A Salt-Tolerant Maize Breeding Method Based on Root Function and Artificial Intelligence Prediction

Abstract

This invention discloses a method for breeding salt-tolerant maize based on root function and artificial intelligence prediction, belonging to the field of crop breeding. It includes constructing a core germplasm resource bank, conducting high-throughput root phenotypic identification under 150mM NaCl stress, screening superior root systems based on quantitative thresholds such as a root-to-shoot ratio increase of more than 20% and lateral root density exceeding the mean by 1.5 times the standard deviation, performing dynamic multi-omics analysis at seven time points (0h, 3h, 6h) after stress to identify early response markers within 3-12h after stress, creating core parents by combining molecular markers and root traits, integrating seedling phenotypic, omics, and genotypic data, training an AI model with a prediction accuracy R² of no less than 0.65, and using the model to eliminate the last 50% of predicted potential materials in the early stages of breeding to guide parent selection. This invention achieves a paradigm shift from empirical screening to precise prediction, significantly improving breeding efficiency and directionality.

Description

Technical Field

[0001] This invention belongs to the field of crop breeding technology, and in particular relates to a method for breeding salt-tolerant maize based on root function and artificial intelligence prediction. Background Technology

[0002] Salt stress is one of the major abiotic stresses limiting crop growth and yield globally. Developing salt-tolerant crop varieties is the most economical and effective strategy for developing and utilizing saline-alkali land and ensuring food security. As an important food and feed crop, the genetic improvement of maize's salt tolerance has always been a focus of attention.

[0003] Currently, maize salt-tolerant breeding mainly relies on several technical routes, such as "traditional breeding methods based on field phenotypic screening, assisted breeding methods based on molecular biotechnology, and technical exploration based on emerging omics and high-throughput phenotyping," but all of them have significant limitations.

[0004] Firstly, there is the traditional breeding method based on field phenotypic screening. This method involves directly planting breeding materials in saline-alkali land or artificially created salt-stressed environments, and screening based on agronomic traits such as survival rate, growth status, and final yield. However, this method has several drawbacks, such as a long cycle, requiring evaluation throughout the entire growing season or even multiple generations; high cost, with large-scale field trials consuming significant amounts of land and manpower; results are easily affected by environmental factors, with interannual differences in climate and soil salinity leading to poor phenotypic repeatability and low selection efficiency; and as a posterior selection method, it cannot predict the potential of materials in the early stages of breeding, resulting in significant resource waste.

[0005] Secondly, molecular biotechnology-based assisted breeding methods have become an important direction, with the development of molecular biology. This involves locating quantitative trait loci (QTLs), identifying salt tolerance-related genes through association analysis, and developing molecular markers for marker-assisted selection (MAS). For example, screening can be performed using markers associated with ion transporter genes (such as ZmSOS1 and ZmNHX). While these methods can improve the accuracy and efficiency of some selections, they also have significant limitations. Currently identified major QTLs or genes are mostly focused on single physiological processes (such as ion compartmentalization), while salt tolerance is a complex quantitative trait regulated by a multi-gene network. Existing markers are mostly developed based on genes stably expressed in the later stages of stress, lacking the ability to capture key early stress response events, resulting in insufficient predictability and universality. Furthermore, while precision technologies such as gene editing can create specific materials, the efficient identification and aggregation of multiple superior alleles still relies on large-scale phenotypic screening.

[0006] Finally, there is the exploration of technologies based on emerging omics and high-throughput phenotyping. In recent years, high-throughput sequencing technologies such as transcriptomics and metabolomics have been applied to elucidate plant salt tolerance mechanisms, while high-throughput phenotyping platforms (such as UAV remote sensing) are gradually being used for field trait collection. However, current research and applications are mostly isolated. Omics research is mostly static or single-time-point analysis, failing to reveal the key early dynamic response networks under salt stress. High-throughput phenotyping mainly focuses on aboveground canopy traits, and in-situ, non-destructive, dynamic monitoring technologies for the root system—the primary sensory and response organ to salt stress—are still immature and lack standardized procedures. More importantly, massive amounts of omics and phenotyping data have not yet been deeply integrated with breeding decisions, failing to construct effective models that can accurately predict complex agronomic traits from early measurable indicators, and the value of the data has not been fully realized.

