Paddy rice drought resistance phenotype detection and genetic analysis method based on phenotype robot

Abstract

The invention discloses a rice drought resistance phenotype detection and genetic analysis method based on a phenotypic robot, which is characterized in that rice hyperspectral phenotypic character data is acquired based on the phenotypic robot, and the relative water content RWC of rice is predicted. Through high-frequency data acquisition, RWC (t) based on a time sequence is obtained, RWC (t) curve change characteristic indexes are calculated, the dynamic change process of a rice experiment material in the whole drought experiment period is quantified, the drought resistance index DTI of the rice experiment material is obtained, and the drought resistance of the rice variety is comprehensively evaluated. DTI as a user-defined phenotypic character can be used for whole genome association analysis (GWAS), and a key gene for controlling the drought resistance of rice is excavated. According to the invention, the problems of low efficiency and low flux of traditional rice phenotype detection are solved, phenotype-gene mining based on the mobile robot is realized, and a new way is expanded for improvement of rice genetic breeding.

Description

Technical Field

[0001] This invention belongs to the fields of crop genetics and breeding, artificial intelligence in production, and agricultural technology promotion, and specifically relates to a method for detecting and analyzing the drought resistance phenotype of rice based on phenotypic robots. Background Technology

[0002] Rice growth is extremely sensitive to water conditions, making the breeding of drought-resistant rice varieties a crucial task in breeding science. However, traditional plant phenotyping methods typically require significant manpower and resources, and the data lacks standardization, hindering objective evaluation. The drought resistance of different rice genotypes is highly correlated with metabolites, but metabolic detection methods (such as GC / LC-MS and NMR) are complex, destructive, and costly. Therefore, non-destructive, high-throughput methods for detecting rice drought resistance phenotypes are essential. Hyperspectral imaging technology is widely used due to its non-destructive and high-throughput properties in plant phenotyping. Using hyperspectral technology, leaf reflectance information and spatial image information can be rapidly collected under non-destructive conditions, enabling high-throughput detection of plant physiological conditions through modeling. Furthermore, combined with GWAS analysis, traits related to spectral reflectance can be explained at the gene level, identifying SNP loci associated with the phenotype.

[0003] Previous studies on hyperspectral and metabolic prediction have often relied on fixed or regionally restricted platforms, which suffer from low throughput and limitations due to space or cost constraints. Currently, hyperspectral acquisition technology based on mobile robotic platforms offers an innovative solution to overcome the limitations of traditional methods, enabling more efficient, higher-throughput, and wider-range high-precision plant phenotypic acquisition under field conditions. By combining these technologies, we can uncover crucial information on drought resistance-related spectral phenotypes, metabolites, and genes in rice, thus advancing the development of drought-resistant rice breeding. Summary of the Invention

[0004] (a) Technical problems to be solved The technical problems to be solved by this invention include the following aspects. Existing rice phenotypic detection methods mainly rely on manual operation, which not only has low detection efficiency and limited data throughput, but also lacks unified standards and is easily affected by human factors, resulting in unstable and unrepeatable results. Furthermore, existing rice metabolite detection methods typically rely on complex sample pretreatment and expensive detection equipment, and most methods are destructive, hindering continuous and dynamic monitoring of samples, resulting in high detection costs and limited application scope. More importantly, existing technologies generally lack the ability to continuously observe the dynamic changes of rice throughout its entire growth cycle. Traditional phenotypic and metabolite detection methods can usually only collect data at a limited number of time points within the rice growth cycle, making it difficult to achieve high-frequency, non-destructive tracking of dynamic changes in phenotypic and metabolite changes during rice growth, thus limiting in-depth research and precise regulation of rice growth mechanisms. Therefore, there is an urgent need for an efficient, standardized, non-destructive method for detecting rice phenotypic and metabolites with continuous dynamic monitoring capabilities.

[0005] (II) Technical Solution To address the aforementioned problems, this invention provides the following technical solution: a method for detecting and analyzing the drought-resistant phenotype of rice based on a phenotypic robot, the specific solution of which is as follows.

