The interactive decision-making and closed-loop learning module is used to present decisions, receive implementation results, and write back actual post-model and health data for periodic model updates.

[0014] Furthermore, the interpretable multi-trait prediction and gene-environment interaction modeling module employs an uncertainty propagation mechanism: using the feature set and its confidence level as input, it generates the confidence interval and probability distribution representation of the genome estimated breeding value (GEBV) for each trait, and outputs a G×E sensitivity score and corresponding confidence level for each individual, which can be used for subsequent economic mapping and risk-aware selection. G represents genes, and E represents the environment, which can be used for subsequent economic mapping and risk-aware selection.

[0015] Furthermore, the interpretable multi-trait prediction and gene-environment interaction modeling module is used to generate an effect attribution report. The effect attribution report includes: gene loci or chromosome segments that significantly contribute to the target trait, key environmental factors and their interaction strengths, and describes the source of attribution evidence and uncertainty in an auditable format to facilitate manual verification and compliance review.

[0016] Furthermore, the economic weight mapping and multi-objective selection optimization module weights the GEBV of each trait based on scenario-based cost-benefit parameters (including fattening costs, breeding inputs, market prices, and risk discounting) and outputs candidate individual ranking and pairing suggestions under constraints such as inbreeding cap, preservation of population diversity, and regulatory site avoidance. At the same time, it provides an expected benefit and constraint satisfaction description for each candidate scheme.

[0017] Furthermore, the pairing simulation and long-term evaluation module performs pedigree-level Monte Carlo simulations based on the candidate pairings to simulate recombination, selection intensity, and genetic drift, and outputs time series and confidence intervals of genetic gain, population inbreeding coefficient, and economic returns over several generations to evaluate short-term and long-term trade-offs.

[0018] Furthermore, after implementation, the interactive decision-making and closed-loop learning module receives actual breeding and production data (including phenotypic, health and environmental records), periodically adjusts the prediction model parameters and uncertainty estimates using Bayesian updates or incremental learning mechanisms, and records versioned model changes and model performance indicators to support traceable model evolution.

[0019] Furthermore, the data quality control and uncertainty estimation module includes: site-level filtering and low-coverage sequencing imputation for genotypes, batch effect correction and measurement error modeling for phenotypes, and generates confidence metadata (including source identifier, confidence score and missing mechanism description) for each input field. The confidence metadata is output along with the integrated data for use by subsequent modules.

[0020] Furthermore, the genomic data includes: phenotype, genotype, environmental and management data;

[0021] The preprocessing includes: performing quality control, imputing missing values, calculating the confidence level for each data item, and outputting integrated data with confidence levels.

[0022] Compared with existing technologies, the beef cattle breeding genome selection auxiliary decision-making system provided by this invention incorporates the auditable effects of gene-environment interactions into the economic weight mapping and pairing optimization process, and evaluates the long-term genetic and economic consequences of candidate schemes through intergenerational simulation. The implementation results and actual post-generation phenotypic data are periodically written back and used for model updates, forming an adaptive closed loop. This allows for the explicit quantification and management of field environmental differences, seasonal fluctuations, and long-term genetic sustainability in decision-making, fundamentally improving scenario adaptability and the sustainability of long-term breeding benefits. Attached Figure Description

[0023] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in this invention. For those skilled in the art, other drawings can be obtained based on these drawings.

[0024] Figure 1 A block diagram of the beef cattle breeding genomic selection auxiliary decision-making system provided in an embodiment of the present invention. Detailed Implementation

[0025] To enable those skilled in the art to better understand the technical solution of the present invention, the present invention will be further described in detail below with reference to the accompanying drawings.

[0026] As attached Figure 1 As shown:

[0027] Example 1:

[0028] A genomic selection-assisted decision-making system for beef cattle breeding includes:

[0029] The data acquisition and quality control module is used to acquire genomic data, preprocess the genomic data and calculate the confidence level to obtain integrated data with confidence level.

[0030] Specifically, the genomic data includes: phenotype, genotype, environmental and management data;

[0031] The preprocessing includes: performing quality control, imputing missing values, calculating the confidence level for each data item, and outputting integrated data with confidence levels.