[0007] In summary, existing technologies suffer from bottlenecks in efficiency, accuracy, and predictability. Therefore, there is an urgent need for a novel breeding method that can systematically integrate root dynamic function analysis, early molecular mechanism discovery under stress, and artificial intelligence predictive decision-making. This would fundamentally improve the efficiency and directionality of salt-tolerant maize breeding, achieving a paradigm shift from empirical screening to precision design. This proposed solution addresses this need. Summary of the Invention

[0008] The purpose of this invention is to address the bottlenecks in efficiency, accuracy, and predictability of existing seedling cultivation technologies, and to provide a salt-tolerant maize breeding method based on root function and artificial intelligence prediction. The core of this method lies in constructing a closed-loop breeding system integrating dynamic identification of root phenotypes, early molecular response analysis, and AI model prediction, thereby achieving a paradigm shift in seedling cultivation from experience-based screening to precise predictive design.

[0009] The present invention achieves the above objectives through the following technical solutions: A method for breeding salt-tolerant maize based on root function and artificial intelligence prediction includes the following steps: S1. Construct a core germplasm resource bank: Collect no less than 200 maize germplasm resources, including salt-tolerant basic materials and salt-sensitive control materials, and obtain their genomic background data; S2. High-throughput precise identification of root phenotypes: Under salt stress conditions of 150mM±10mMNaCl, the materials were cultured in the seedling stage, and the root phenotype imaging system was used to dynamically acquire root architecture, dynamic growth and functional indicators. S3. Dynamic multi-omics analysis: Salt-tolerant and salt-sensitive comparative materials were selected. Root tip tissue samples were collected at seven time points after salt stress: 0h, 3h, 6h, 12h, 24h, 48h and 7d. Temporal transcriptomics and metabolomics analyses were performed. Early omics marker data and genotype data were obtained. The focus was on analyzing early response gene modules and metabolic pathways within 3-12h after stress. S4. Creating core parental materials: Combining molecular marker-assisted selection and root phenotypic screening based on step S2, wherein the quantitative standard for root screening is: under the salt stress, materials with a root cap increase of more than 20% compared with the control group and a lateral root density higher than the population average by more than 1.5 standard deviations are identified as having excellent root traits; homozygous inbred lines are rapidly obtained through haploid breeding technology; S5. Construct an AI prediction model for salt tolerance: Using the seedling root phenotypic data obtained in step S2, the early omics marker data and genotype data obtained in step S3 as input features, and the field relative yield data of the corresponding material in saline-alkali land environment as prediction labels, train the machine learning model; the prediction coefficient of determination (R²) of the model for the relative yield of the material in saline-alkali land (with the yield of the control land as 100%) on the independent validation set is not less than 0.65. S6. Model-driven efficient breeding: Using the trained AI prediction model, the salt tolerance potential of breeding generation materials is predicted and screened at the seedling stage. Individuals with predicted potential in the bottom 50% are eliminated, guiding the selection of parental pairs and accelerating the breeding cycle of salt-tolerant maize varieties.

[0010] Furthermore, in step S1, the resource library should have no fewer than 200 copies to ensure sufficient genetic diversity, avoid genetic bottlenecks caused by a narrow material base, and significantly increase the probability of discovering rare and superior alleles from the source. This is the material basis for all subsequent innovations.

[0011] Furthermore, in step S2, the salt stress concentration is 150 mM ± 10 mM. This concentration has been verified by research to be the most sensitive and reproducible "golden concentration" for distinguishing differences in salt tolerance among maize materials. Below this range, the difference is not obvious; above this range, it may cause generalized damage and mask the true salt tolerance mechanism. Furthermore, in step S3, the focus is on analyzing 3-12 hours after stress, upgrading the traditional static snapshot of multi-omics to dynamic analysis that captures early signaling events. This time window is the core stage of salt stress signal transduction and initial response, and the biomarkers mined from this window are highly predictive, far exceeding the genes expressed during the stable stress period.