[0006] A method for detecting and analyzing the drought resistance phenotype of rice based on phenotypic robots includes the following steps: Step A: Select some varieties from the core germplasm resource bank of rice to conduct a drought experiment on rice populations. The following operations were performed on each variety material for different drought days: (1) Use a phenotypic robot equipped with a hyperspectral camera to collect hyperspectral reflectance information of each variety material and process it to obtain hyperspectral phenotypic trait data; (2) Manually take leaf samples of each variety material and measure the contents of glucose, galactose and fructose; (3) Manually measure the fresh weight FW, fresh weight after soaking FWAS and dry weight DW of each variety material and calculate the relative water content RWC. Step B: Based on the hyperspectral phenotypic data of each variety and the corresponding number of drought days, a drought day prediction model is constructed; based on the hyperspectral phenotypic data of each variety and the contents of glucose, fructose and galactose, glucose content prediction model, fructose content prediction model and galactose content prediction model are constructed respectively. Step C: Combining the four indicators of drought days, glucose content, galactose content, and fructose content from Step A, regression analysis is performed on the relative water content (RWC) obtained in Step A to construct an accurate prediction model for the relative water content (RWC) of rice. Step D: Using a phenotypic robot equipped with a hyperspectral camera, high-frequency hyperspectral data is automatically collected from field crops to obtain hyperspectral data of field crops at different times. This data is then input into multiple prediction models constructed in Step B to obtain the predicted number of drought days, glucose content, galactose content, and fructose content. This data is then input into the relative water content (RWC) accurate prediction model constructed in Step C to obtain the predicted relative water content (RWC). Based on the results at different times, the time-series relative water content (RWC(t)) of the experimental materials is obtained. Step E: Using the relative water content RWC(t) based on the time series obtained in Step D, calculate the characteristic index of the RWC(t) curve change, quantify the dynamic change process of each rice variety during the entire drought experiment period, and obtain the rice drought resistance performance index DTI, which is used to comprehensively evaluate the drought resistance of rice varieties. Step F involves using the rice drought resistance index (DTI) obtained in Step E as a custom phenotypic trait, and combining it with the whole genome sequence of the rice varieties mentioned in Step A to perform a genome-wide association analysis (GWAS) to identify significantly related gene loci and mine rice drought resistance-related genes.

[0007] Preferably, the method for obtaining hyperspectral phenotypic data of each variety of material in step A is as follows: the mounted hyperspectral camera continuously scans along the direction of travel of the phenotyping robot, and collects the hyperspectral reflectance information of the entire planting unit of plants at one time. The data is stored in binary data stream mode, and the total reflectance T and average reflectance A of the plant leaves under each band are obtained through image segmentation, image masking and data processing, as well as their derived first derivative, second derivative and logarithm.

[0008] Preferably, the method for constructing the drought day prediction model in step B is as follows: the hyperspectral phenotypic data collected for different drought days are divided into several classes. A random dataset partitioning method is adopted. First, a competitive adaptive reweighted sampling algorithm is used to perform preliminary feature screening on the massive hyperspectral phenotypic traits to reduce the number of features. Then, the data is input into a random forest classifier to construct a classification model and output the feature importance coefficients. Features with high importance coefficients are selected from these models. Finally, the data using these features are used to build a model through the random forest classifier.

[0009] Preferably, the specific method for constructing the glucose content prediction model, fructose content prediction model, and galactose content prediction model in step B is to use different data preprocessing, feature selection, data partitioning, and regression methods in combination, select the method with a high coefficient of determination and a small number of features, and output the optimal metabolite prediction model.

[0010] Preferably, the linear regression formula for predicting relative water content (RWC) in step C using the number of drought days, glucose, fructose, and galactose is: ; Among them, DryDays represents the number of dry days; Glucose represents the glucose content; Galactose represents the fructose content; and Fructose represents the galactose content.

[0011] Preferably, the rice drought resistance index (DTI) in step E is calculated using the dynamic change curve based on the time series RWC(t) described in step D, to measure the material's moisture retention and recovery ability. The formula is: ; Among them, AUC dry The area under the relative water content curve during the dry period; T d Duration of drought; RWC final The relative water content that is stable after rehydration; RWC min The relative water content at the point of most severe drought stress; T r The time from the start of rehydration to when the moisture content tends to stabilize.