[0032] Furthermore, each record is assigned a globally unique identifier (individual ID, sample batch, timestamp), a unified time base is adopted, and format standardization is performed (e.g., JSON / Parquet structured fields). Basic syntax validation and range checks are implemented before data is entered into the database. Extensible data access tools (API, batch upload, sensor stream) and metadata registration (data source, measurement device, sampling time) are provided. The output is a standardized raw data stream normalized by field and marked with source identifiers, and written to a traceable data lake or relational / column database, providing a unified interface and auditing chain for subsequent quality control and feature engineering.

[0033] Genotypes are filtered at the site and sample levels (call rate threshold, missing rate, minimum allele frequency), and low-coverage sequencing samples are subjected to positional imputation and confidence estimation based on high-quality in-site references (imputation confidence score).

[0034] The output is "high-confidence integrated data", which means that the original observations and the corresponding uncertainty / source label are passed downstream together, supporting confidence-based weighted inference and risk-sensitive decision-making.

[0035] The functional annotation and feature construction module is used to perform gene functional annotation on the integrated data with confidence, construct linkage fragments and temporal phenotypic features, and output an interpretable feature set;

[0036] Specifically, the integrated data with confidence levels is transformed into a multi-scale feature set that is interpretable and can be directly used by the model;

[0037] In terms of genes, this includes: mapping SNPs to gene coding regions, regulatory elements, conservation scores, and annotation of known pathogenic sites; and constructing chromosome segments or haplotype blocks to capture local linkage information.

[0038] Phenotypic / temporal aspects are extracted using curve fitting (e.g., growth curve slope, rising / steady-state parameters) and event aggregation (reproductive window statistics).

[0039] For environmental aspects, cumulative exposure, extreme days, and lag indicators related to phenotypes are calculated. Each feature simultaneously retains the uncertainty of its source, construction method, and propagation (input from the previous module and propagated through rules).

[0040] Furthermore, the output is a matrix representation (rows: individuals, columns: features) with accompanying feature metadata, facilitating interpretable model reading, manual review, and traceability.

[0041] An interpretable multi-trait prediction and gene-environment interaction modeling module is used to output the genomic estimated breeding value and its confidence interval and G×E sensitivity score for each individual based on the interpretable feature set;

[0042] Specifically, the interpretable multi-trait prediction and gene-environment interaction modeling module adopts an uncertainty propagation mechanism: taking the feature set and its confidence level as input, it generates the confidence interval and probability distribution representation of the genome estimated breeding value (GEBV) for each trait, and outputs the G×E sensitivity score and corresponding confidence level for each individual for subsequent economic mapping and risk-aware selection.

[0043] The interpretable multi-trait prediction and gene-environment interaction modeling module is used to generate effect attribution reports. The effect attribution reports include: gene loci or chromosome segments that significantly contribute to the target trait, key environmental factors and their interaction strengths, and describe the sources and uncertainties of attribution evidence in an auditable format to facilitate manual verification and compliance review.

[0044] Furthermore, a multi-trait modeling framework is employed. Inputs include features and their confidence levels, while outputs are point estimates and confidence intervals / probability distributions for each trait, generating a G×E sensitivity score (characterizing the strength of an individual trait's response to environmental fluctuations). Simultaneously, interpretable effect attributions are generated—listing major contributing markers, chromosomal segments, or key environmental factors and their relative contributions, along with evidence and uncertainty explanations for the attribution results. This module emphasizes "auditability": providing source tracing, parameter versions, and model performance metrics for the output.

[0045] The economic weight mapping and multi-objective selection optimization module is used to map the breeding value into a scenario-based economic value and generate selection priorities and candidate pairings under inbreeding and diversity constraints.

[0046] Specifically, the economic weight mapping and multi-objective selection optimization module weights the GEBV of each trait based on scenario-based cost-benefit parameters (including fattening costs, breeding inputs, market prices and risk discounting) and outputs candidate individual ranking and pairing suggestions under constraints such as inbreeding upper limit, preservation of population diversity and regulatory site avoidance. At the same time, it provides expected benefits and constraint satisfaction descriptions for each candidate scheme.

[0047] Furthermore, the multi-trait GEBV is linked to a scenario-based economic value system to generate economic indicators that can be used for ranking and pairing. First, the user or scenario protocol defines the trait weights and cost-benefit parameters (fattening profits, market prices, breeding costs, risk discounting, etc.). The module maps GEBV to expected economic returns while considering uncertainty. Then, multi-objective optimization is performed under a set of constraints (upper limit of inbreeding in a single generation, minimum threshold of population diversity, avoidance of regulations / pathogenic sites, and availability of breeding stock). The output is a priority list of candidate individuals and pairing schemes, along with the expected net return, risk measure, and constraint satisfaction description for each scheme. The implementation can employ mixed-integer programming, hierarchical greedy algorithms, or heuristic algorithms to meet interpretability and computational efficiency requirements, and sensitivity analysis is provided to evaluate the stability of schemes under different economic parameters.

