Screening and breeding methods for key disease-resistant phenotypes in oval pomfret based on Mendelian randomization

Abstract

This invention discloses a method for screening and breeding key disease-resistant phenotypes in oval pomfret based on Mendelian randomization, belonging to the field of aquaculture breeding. The method includes obtaining disease-resistant phenotype data, external phenotype data, and genotype data of a reference population of oval pomfret; performing genome-wide association analysis on the genotype data based on the disease-resistant phenotype data and external phenotype data respectively to obtain a set of disease-resistant SNP loci and a set of external phenotype-related SNP loci; using the set of external phenotype-related SNP loci as an instrumental variable and the set of disease-resistant SNP loci as an outcome information source, performing Mendelian randomization analysis on the external phenotype and the disease-resistant phenotype to screen out gill area, which has a causal relationship with the disease-resistant trait; and based on the gill area, selecting oval pomfret for disease resistance against *Erythrophagus iridis* to improve the accuracy of disease resistance trait prediction.

Description

Technical Field

[0001] This invention belongs to the field of aquaculture breeding, specifically relating to a method for screening and breeding key disease-resistant phenotypes in oval pomfret based on Mendelian randomization. Background Technology

[0002] The oval pomfret (Trachinotus ovatus) is one of my country's important marine aquaculture fish species. With the development of intensive and high-density aquaculture, disease problems have become increasingly prominent. *Amyloodinium ocellatum*, a highly lethal parasite, primarily infects the gills, skin, and fins of fish, causing mass mortality in a short period and severely impacting the oval pomfret aquaculture industry. Current control methods largely rely on chemical drugs, but these have limitations in efficacy, are prone to drug resistance, and pose environmental risks. Therefore, conducting disease-resistant breeding is a crucial approach to improving the species' disease resistance and reducing aquaculture losses.

[0003] In existing disease-resistant breeding techniques, simple disease resistance phenotypes such as survival or death are usually used as the basis for selection. Although such phenotypes can reflect the disease resistance level of an individual, the phenotypic information is relatively coarse and cannot meet the needs of precision breeding.

[0004] Conventional correlation analysis can reveal the statistical association between extrinsic phenotypes and disease resistance traits, but it struggles to distinguish between correlation and causation, and is easily influenced by environmental and confounding factors. Mendelian randomization, a method that uses genetic variation as an instrumental variable for causal inference, can improve the reliability of causal judgments. However, current technology lacks a technical solution for applying Mendelian randomization to the screening of phenotypes for resistance to the eye-spotted amygdalocha in pompanocytic tamarinus and further for use in breeding decisions. Summary of the Invention

[0005] The purpose of this invention is to provide a method for selecting key phenotypes based on Mendelian randomization and using them for disease resistance breeding of oval pompanophora against eye-spotted starch egg whipworm, in order to solve the problems of coarse disease resistance phenotypes, difficulty in determining key phenotypes, and low efficiency of disease resistance breeding in the prior art.

[0006] This invention is achieved through the following technical solution. A method for screening and breeding key disease-resistant phenotypes in oval pomfret based on Mendelian randomization, the method comprising the following steps: S1. Obtain disease resistance phenotype data, external phenotype data and genotype data of the reference population of oval pomfret, wherein the external phenotype data includes at least gill area; S2. Based on the disease resistance phenotype data and the external phenotype data, perform genome-wide association analysis on the genotype data to obtain the set of disease resistance-related SNP sites and the set of external phenotype-related SNP sites. S3. Using the set of SNP sites related to the external phenotype as an instrumental variable and the set of SNP sites related to disease resistance as the source of outcome information, perform Mendelian randomization analysis on the external phenotype and the disease resistance phenotype to screen out the target external phenotype gill area that has a causal relationship with the disease resistance trait from the external phenotype. S4. Based on the gill area determined in step S3, carry out disease-resistant breeding operations against the eye-spotted amyloliquefaciens in the oval pomfret.

[0007] Furthermore, the positive causal effect is manifested as follows: for every unit increase in gill area, the disease resistance increases by 0.0915.

[0008] Furthermore, in step S1, the disease resistance phenotype data is obtained by artificially infecting *Echinochloa spp.* and includes at least one of the following: a binary trait divided by survival / death, a hierarchical trait divided by death time, and a continuous trait recorded by death time.

[0009] Furthermore, in step S1, the external phenotypic data also includes one or more of the following: pectoral fin area, anterior pelvic fin height, anterior pelvic fin width, fork length and height, anterior pectoral fin area, anterior pectoral fin height, anterior pectoral fin width, and head length.

[0010] Furthermore, in step S2, the genome-wide association analysis employs a mixed linear model and combines principal component analysis results with a kinship matrix to correct for the population structure.

[0011] Furthermore, in step S3, the Mendelian randomization analysis employs one or more of the following methods: inverse variance weighting, MR Egger method, and weighted median method.

[0012] Furthermore, in step S4, the disease resistance breeding operation includes: incorporating the gill area as a fixed effect into a predictive model for genomic selection of the disease resistance phenotype, so as to predict breeding values, screen individuals, select parents, or evaluate families.