[0012] Preferably, the GWAS process in step F specifically involves: performing genotype filling preprocessing on the sequencing data, selecting SNPs with a deletion rate of less than 20% for subsequent analysis, using a mixed linear model in the GWAS analysis process, setting a threshold, and screening out significant SNP sites that exceed the threshold after Boltzmann test.

[0013] (III) Beneficial Effects Compared with the prior art, the present invention has at least the following positive technical effects.

[0014] Existing methods for metabolite detection often rely on destructive sampling and complex metabolic detection techniques such as gas chromatography / liquid chromatography (GC / LC), gas chromatography / liquid chromatography-mass spectrometry (GC / LC-MS), and nuclear magnetic resonance (NMR). These methods not only may damage plants but also suffer from high time and labor costs and operational complexity. While there are numerous studies on metabolite modeling and prediction using hyperspectral imaging, research specifically on drought-related metabolites in rice is limited. This invention constructs a metabolite regression model based on top-view 900-1700 nm hyperspectral data collected by a mobile robot, combined with leaf metabolic data collected during the experiment. The results show that by using different combinations of data preprocessing, feature selection, data partitioning, and regression methods, the optimal prediction models for glucose, fructose, and galactose were found to utilize only 0.89%, 2.12%, and 2.34% of all 896 spectral indices, respectively, achieving R² determinations of 0.7676, 0.7956, and 0.8535 on the test set.

[0015] Traditional methods for determining the relative water content of rice require harvesting and weighing the fresh weight (FW), fresh weight after soaking (FWAS), and dry weight (DW) of the above-ground parts of the rice plant. This process is cumbersome, time-consuming, labor-intensive, and directly damages the plant, making continuous measurement impossible. This invention, however, directly determines the relative water content of the plant by non-destructively collecting hyperspectral data from the plant using a robot. Preliminary models predicting the number of drought days, glucose, fructose, and galactose content are established, and these four indicators are predicted using the collected spectral data. Regression analysis is then performed on the relative water content of the plant based on these four indicators. Ultimately, the predicted relative water content (RWC)² compared to the manually measured value reached 0.69 on the test set. This value is higher than the results obtained using any one of the four indicators alone, indicating that the combined indicators better reflect the drought severity of the plant.

[0016] Traditional phenotypic and metabolite detection methods suffer from low efficiency, low throughput, and poor convenience. The time required for a single, complete data collection is excessive, and the experimental environment may have already undergone significant changes. Furthermore, data collection is mostly limited to a few specific time points during rice growth, failing to accurately capture dynamic changes in rice growth processes, such as diurnal rhythm variations or physiological changes in response to subtle temperature differences. This paper utilizes a phenotypic robot equipped with a hyperspectral camera to automatically collect high-frequency hyperspectral data from field crops. This data, obtained at different times, yields relative water content (RWC(t)) based on time series data. The characteristic index of RWC(t) curve changes is calculated, and a comprehensive drought resistance index (DTI) is derived. This index quantifies the dynamic changes of various rice varieties throughout the entire drought experimental period, providing a more comprehensive and accurate assessment of the drought resistance performance of rice varieties. Attached Figure Description

[0017] Figure 1 This is a flowchart of the method of the present invention.

[0018] Figure 2 RWC modeling and prediction results for various dataset partitioning methods and regression method combinations.

[0019] Figure 3 The curve is the relative water content RWC(t) based on the time series.

[0020] Figure 4 This is a schematic diagram of colocalized chromosome sites. Detailed Implementation

[0021] The present invention will be further described below with reference to the accompanying drawings and implementation examples.

[0022] This invention discloses a method for detecting and analyzing the drought resistance phenotype of rice based on phenotypic robots, such as... Figure 1The flowchart shown is a method for detecting and analyzing the drought resistance phenotype of rice based on a phenotypic robot provided in this embodiment, which includes the following steps.