[0048] The pairing simulation and long-term evaluation module is used to perform intergenerational simulation of the candidate pairings and output long-term genetic and economic trajectories.

[0049] Specifically, the pairing simulation and long-term evaluation module performs pedigree-level Monte Carlo simulations based on the candidate pairings, simulating recombination, selection intensity, and genetic drift, and outputs time series and confidence intervals of genetic gain, population inbreeding coefficient, and economic returns over several generations to evaluate short-term and long-term trade-offs.

[0050] Furthermore, prospective generational simulations are performed on the selected schemes to assess long-term consequences. The simulations model recombination, genetic variation, selection pressure, and genetic drift at the pedigree level, supporting Monte Carlo repeats to quantify uncertainty. Inputs include candidate pairings, initial population gene distribution, reproductive strategies, and environmental scenarios (e.g., assumptions of rising temperatures or disease outbreaks). Outputs are time series data over several generations (configurable, e.g., 5–20 years): trait mean GEBV, population inbreeding coefficient, genetic diversity indicators, and economic return trajectories, along with confidence intervals. This module helps decision-makers weigh short-term gains against long-term genetic sustainability, avoiding long-term degradation caused by local optima, while supporting robustness comparisons across different scenarios.

[0051] The interactive decision-making and closed-loop learning module is used to present decisions, receive implementation results, and write back actual post-model and health data for periodic model updates.

[0052] Specifically, the data quality control and uncertainty estimation module includes: site-level filtering and low-coverage sequencing imputation for genotypes, batch effect correction and measurement error modeling for phenotypes, and generates confidence metadata (including source identifier, confidence score and missing mechanism description) for each input field. The confidence metadata is output along with the integrated data for use by subsequent modules.

[0053] Furthermore, the model output, simulation results, and pairing suggestions are presented in the form of visual dashboards and reports, supporting user interaction (weight adjustment, constraint modification, manual prioritization / exclusion). After implementation, actual representative, health, and management data are periodically written back to the system, forming supervisory signals for incremental / Bayesian updates and parameter calibration of the predictive model. The system records model versions, training data snapshots, and performance metrics to meet traceability and compliance requirements. The closed-loop learning mechanism includes data ingestion, quality reassessment, model retraining / fine-tuning, and deployment rollback strategies, thereby enabling adaptive model improvement with production practice, gradually reducing predictive uncertainty and improving recommendation accuracy.

[0054] As shown above, the system in this application incorporates the auditable effects of gene-environment interactions into the economic weight mapping and pairing optimization process, and evaluates the long-term genetic and economic consequences of candidate schemes through intergenerational simulation. The implementation results and actual post-generation phenotypes are periodically written back and used for model updates, forming an adaptive closed loop. This allows for the explicit quantification and management of on-site environmental differences, seasonal fluctuations, and long-term genetic sustainability in decision-making, fundamentally improving scenario adaptability and the sustainability of long-term breeding benefits.

[0055] Example 2:

[0056] Scenario: Large-scale commercial breeding - commercial fattening farm (large-scale implementation);

[0057] Objectives and background:

[0058] Objective: To improve selection accuracy, economic returns, and suppress inbreeding accumulation within a commercial system primarily focused on fattening and marketable meat quality. Farm size: 5,000 breeding cows, with approximately 2,000 candidate bulls / cows annually.

[0059] 1. Data Collection and Unification

[0060] Sample and frequency: Genotyping is performed on approximately 2,000 candidate individuals each year; the phenotypes of cows and their offspring are continuously recorded for 3 years (baseline) before entering the productive selection cycle.

[0061] Genotyping: 50K SNP chips (per sample) were used for candidate breeding stock, and low-coverage sequencing 4× was performed on 50 male breeding bulls in the station (for constructing high-quality references and paternal retesting).

[0062] Phenotyps: birth weight, weaning weight, stage gain (ADG), daily feed intake / feed conversion ratio, post-slaughter marbling score, pregnancy rate (narrow window detection), etc. Weighing is automated once a month, and feed intake is continuously collected (Poisson events and daily summaries).