[0012] Furthermore, in step S4, the root-to-shoot ratio increases by 20% and the lateral root density exceeds the mean by 1.5 standard deviations. This fuzzy concept of superior root system is quantified into a precise, industrially-grade objective standard. This threshold can efficiently identify individuals with genuine physiological advantages, greatly enhancing positive selection pressure and breeding efficiency.

[0013] Furthermore, in step S5, the model prediction determination coefficient R² ≥ 0.65 sets a hard technical indicator for model effectiveness. Achieving this accuracy means that the model can reliably replace most mid-term field phenotypic identification, shortening the breeding cycle from several years to several months, saving a lot of land and labor costs, and is the key to the transformation of the breeding paradigm.

[0014] Furthermore, in step S5, the elimination of the remaining 50% of materials, combined with model accuracy, forms a high-efficiency decision-making mechanism with controllable risk. While ensuring no optimal materials are overlooked, this systematically and on a large scale reduces the cost of later-stage experiments, making high-throughput breeding economically feasible.

[0015] Furthermore, in step S1, the ratio of the salt-tolerant base material to the salt-sensitive control material is controlled between 1:1 and 3:1 to ensure the effectiveness of the genetic analysis.

[0016] Furthermore, in step S2, the root phenotype identification is continuously and dynamically monitored from day 7 to day 21 of the seedling stage, and the calculation of the root growth rate change rate is based on the data from day 3 to day 7 after stress.

[0017] Furthermore, in step S3, the early omics marker data includes: expression data of genes whose expression levels are upregulated by more than 2-fold within 6 hours after stress in salt-tolerant materials and continue to be upregulated for up to 24 hours.

[0018] Furthermore, in step S4, the molecular marker-assisted selection specifically involves simultaneous screening using at least three functional KASP markers associated with the salt tolerance major genes ZmSOS1 and / or ZmNHX4.

[0019] Furthermore, in step S5, the input feature dimension of the AI ​​prediction model is controlled to be between 15 and 30 after feature selection, including 5 to 10 key root phenotypic indicators, 3 to 5 early response gene expression level indicators, and 2 to 4 key metabolite indicators.

[0020] Furthermore, in step S6, the elimination of individuals with the predicted potential in the bottom 50% specifically means that the screening is completed at 20 days of the seedling stage, thereby reducing the size of the population entering the saline-alkali land test by 40%-60%.

[0021] A system for implementing salt-tolerant maize breeding methods, comprising: The germplasm resource management module is used to store and manage the genetic information and phenotypic data of the core germplasm resource bank; The high-throughput root phenotyping platform includes hardware devices for seedling salt stress culture and root dynamic imaging, as well as software modules for image analysis and phenotypic index extraction. The multi-omics data analysis module has a built-in dedicated algorithm for identifying genes whose expression levels change significantly within 3-12 hours after stress. The artificial intelligence prediction module integrates a machine learning model with a prediction determination coefficient (R²) of not less than 0.65 as described in claim 1; The breeding decision support module is used to automatically output parent selection recommendations and screening lists based on prediction results.

[0022] Furthermore, it also includes a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, is capable of implementing steps S5 and S6 in the salt-tolerant maize breeding method as described in any one of claims 1 to 7.

[0023] Beneficial effects: This invention is reasonably designed and has the following beneficial effects: 1. In this invention, the systemic function of the root system—the primary sensory and responsive organ to salt stress—is used as the core selection indicator, overcoming the lag of traditional methods that only focus on the final phenotype of the aboveground parts. By setting quantitative thresholds such as a 20% increase in the root-to-shoot ratio and a lateral root density exceeding the mean by 1.5 standard deviations, objective and efficient screening of key physiological traits for salt tolerance is achieved, thereby improving the salt tolerance foundation of materials from the source. 2. In this invention, by setting a 150mM NaCl stress concentration and an early time-series analysis window of 3-12 hours, the method can accurately capture key molecular events in salt stress signal transduction and initial response. The resulting early response markers, which rapidly upregulated by 2-fold within 6 hours, exhibit stronger predictive power and breeding application value compared to genes stably expressed in the later stages of stress, achieving dynamic and in-depth analysis of salt tolerance mechanisms and prospective marker discovery. 3. In this invention, an AI prediction model is constructed by integrating multi-dimensional early data, and a hard target of R² ≥ 0.65 is set for its prediction accuracy, enabling high-confidence prediction of the final yield potential of materials at the seedling stage. The early screening strategy implemented accordingly, which eliminates the remaining 50% of materials, can systematically reduce the scale of costly field testing by 40%-60%. This marks a fundamental leap from extensive, low-yield experience-based screening to precision-guided predictive breeding, shortening the breeding cycle by more than 40% and significantly saving land, labor, and time costs. 4. In this invention, the parameters of each step are defined numerically and in a standardized manner, such as the size of the resource pool, stress concentration, time point, screening threshold, and model accuracy. This transforms the breeding process, which originally relied on personal experience, into a stable, controllable, and repeatable industrial production line, greatly improving the operability and scalability of the technical solution. Attached Figure Description