[0023] Step S1: Select some varieties from the core germplasm resource bank of rice to conduct a drought experiment on rice populations. The following operations were performed on each variety material for different drought days: (1) Use a phenotyping robot equipped with a hyperspectral camera to collect hyperspectral reflectance information of each variety material and process it to obtain hyperspectral phenotypic data; (2) Manually take leaf samples of each variety material and measure the contents of glucose, galactose and fructose; (3) Manually measure the fresh weight FW, fresh weight after soaking FWAS and dry weight DW of each variety material and calculate the relative water content RWC.

[0024] Specifically, the mounted hyperspectral camera continuously scans along the direction of the phenotypic robot's movement, collecting hyperspectral reflectance information of the entire planting unit at once. The data is stored in binary data stream format, and the total reflectance T and average reflectance A of the plant leaves under each band are obtained through image segmentation, image masking, and data processing, as well as their derived first derivative, second derivative, and logarithm.

[0025] Step S2: Based on the hyperspectral phenotypic data of each variety and the corresponding number of drought days, a drought day prediction model is constructed; based on the hyperspectral phenotypic data of each variety and the contents of glucose, fructose and galactose, glucose content prediction model, fructose content prediction model and galactose content prediction model are constructed respectively.

[0026] Specifically, the method for constructing a drought day prediction model is as follows: Hyperspectral phenotypic data collected on days 0, 3, 7, 10, and 14 of a drought period are divided into 5 categories. A random dataset partitioning method is adopted. First, a competitive adaptive reweighted sampling algorithm is used to perform preliminary feature screening on the massive hyperspectral phenotypic data to reduce the number of features. Then, the data is input into a random forest classifier to construct a classification model and output feature importance coefficients. Features with high importance coefficients are selected from these models. Finally, the data using these features are used to build a model through the random forest classifier. This model can predict the number of drought days from hyperspectral phenotypic data.

[0027] Specifically, the method for constructing a metabolite content prediction model is as follows: Spectral indices are preprocessed and variables are selected. Combined with metabolite content data, a regression model is constructed, and the optimal metabolite prediction model is determined. This includes the following steps: Data preprocessing uses four methods, including no preprocessing: None, Smoothing Convolution (SG), Standard Normal Variable Exchange (SNV), and Multivariate Scatter Correction (MSC). The preprocessed data is divided into three datasets using three methods: Random, KS, and SPXY. The training set after division is then classified using four feature reduction methods: None, Competitive Adaptive Reweighted Sampling (CARS), Continuous Projection Algorithm (SPA), and ReliefF Feature Selection. Finally, five regression methods are used: Partial Least Squares Regression (PLSR), Ordinary Least Squares Regression (OLSR), Ridge Regression (RR), Bayesian Ridge Regression (BRR), and Support Vector Regression (SVR) to construct the regression model. The regression model construction employs the T-squared method commonly used in statistics. The test method compared the metabolite content of rice leaves in the drought group and the control group, and selected the metabolites with significant differences between the drought and control groups. Finally, the optimal metabolite prediction model was determined by indicators such as coefficient of determination, normalized root mean square error, relative prediction bias, and number of features.

[0028] Step S3: Combining the four indicators of drought days, glucose content, galactose content, and fructose content from step S1, regression analysis is performed on the relative water content (RWC) obtained in step S1 to construct an accurate prediction model for rice relative water content (RWC).

[0029] Specifically, regression analysis was performed on the RWC of the validation samples based on the predicted number of dry days, glucose content, galactose content, and fructose content, and outliers were removed. No data preprocessing or feature reduction was performed. Three dataset partitioning methods (Random, KS, and SPXY) and five regression methods (Partial Least Squares Regression (PLSR), Ordinary Least Squares Regression (OLSR), Ridge Regression (RR), Bayesian Ridge Regression (BRR), and Support Vector Regression (SVR)) were selected. The modeling results of the various dataset partitioning and regression method combinations are shown below. Figure 2 As shown, the optimal regression model is a combination of Bayesian Ridge Regression (BRR) and KS dataset partitioning. This regression model can accurately predict the relative water content of plants and can serve as an important reference for assessing the drought severity of rice.