[0063] Environment and Management: Each cattle shed is equipped with temperature and humidity sensors (per hour), feed formulation records (per batch), and stress events (transportation, vaccines) are recorded.

[0064] Genotype quality control thresholds: sample call rate ≥ 98%; SNP site deletion rate ≤ 2%; minimum allele frequency (MAF) ≥ 0.02.

[0065] Low-coverage interpolation: Interpolation is performed bit by bit based on high-quality bull reference and core-type reference within the site, with an average interpolation accuracy of ≥ 98% for target sites.

[0066] Phenotypic error modeling: Batch effects were corrected for by month-batch; the standard errors of the automatic recording equipment were estimated using historical calibration data, which were ±0.5 kg body weight and ±3% dietary intake.

[0067] Output: Each input field is accompanied by metadata: source, confidence score (0–1), and description of the missing mechanism.

[0068] 2. Functional annotation and feature construction:

[0069] Note: SNPs are mapped to gene coding regions, known regulatory elements, and conservation scores.

[0070] Linkage fragments: Constructing chromosome segment haplotype blocks to capture population-specific linkage markers.

[0071] Temporal characteristics: The slope and peak time points were obtained by fitting the ADG curve; the environmental exposure index was calculated as "cumulative number of high-temperature days".

[0072] Output: An interpretable feature set, with each item retaining its source and uncertainty.

[0073] 3. Explainable multi-trait prediction and G×E modeling:

[0074] Input: The above feature set and its confidence level.

[0075] Output: For each individual, output the GEBV (with confidence interval) of multiple traits and the G×E sensitivity score (indicating the sensitivity of the trait to environmental fluctuations).

[0076] Results (compared to control): Selection accuracy (overall target index) improved from 0.55 to 0.62 (a relative improvement of approximately 12.7%); the average confidence interval width of GEBV decreased by approximately 15%, improving decision credibility.

[0077] 4. Economic weight mapping and multi-objective selection optimization:

[0078] Scenario-specific parameters: market price, fattening cost, breeding cost, and risk discount (annualized). Target index = 0.5 × fattening return + 0.3 × reproductive performance + 0.2 × disease resistance (weights adjustable).

[0079] Constraints: Inbreeding rate per generation capped at 1.2%; equivalent diversity threshold for the population to be preserved; avoidance of designated pathogenic / genetic defect sites.

[0080] Output: Annual candidate ranking and pairing list, accompanied by an explanation of the expected net profit increment and constraint satisfaction for each option. The expected net profit increase per capita is approximately RMB 150 per head sold.

[0081] 5. Paired simulation and long-term evaluation:

[0082] Simulation parameters: simulated recombination, selection intensity, genetic drift, simulation generations 5 years (approximately 3 generations).

[0083] Output: Target index genetic gain over 5 years: Implementation plan predicted 0.15 (normalized units) vs. control 0.12 (relative increase of 25%); cumulative rate of single-generation inbreeding decreased from 0.90% to 0.75% (relative reduction of approximately 16.7%).

[0084] 6. Interactive decision-making and closed-loop learning:

[0085] Implementation: The decision dashboard displays candidate pairings and long-term predictions; post-implementation profiles, health events, and environmental records are periodically (quarterly) rewritten; the model uses Bayesian update / incremental learning for parameter fine-tuning and is versioned.

[0086] Results feedback: After the first two years of write-back, the model's confidence in specific G×E sensitive sites improved, further enhancing the selection accuracy by approximately 2–3% in the following year.

[0087] Example 3:

[0088] Small-scale breeding farms (prioritizing resilience across multiple scenarios / locations):

[0089] Objectives and background:

[0090] Objective: To prioritize improving reproductive rate, disease resistance, and stability (low G×E susceptibility) in a multi-environment grazing system. Farm size: Three geographically dispersed small farms totaling 800 cows, with approximately 600 candidate individuals per year.

[0091] 1. Data Collection and Unification:

[0092] Genotype: Low-coverage sequencing 2× (cost control) was used for imputation, combined with public reference and in-field retest samples.

[0093] Phenotypic characteristics: Focus on recording reproductive traits (conception rate, calving interval during lactation), body condition score, and clinical disease events; weighing frequency is once every two months, and reproductive events are recorded as service / diagnosis / production records.