[0024] Figure 1 This is a general technical roadmap for the present invention; Figure 2 This is a roadmap of the first-stage branch technology of the present invention; Figure 3 This is a branch technology roadmap for the second stage of this invention; Figure 4 This is a roadmap for the third stage of the technology of this invention. Detailed Implementation

[0025] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0026] In the description of this invention, it should be understood that the terms "longitudinal", "lateral", "length", "width", "thickness", "upper", "lower", "front", "rear", "left", "right", "vertical", "horizontal", "top", "bottom", "inner", "outer", etc., indicate the orientation or positional relationship based on the orientation or positional relationship shown in the accompanying drawings. They are only for the convenience of describing this invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, they should not be construed as limitations on this invention.

[0027] Example 1 A method for breeding salt-tolerant maize based on root function and artificial intelligence prediction includes the following steps: S1. Construct a core germplasm resource bank: Collect no less than 200 maize germplasm resources, including salt-tolerant basic materials and salt-sensitive control materials, and obtain their genomic background data; S2. High-throughput precise identification of root phenotypes: Under salt stress conditions of 150mM±10mMNaCl, the materials were cultured in the seedling stage, and the root phenotype imaging system was used to dynamically acquire root architecture, dynamic growth and functional indicators. S3. Dynamic multi-omics analysis: Salt-tolerant and salt-sensitive comparative materials were selected. Root tip tissue samples were collected at seven time points after salt stress: 0h, 3h, 6h, 12h, 24h, 48h and 7d. Temporal transcriptomics and metabolomics analyses were performed. Early omics marker data and genotype data were obtained. The focus was on analyzing early response gene modules and metabolic pathways within 3-12h after stress. S4. Creating core parental materials: Combining molecular marker-assisted selection and root phenotypic screening based on step S2, the quantitative criteria for root screening are: under salt stress, materials with a root cap increase of more than 20% compared to the control group and a lateral root density more than 1.5 standard deviations higher than the population average are identified as having excellent root traits; homozygous inbred lines are rapidly obtained through haploid breeding technology; S5. Construct an AI prediction model for salt tolerance: Using the seedling root phenotypic data obtained in step S2, the early omics marker data and genotype data obtained in step S3 as input features, and the field relative yield data of the corresponding material in saline-alkali land environment as prediction labels, train the machine learning model; the prediction coefficient of determination (R²) of the model on the independent validation set for the relative yield of the material in saline-alkali land (with the yield of the control land as 100%) is not less than 0.65. S6. Model-driven high-efficiency breeding: Using trained AI prediction models, the salt tolerance potential of breeding generation materials is predicted and screened during the seedling stage. Individuals with predicted potential in the bottom 50% are eliminated, guiding the selection of parental pairs and accelerating the breeding cycle of salt-tolerant maize varieties.

[0028] In step S1, the resource library should have no fewer than 200 copies to ensure sufficient genetic diversity, avoid genetic bottlenecks caused by a narrow material base, and significantly increase the probability of discovering rare and superior alleles from the source. This is the material basis for all subsequent innovations.