[0030] Step S4: Using a phenotypic robot equipped with a hyperspectral camera, high-frequency hyperspectral data is automatically collected from field crops to obtain hyperspectral data of the field crops at different times. This data is then input into multiple prediction models constructed in step S2 to obtain the predicted number of drought days, glucose content, galactose content, and fructose content. This data is then input into the relative water content (RWC) accurate prediction model constructed in step S3 to obtain the predicted relative water content (RWC). Based on the results at different times, the time-series relative water content (RWC(t)) of the experimental materials is obtained.

[0031] Step S5: Using the relative water content RWC(t) based on the time series obtained in step S4, calculate the characteristic index of the RWC(t) curve change, quantify the dynamic change process of each rice variety during the entire drought experiment period, and obtain the rice drought resistance performance index DTI, which is used to comprehensively evaluate the drought resistance of rice varieties.

[0032] Specifically, using the RWC-time curve, the following are calculated: area under the moisture curve during the drought period, duration of drought, stable moisture content after rehydration, moisture content at the point of most severe drought stress, and time from the start of rehydration to when moisture content stabilizes. The RWC-time curve and its characteristic indicators are as follows: Figure 3 As shown. The final drought resistance index (DTI) of rice is calculated using the following formula: ; Among them, AUC dry The area under the relative water content curve during the dry period; T d Duration of drought; RWC final The relative water content that is stable after rehydration; RWC min The relative water content at the point of most severe drought stress; T r The time from the start of rehydration to when the moisture content tends to stabilize.

[0033] Step S6: Using the rice drought resistance index DTI obtained in step S5 as a custom phenotypic trait, a genome-wide association analysis (GWAS) is performed on the whole genome sequences of the rice varieties mentioned in step S1 to identify significantly related gene loci and mine rice drought resistance-related genes.

[0034] Specifically, the rice drought resistance index (DTI) was used as a custom phenotypic trait. It was then used in GWAS analysis along with genotype data. Manhattan plots and QQ-plots were used to analyze the enriched characteristic spectral indices and metabolite types, uncovering new genetic associations between the characteristic spectral indices and known and unknown metabolites. Based on the corresponding characteristic spectral indices, genes at these chromosomes were located, and overexpression experiments were conducted to analyze changes in the content of related metabolites after overexpression. This further validated the genetic association between the characteristic spectral indices and metabolites, enabling the localization of functional genes responding to drought stress in rice. Figure 4 The diagram shows co-localized chromosomes, where multiple characteristic spectral indices co-localize gene loci on chromosome 6 (Chr06). Among the co-localized loci on chromosome Chr06, there are SNPs that are significantly associated with the characteristic spectral indices.

[0035] The specific examples described in this application are merely illustrative of the spirit of the invention. Those skilled in the art to which this invention pertains may make various modifications or additions to the specific examples described herein, or substitute them by similar means, without departing from the spirit of the invention or exceeding the scope defined by the appended claims.

Claims

1. A method for detecting and genetically analyzing rice drought resistance phenotypes based on phenotypic robots, characterized in that, Includes the following steps: Step A: Select some varieties from the core germplasm resource bank of rice to conduct a drought experiment on rice populations. The following operations were performed on each variety material under different drought days: (1) Use a phenotypic robot equipped with a hyperspectral camera to collect hyperspectral reflectance information of each variety material and process it to obtain hyperspectral phenotypic trait data; (2) Manually take leaf samples of each variety material and measure the contents of glucose, galactose and fructose; (3) Manually measure the fresh weight FW, fresh weight after soaking FWAS and dry weight DW of each variety material and calculate the relative water content RWC. Step B: Based on the hyperspectral phenotypic data of each variety and the corresponding number of drought days, a drought day prediction model is constructed; based on the hyperspectral phenotypic data of each variety and the contents of glucose, fructose and galactose, glucose content prediction model, fructose content prediction model and galactose content prediction model are constructed respectively. Step C: Combining the four indicators of drought days, glucose content, galactose content, and fructose content from Step A, regression analysis is performed on the relative water content (RWC) obtained in Step A to construct an accurate prediction model for the relative water content (RWC) of rice. Step D: Using a phenotypic robot equipped with a hyperspectral camera, high-frequency hyperspectral data is automatically collected from the field crop under test. The hyperspectral data of the field crop at different times is obtained and input into the four prediction models constructed in step B to obtain the predicted number of drought days, glucose content, galactose content and fructose content. Then, it is input into the relative water content (RWC) accurate prediction model constructed in step C to obtain the predicted relative water content (RWC). Based on the results at different times, the time-series relative water content (RWC(t)) of the experimental material is obtained. Step E: Using the relative water content RWC(t) based on the time series obtained in Step D, calculate the characteristic index of the RWC(t) curve change, quantify the dynamic change process of each rice variety during the entire drought experiment period, and obtain the rice drought resistance performance index DTI, which is used to comprehensively evaluate the drought resistance of rice varieties. Step F involves using the rice drought resistance index (DTI) obtained in Step E as a custom phenotypic trait, and combining it with the whole genome sequence of the rice varieties mentioned in Step A to perform a genome-wide association analysis (GWAS) to identify significantly related gene loci and mine rice drought resistance-related genes.