[0094] Environment: Pasture rainfall / temperature sensors (daily values), grazing stress indicators (rotation grazing frequency).

[0095] Low-coverage imputation target accuracy ≥ 95%; event window validation for reproductive records; phenotypic observation confidence interval estimation: conception rate detection error ±2%. Output confidence metadata is transmitted with the sample.

[0096] 2. Functional annotation and feature construction:

[0097] Construct a "stability index" characterized by robustness (low performance variance); combine chromosome segment and disease site annotation.

[0098] 3. Explainable multi-trait prediction and G×E modeling:

[0099] Output the reproductive GEBV (with confidence interval), disease resistance score, and G×E susceptibility for each individual.

[0100] Results: Selection accuracy improved from 0.42 to 0.48 (a relative improvement of approximately 14.3%); the average GEBV interval width decreased by approximately 12%.

[0101] 4. Economic mapping and constraint optimization:

[0102] The scenario-based weighting is biased towards long-term reproductive benefits; the goal is to reduce the replacement rate. The output pairing scheme provides an expected annualized increase in profit per head of approximately 100 yuan.

[0103] 5. Paired simulation and long-term evaluation:

[0104] 10-year pedigree simulation: The implementation scheme shows that the cumulative rate of inbreeding is slowed down by about 20%, the reproductive stability of offspring is improved, and the replacement rate is reduced by 3 percentage points (e.g., from 18% to 15%).

[0105] 6. Closed-loop write-back:

[0106] The small farm uploads reproductive and disease events via mobile devices. The model is updated periodically, identifying four chromosome segments that significantly contribute to reproductive stability and prioritizing the avoidance of inbreeding in the following year.

[0107] Data Acquisition:

[0108] Genetic data: 50K SNP array or low-coverage sequencing (2×–4×), sequenced by a compliant laboratory and generated raw FASTQ, which has been called and quality controlled.

[0109] Phenotypic data: Automated weighing scales, RFID event logs, exported from ranch management systems, slaughterhouse inspection reports. Frequency: Monthly / bi-monthly weight reports; real-time recording of breeding events.

[0110] Environmental data: Temperature, humidity, and precipitation are sampled hourly / daily by sensors; feed formulations and batch records are manually uploaded.

[0111] Quality control: sample call rate, SNP missing rate, MAF threshold, phenotypic batch effect correction, uncertainty metadata (source, confidence level, missing mechanism).

[0112] A comprehensive comparison is shown in Table 1 below (conventional genome selection vs. the system used in this application).

[0113] Table 1

[0114] Metrics (Contextualized) Conventional genome selection (control) This application system (Example A / B) Notes Sample Genotyping 50K chips or equivalent 50K chips + 4× low-coverage references / 2× low-coverage (small field) More reference sequences + interpolation Phenotyping and Environmental Utilization Main phenotype, minor environmental factors Multi-time phenotypes + environmental sensing (hourly / daily) Support G×E Select Accuracy (Overall Index) 0.55 (A) / 0.42 (B) 0.62 (A) / 0.48 (B) A: Relative increase ≈ 12.7%; B: ≈ 14.3% GEBV confidence interval width Based on point estimation, no systematic transmission was achieved. Interval-based output reduces the average width by approximately 12–15%. Improve decision credibility Economic gain (expected per head) Baseline A: +150 yuan / head sold; B: +120 yuan / head / year Conservative estimate Cumulative inbreeding rate per generation Approximately 0.90% (A) Approximately 0.75% (A) A relative decrease of approximately 16.7%. Multigenerational genetic gain (5 years) 0.12 (standardized) 0.15 (standardized) A relative increase of 25% Replacement rate (small field) 18% 15% Replacement rate decreased by 3 percentage points (multi-environment robust strategy) Quantification of decision-making risks None / Limited GEBV probability distribution + risk perception score Quantifiable risks and scenario comparisons Model adaptability Typically offline periodic updates Closed-loop writeback + Bayesian / incremental update Dynamic correction based on practice Auditability and Explanatoryness Low (black box or point estimate) Effect attribution report (site / environment / uncertainty) It helps with compliance and scientific research validation.

[0115] As shown above, the system retains and transmits uncertainty information in each stage of data acquisition, quality control, feature construction, and prediction, outputting breeding values ​​in the form of confidence intervals / probability distributions. The decision-making process is based on probabilistic indicators rather than single-point estimates. This allows decision-makers to identify individuals and programs with high uncertainty, optimize mating within acceptable risk ranges, and significantly reduce erroneous selections and unexpected genetic consequences caused by measurement errors or data defects, thereby improving selection robustness and interpretability.