[0029] In step S2, the salt stress concentration is 150 mM ± 10 mM. This concentration has been verified by research to be the most sensitive and reproducible "golden concentration" for distinguishing differences in salt tolerance among maize materials. Below this range, the differences are not significant; above this range, it may cause generalized damage and mask the true salt tolerance mechanism. In step S3, the focus is on analyzing 3-12 hours after stress, upgrading the traditional static snapshot of multi-omics to dynamic analysis that captures early signaling events. This time window is the core stage of salt stress signal transduction and initial response, and the biomarkers mined from this window are highly predictive, far exceeding the genes expressed during the stable stress period.

[0030] In step S4, the root-to-shoot ratio increases by 20% and the lateral root density exceeds the mean by 1.5 standard deviations. This quantifies the vague concept of superior root system into a precise and industrially screenable objective standard. This threshold can efficiently identify individuals with real physiological advantages, greatly improving positive selection pressure and breeding efficiency.

[0031] In step S5, the model prediction determination coefficient R² ≥ 0.65 sets a hard technical indicator for model effectiveness. Achieving this accuracy means that the model can reliably replace most mid-term field phenotypic identification, shortening the breeding cycle from several years to several months, saving a lot of land and labor costs, and is the key to the transformation of the breeding paradigm.

[0032] In step S5, the elimination of the remaining 50% of materials, combined with model accuracy, forms a high-efficiency decision-making mechanism with controllable risk. While ensuring no optimal materials are overlooked, this systematically and on a large scale reduces the cost of later-stage experiments, making high-throughput breeding economically feasible.

[0033] In step S1, the ratio of salt-tolerant base material to salt-sensitive control material is controlled between 1:1 and 3:1 to ensure the effectiveness of genetic analysis.

[0034] In step S2, root phenotypic identification was carried out under continuous dynamic monitoring from day 7 to day 21 of the seedling stage, and the calculation of the root growth rate change rate was based on the data from day 3 to day 7 after stress.

[0035] In step S3, the early omics marker data includes: expression data of genes whose expression levels are upregulated more than 2-fold within 6 hours after stress and persist for up to 24 hours in salt-tolerant materials.

[0036] In step S4, the molecular marker-assisted selection specifically involves simultaneous screening using at least three functional KASP markers associated with the salt tolerance major genes ZmSOS1 and / or ZmNHX4.

[0037] In step S5, the input feature dimension of the AI ​​prediction model is controlled between 15 and 30 after feature selection, including 5 to 10 key root phenotypic indicators, 3 to 5 early response gene expression level indicators, and 2 to 4 key metabolite indicators.

[0038] In step S6, individuals with predicted potential in the bottom 50% are eliminated. Specifically, screening is completed 20 days after the seedling stage, reducing the population size entering the saline-alkali land test by 40%-60%.

[0039] A system for implementing salt-tolerant maize breeding methods, comprising: The germplasm resource management module is used to store and manage the genetic information and phenotypic data of the core germplasm resource bank; The high-throughput root phenotyping platform includes hardware devices for seedling salt stress culture and root dynamic imaging, as well as software modules for image analysis and phenotypic index extraction. The multi-omics data analysis module has a built-in dedicated algorithm for identifying genes whose expression levels change significantly within 3-12 hours after stress. The artificial intelligence prediction module integrates a machine learning model with a prediction determination coefficient (R²) of not less than 0.65 as claimed in claim 1. The breeding decision support module is used to automatically output parent selection recommendations and screening lists based on prediction results.

[0040] It also includes a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, enables the implementation of steps S5 and S6 in the salt-tolerant maize breeding method as described in any one of claims 1 to 7.

[0041] Example 2 1. Experimental Materials and Preliminary Preparations Plant materials: A total of 220 maize inbred lines were collected and created. Among them, 100 were known salt-tolerant germplasm, such as local varieties from saline-alkali land and reported salt-tolerant materials; 100 were commonly used but salt-sensitive backbone inbred lines, such as B73, Mo17, and Zheng58; and the remaining 20 were control materials with known significant differences in salt tolerance, such as the extremely salt-tolerant material C1 and the extremely sensitive material S1, used for system calibration.

[0042] Salt stress solution: Prepare a 150 mM salt solution using analytical grade sodium chloride (NaCl), using deionized water, and adjust the pH to be consistent with the normal Hogland nutrient solution (approximately 6.0) using a pH meter.