2. The method for detecting and genetically analyzing rice drought resistance phenotypes based on phenotypic robots according to claim 1, characterized in that, In step A, the hyperspectral phenotypic data of each variety of material is obtained as follows: the mounted hyperspectral camera continuously scans along the direction of the phenotypic robot to collect the hyperspectral reflectance information of the entire planting unit at one time. The data is stored in binary data stream mode. The total reflectance T and average reflectance A of the plant leaves under each band are obtained through image segmentation, image masking and data processing, as well as their derived first derivative, second derivative and logarithm.

3. The method for detecting and genetically analyzing rice drought resistance phenotypes based on phenotypic robots according to claim 1, characterized in that, The method for constructing the drought day prediction model in step B is as follows: the hyperspectral phenotypic data collected for different drought days are divided into several classes. A random dataset partitioning method is adopted. First, a competitive adaptive reweighted sampling algorithm is used to perform preliminary feature screening on the massive hyperspectral phenotypic traits to reduce the number of features. Then, the data is input into a random forest classifier to construct a classification model and output the feature importance coefficients. Features with high importance coefficients are selected from these models. Finally, the data using these features are used to build a model through the random forest classifier.

4. The method for detecting and genetically analyzing rice drought resistance phenotypes based on phenotypic robots according to claim 1, characterized in that, The specific method for constructing the glucose content prediction model, fructose content prediction model, and galactose content prediction model in step B is to use different data preprocessing, feature selection, data partitioning, and regression methods in combination, select the method with a high coefficient of determination and a small number of features, and output the optimal metabolite prediction model.

5. The method for detecting and genetically analyzing rice drought resistance phenotypes based on phenotypic robots according to claim 1, characterized in that, The linear regression formula for predicting relative water content (RWC) in step C using the number of drought days, glucose, fructose, and galactose is as follows: ； Among them, DryDays represents the number of dry days; Glucose represents the glucose content; Galactose represents the fructose content; and Fructose represents the galactose content.

6. The method for detecting and genetically analyzing rice drought resistance phenotypes based on phenotypic robots according to claim 1, characterized in that, The rice drought resistance index (DTI) in step E is calculated using the dynamic change curve based on the time series RWC(t) described in step H, to measure the material's moisture retention and recovery ability. The formula is: ； Among them: AUC dry The area under the relative water content curve during the dry period; T d Duration of drought; RWC final The relative moisture content that is stable after rehydration; RWC min The relative water content at the point of most severe drought stress; T r The time from the start of rehydration to when the moisture content tends to stabilize.

7. The method for detecting and genetically analyzing rice drought resistance phenotypes based on phenotypic robots according to claim 1, characterized in that, The GWAS process in step F is as follows: the sequencing data is preprocessed with genotype filling, and SNPs with a deletion rate of less than 20% are selected for subsequent analysis. The GWAS analysis process uses a mixed linear model, sets a threshold, and screens out significant SNP sites that exceed the threshold after Boltzmann test.

Images