[0116] The foregoing has only described certain exemplary embodiments of the present invention by way of illustration. Undoubtedly, those skilled in the art can modify the described embodiments in various ways without departing from the spirit and scope of the present invention. Therefore, the foregoing drawings and descriptions are illustrative in nature and should not be construed as limiting the scope of protection of the claims of the present invention.

Claims

1. A genomic selection-assisted decision-making system for beef cattle breeding, characterized in that, include: The data acquisition and quality control module is used to acquire genomic data, preprocess the genomic data and calculate the confidence level to obtain integrated data with confidence level. The functional annotation and feature construction module is used to perform gene functional annotation on the integrated data with confidence, construct linkage fragments and temporal phenotypic features, and output an interpretable feature set; An interpretable multi-trait prediction and gene-environment interaction modeling module is used to output the genomic estimated breeding value and its confidence interval and G×E sensitivity score for each individual based on the interpretable feature set; The economic weight mapping and multi-objective selection optimization module is used to map the breeding value into a scenario-based economic value and generate selection priorities and candidate pairings under inbreeding and diversity constraints. The pairing simulation and long-term evaluation module is used to perform intergenerational simulation of the candidate pairings and output long-term genetic and economic trajectories. The interactive decision-making and closed-loop learning module is used to present decisions, receive implementation results, and write back actual post-model and health data for periodic model updates.

2. The beef cattle breeding genomic selection auxiliary decision-making system according to claim 1, characterized in that, The interpretable multi-trait prediction and gene-environment interaction modeling module employs an uncertainty propagation mechanism: using the feature set and its confidence level as input, it generates the confidence interval and probability distribution representation of the genomic estimated breeding value for each trait, and outputs a G×E sensitivity score and corresponding confidence level for each individual, where G represents genes and E represents the environment, for subsequent economic mapping and risk-aware selection.

3. The beef cattle breeding genomic selection auxiliary decision-making system according to claim 1, characterized in that, The interpretable multi-trait prediction and gene-environment interaction modeling module is used to generate effect attribution reports. The effect attribution reports include: gene loci or chromosome segments that significantly contribute to the target trait, key environmental factors and their interaction strengths, and describe the sources and uncertainties of attribution evidence in an auditable format to facilitate manual verification and compliance review.

4. The beef cattle breeding genomic selection auxiliary decision-making system according to claim 1, characterized in that, The economic weight mapping and multi-objective selection optimization module weights the GEBV of each trait based on scenario-based cost-benefit parameters and outputs candidate individual ranking and pairing suggestions under constraints such as inbreeding upper limit, preservation of population diversity and regulatory site avoidance. At the same time, it provides expected benefits and constraint satisfaction descriptions for each candidate scheme.

5. The beef cattle breeding genomic selection auxiliary decision-making system according to claim 1, characterized in that, The pairing simulation and long-term evaluation module performs pedigree-level Monte Carlo simulations based on the candidate pairings, simulating recombination, selection intensity, and genetic drift, and outputs time series and confidence intervals of genetic gain, population inbreeding coefficient, and economic returns over several generations to evaluate short-term and long-term trade-offs.

6. The beef cattle breeding genomic selection auxiliary decision-making system according to claim 1, characterized in that, After implementation, the interactive decision-making and closed-loop learning module receives actual breeding and production data, periodically adjusts the prediction model parameters and uncertainty estimates using Bayesian updates or incremental learning mechanisms, and records versioned model changes and model performance indicators to support traceable model evolution.

7. The beef cattle breeding genomic selection auxiliary decision-making system according to claim 1, characterized in that, The data quality control and uncertainty estimation module includes: site-level filtering and low-coverage sequencing interpolation for genotypes, batch effect correction and measurement error modeling for phenotypes, and generates confidence metadata for each input field. The confidence metadata is output along with the integrated data for use by subsequent modules.

8. The beef cattle breeding genomic selection auxiliary decision-making system according to claim 1, characterized in that, The genomic data includes: phenotype, genotype, environmental and management data; The preprocessing includes: performing quality control, imputing missing values, calculating the confidence level for each data item, and outputting integrated data with confidence levels.

Images