[0043] Culture system: A customized high-throughput root phenotyping system is used. The system consists of a transparent acrylic root box (internal dimensions: 30cm high × 20cm wide × 1.5cm thick), an automated nutrient solution circulation and replacement module, and a high-resolution scanner fixed on a track.

[0044] Main reagents and instruments: RNA extraction kit, reverse transcription kit, high-throughput sequencing service (for transcriptomics), liquid chromatography-mass spectrometry (LC-MS, for non-targeted metabolomics), KASP genotyping kit, and computer workstation for AI modeling (configured with at least an Intel i7 processor and 32GB of memory).

[0045] 2. Core Experimental Procedure Phase 1: Precise identification of root phenotypes and dynamic omics analysis (taking tolerance / sensitivity control materials C1 and S1 as examples) Step 1: Seedling salt stress and root dynamic imaging After disinfecting and germinating C1 and S1 seeds, they were sown in root boxes containing standard quartz sand substrate. Each material was replicated in 6 boxes and placed in an artificial climate chamber (28℃ / 22℃, day / night, 16 hours of light). When the seedlings reached the three-leaf stage (approximately 10 days after sowing), treatment began. Three boxes were replaced with a nutrient solution containing 150 mM NaCl and labeled as the stress group, while the other three boxes continued to use the normal nutrient solution and were labeled as the control group. After the process begins, each root box is automatically scanned at the same time every day for 7 consecutive days, and the images are automatically uploaded to the server.

[0046] Step 2: Root image analysis and phenotypic extraction Images were processed in batches using image analysis software developed in Python (integrating vision and image processing libraries), automatically separating the root system from the background and calculating the following key phenotypic parameters: total root length (TRL), root surface area (RSA), average root diameter (AvgDia), root-to-shoot ratio (RSR, calculated from simultaneously captured aboveground images), and root growth rate (RGR) calculated based on images from two consecutive days. Output results: On day 4 of treatment, the RGR of salt-resistant material C1 decreased by about 15% under stress, while the RGR of sensitive material S1 decreased by 60%.

[0047] Step 3: Time-series sample acquisition and multi-omics analysis In another batch of parallel experiments, samples were taken from the root tips of C1 and S1 at 0h, 3h, 6h, 12h and 24h after salt stress treatment, respectively, with 3 biological replicates per material at each time point, and the samples were flash-frozen in liquid nitrogen. Transcriptome analysis: Total RNA was extracted, and after passing quality control, a cDNA library was constructed and sequenced. Bioinformatics analysis focused on screening genes that were significantly upregulated in C1 within 3-6 hours (log2FC>1, i.e., a 2-fold change), while showing no significant change or downregulation in S1. Metabolomics analysis: After grinding and extracting metabolites, the samples were analyzed by LC-MS. Principal component analysis (PCA) and orthogonal partial least squares discriminant analysis (OPLS-DA) were used to identify differentially accumulated metabolites, such as proline and γ-aminobutyric acid, in C1 root tips within 12 hours. Key outputs: A transcription factor gene (tentatively named ZnERF1) was identified, whose expression level was upregulated 3.5-fold in C1 at 6 h after stress, and its expression level in the root tip was found to be significantly positively correlated with the previously measured lateral root density (r=0.82). This gene and its associated metabolites (such as succinic acid) were identified as candidate early predictive markers.

[0048] Phase Two: Building and Validating AI Prediction Models Step 1: Construct the training dataset Using the phenotyping and omics platform established in the first phase, all materials in the 220 core germplasm resource bank were subjected to a uniform 7-day treatment with 150mM NaCl at the seedling stage (starting from 10 days of seedling age). Ten core root phenotypic data were obtained for each material, such as total root length, root-shoot ratio, and RGR decline rate on day 7. The expression level of ZnERF1 gene (qPCR validation) and the two key metabolites (LC-MS targeted detection) were measured in all materials. The above 220 materials were planted in a moderately saline-alkali experimental field (soil electrical conductivity ECe=4.0dS / m). After maturity, the relative yield of each material in saline-alkali land was measured, that is, the ratio of the yield per plant in saline-alkali land to the yield per plant in the control normal plot.

[0049] Step 2: Model Training and Optimization The dataset was randomly divided into a training set (176 sets, accounting for 80%) and an independent validation set (44 sets, accounting for 20%).

[0050] Using root phenotypic data, ZnERF1 expression level, metabolite content, and existing KASP marker data of three ZmSOS1 genotypes as initial features, and relative yield of saline-alkali land as the target variable, a random forest regression model was trained. By adjusting the model hyperparameters, such as the number of decision trees and maximum depth, through grid search and cross-validation, the final model achieved a coefficient of determination R² of 0.68 for predicting relative yield on the independent validation set, with a root mean square error (RMSE) of 8.5%, meeting the preset technical requirement of R² ≥ 0.65.

[0051] Phase 3: Model-driven applications of efficient breeding Step 1: Early Screening Create a new breeding population containing 500 F2 individual plants; At 20 days of seedling stage, only a non-destructive root scan of each individual plant is needed to obtain 5 key phenotypes and leaf samples for qPCR detection of ZnERF1 expression levels.

[0052] The acquired data is input into the trained AI prediction model, and the model immediately outputs the predicted relative yield value for each individual plant. Based on the predicted values, individuals were sorted from low to high, and the bottom 50% (approximately 250 plants) were eliminated. The remaining plants were predicted to be high-potential individuals and transplanted to saline-alkali land for subsequent field testing.

[0053] Step 2: Validation of Results After one growing season, compare the actual performance of the eliminated plants and the retained plants in the field. Data projections: In the retained population, the proportion of individuals with actual relative yields higher than the population average is expected to exceed 65%; while in the eliminated population, this proportion is expected to be less than 20%, demonstrating that the model screening significantly enriches superior individuals.

[0054] Efficiency assessment: In this breeding cycle, 50% of the materials were eliminated early, saving about 50% of the field trial land, management and identification costs. Furthermore, the investment of superior resources was concentrated on high-potential materials, which accelerated the purification and stabilization process of superior lines.

[0055] 3. Experiment Summary and Extension The above embodiments demonstrate the complete closed loop of this invention, from precise phenotypic identification and mechanism discovery to intelligent application. By setting numerical thresholds such as 150mM NaCl stress, early sampling at 3-12h, and a 20% increase in root-to-shoot ratio, and by achieving an accuracy of R²≥0.65 in the final prediction model, this scheme successfully transforms the complex salt-tolerant breeding process into a standardized and predictable industrial process.

[0056] Those skilled in the art will understand that various modifications and alterations can be made to the above embodiments without departing from the spirit and scope of the invention. For example, the ZnERF1 gene can be replaced with other equivalent early response genes identified by this method; the machine learning model can also employ XGBoost or lightweight neural networks, etc. All such modifications and alterations fall within the scope of protection of the invention as defined by the appended claims.

[0057] It will be apparent to those skilled in the art that the present invention is not limited to the details of the exemplary embodiments described above, and that the invention can be implemented in other specific forms without departing from its spirit or essential characteristics. Therefore, the embodiments should be considered in all respects as exemplary and non-limiting, and the scope of the invention is defined by the appended claims rather than the foregoing description. Thus, all variations falling within the meaning and scope of equivalents of the claims are intended to be included within the present invention. No reference numerals in the claims should be construed as limiting the scope of the claims.

[0058] Furthermore, it should be understood that although this specification describes embodiments, not every embodiment contains only one independent technical solution. This narrative style is merely for clarity. Those skilled in the art should consider the specification as a whole, and the technical solutions in each embodiment can also be appropriately combined to form other embodiments that can be understood by those skilled in the art.

Claims

1. A method for breeding salt-tolerant maize based on root function and artificial intelligence prediction, characterized in that, Includes the following steps: S1. Construct a core germplasm resource bank: Collect no less than 200 maize germplasm resources, including salt-tolerant basic materials and salt-sensitive control materials, and obtain their genomic background data; S2. High-throughput precise identification of root phenotypes: Under salt stress conditions of 150mM±10mMNaCl, the materials were cultured in the seedling stage, and the root phenotype imaging system was used to dynamically acquire root architecture, dynamic growth and functional indicators. S3. Dynamic multi-omics analysis: Salt-tolerant and salt-sensitive comparative materials were selected. Root tip tissue samples were collected at seven time points after salt stress: 0h, 3h, 6h, 12h, 24h, 48h and 7d. Temporal transcriptomics and metabolomics analyses were performed. Early omics marker data and genotype data were obtained. The focus was on analyzing early response gene modules and metabolic pathways within 3-12h after stress. S4. Creating core parental materials: Combining molecular marker-assisted selection and root phenotypic screening based on step S2, wherein the quantitative standard for root screening is: under the salt stress, materials with a root cap increase of more than 20% compared with the control group and a lateral root density higher than the population average by more than 1.5 standard deviations are identified as having excellent root traits; homozygous inbred lines are rapidly obtained through haploid breeding technology; S5. Construct an AI prediction model for salt tolerance: Using the seedling root phenotypic data obtained in step S2, the early omics marker data and genotype data obtained in step S3 as input features, and the field relative yield data of the corresponding material in saline-alkali land environment as prediction labels, train the machine learning model; the prediction coefficient of determination (R²) of the model for the relative yield of the material in saline-alkali land (with the yield of the control land as 100%) on the independent validation set is not less than 0.

65. S6. Model-driven efficient breeding: Using the trained AI prediction model, the salt tolerance potential of breeding generation materials is predicted and screened at the seedling stage. Individuals with predicted potential in the bottom 50% are eliminated, guiding the selection of parental pairs and accelerating the breeding cycle of salt-tolerant maize varieties.

2. The method for breeding salt-tolerant maize according to claim 1, characterized in that: In step S1, the ratio of the salt-tolerant base material to the salt-sensitive control material is controlled between 1:1 and 3:1 to ensure the effectiveness of the genetic analysis.

3. The method for breeding salt-tolerant maize according to claim 1, characterized in that: In step S2, the root phenotype identification is continuously and dynamically monitored from day 7 to day 21 of the seedling stage, and the calculation of the root growth rate change rate is based on the data from day 3 to day 7 after stress.

4. The method for breeding salt-tolerant maize according to claim 1, characterized in that: In step S3, the early omics marker data includes: expression data of genes whose expression levels are upregulated by more than 2-fold within 6 hours after stress and continue to be upregulated for up to 24 hours in salt-tolerant materials.

5. The method for breeding salt-tolerant maize according to claim 1, characterized in that: In step S4, the molecular marker-assisted selection specifically involves simultaneous screening using at least three functional KASP markers associated with the salt tolerance major genes ZmSOS1 and / or ZmNHX4.

6. The method for breeding salt-tolerant maize according to claim 1, characterized in that: In step S5, the input feature dimension of the AI ​​prediction model is controlled between 15 and 30 after feature selection, including 5 to 10 key root phenotypic indicators, 3 to 5 early response gene expression level indicators, and 2 to 4 key metabolite indicators.

7. The method for breeding salt-tolerant maize according to claim 1, characterized in that: In step S6, the elimination of individuals with the predicted potential in the bottom 50% specifically means that the screening is completed at 20 days of the seedling stage, thereby reducing the size of the population entering the saline-alkali land test by 40%-60%.

8. A system for implementing the salt-tolerant maize breeding method according to any one of claims 1 to 7, characterized in that, include: The germplasm resource management module is used to store and manage the genetic information and phenotypic data of the core germplasm resource bank; The high-throughput root phenotyping platform includes hardware devices for seedling salt stress culture and root dynamic imaging, as well as software modules for image analysis and phenotypic index extraction. The multi-omics data analysis module has a built-in dedicated algorithm for identifying genes whose expression levels change significantly within 3-12 hours after stress. The artificial intelligence prediction module integrates a machine learning model with a prediction determination coefficient (R²) of not less than 0.65 as described in claim 1; The breeding decision support module is used to automatically output parent selection recommendations and screening lists based on prediction results.

9. The system of the salt-tolerant maize breeding method according to claim 8, characterized in that: It also includes a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, enables the implementation of steps S5 and S6 in the salt-tolerant maize breeding method as described in any one of claims 1 to 7.

Images