[0013] Furthermore, the prediction model is the Bayes Lasso model.

[0014] Furthermore, when incorporating the gill area as a fixed effect into the prediction model, the significant GWAS SNP sites related to the disease resistance phenotype obtained in step S2 are also incorporated as fixed effects.

[0015] Furthermore, in step S1, the genotype data is obtained by performing low-depth whole-genome resequencing on individuals in the reference population and then filling in the genotypes using a reference haplotype database.

[0016] The beneficial effects of this invention compared to the prior art are as follows: This invention introduces Mendelian randomization into the breeding of oval pomfret resistant to the eye-spotted amygdaloides, enabling the screening of key phenotypes causally related to disease resistance from multiple extrinsic phenotypes. This provides a new technical approach for precise phenotypic screening of disease resistance in aquatic animals. In one embodiment, incorporating GWAS significant loci and gill area as fixed effects into the genomic selection model further improves the prediction accuracy, reaching a maximum of 0.753.

[0017] The gill area obtained by screening in this invention has a positive causal relationship with disease resistance traits, and can be used as a key indicator in disease resistance breeding. It can be used for screening disease-resistant individuals, selecting parent lines, and evaluating family lines, thereby improving the accuracy and targeting of breeding.

[0018] This invention further utilizes the key phenotype for breeding value prediction, which can be combined with genomic selection models to improve the accuracy of disease resistance trait prediction and enhance its practical breeding application value. Attached Figure Description

[0019] Figure 1 This is a schematic diagram illustrating the acquisition and identification of external phenotypes. Figure 2 This is a schematic diagram showing the correlation analysis results between external phenotype and disease resistance phenotype; Figure 3 A schematic diagram showing the results of Mendelian randomization analysis of external phenotypes and disease resistance phenotypes; Figure 4 A schematic diagram of a model for predicting disease resistance breeding values ​​using key phenotypes. Detailed Implementation

[0020] The present invention will be further described below with reference to specific embodiments, but the scope of protection of the present invention is not limited to the following embodiments.

[0021] Example 1: Reference population construction and acquisition of disease resistance phenotype Healthy individuals of the oval pomfret were selected to construct an experimental population. Experimental materials were obtained from Guangxi Jinggong Marine Technology Co., Ltd. The reference population consisted of 1500 oval pomfrets with an average weight of 144±32 g. The host fish required for culturing the eye-spotted amylopectinus were cultured in 300 L tanks, with the water temperature maintained at a constant 26±0.5 ℃ using heating equipment. The oval pomfrets required for the reference population were temporarily housed in indoor cement tanks measuring 6×7×2 m. They were fed commercial feed at 8:00, 12:00, and 18:00 daily, at a rate of 3% of their body weight. Daily water changes, bottom cleaning, and water quality monitoring were performed.

[0022] After culturing *Eriocheir oryzae* to the infection stage, the experimental population was artificially infected, and the survival status, death time, and disease resistance level of each individual were recorded to form disease resistance phenotypic data.

[0023] In a preferred embodiment, the disease resistance phenotype data includes: (1) Disease resistance traits classified into two categories based on survival and mortality; (2) Disease resistance traits classified according to the time of death; (3) Disease resistance traits formed by continuous recording of death time.

[0024] As one specific implementation, after *Heliotropium indicum* was uniformly cultured to the turcellus stage, the median lethal dose (LD50) was determined. The experiment included one control group and five infection dose gradient groups, with 10 fish per group and three replicates per group. The infection doses were 4,000, 8,000, 12,000, 16,000, and 32,000 turcells / fish, respectively. Six hours after infection, the fish were transferred to parasite-free rearing tanks, and the mortality time and infection symptoms of each group were recorded. The LD50 was calculated based on the final number of survivors in each group after 7 days. The results showed that the LD50 of *Heliotropium indicum* against *Siniperca ovatus* was 9,850 turcells / fish.

[0025] When constructing the disease-resistant reference population, 1000 oval pomfret were randomly selected and subjected to the aforementioned LD. 50 Group challenge was conducted using an infectious dose. The spores were evenly added to an indoor cement tank. Two hours after infection, the oval pompano were transferred to a new, worm-free cement tank, with the water changed every 3 days. Post-infection environmental conditions were controlled as follows: water temperature 24.4–25.7 ℃, salinity 24–26, pH 7.3–7.5, and dissolved oxygen 2.9–5.1 mg / mL. The first death occurred on the 3rd day post-infection. After the onset of clinical symptoms, individuals were observed every 2 hours, and subsequent deaths were promptly collected, with weight, length, width, time of death, and external phenotypic photographs recorded.

[0026] Among them, binary traits are divided into resistance and susceptibility based on whether the individual survives; graded traits are divided into 0 to 13 levels based on the time of death, where level 0 represents the most susceptible and level 13 represents the most resistant; continuous traits directly use the time of death (h) as the phenotypic value.

[0027] Example 2: External Phenotype Collection and Candidate Phenotype Screening Images were captured for each individual in the reference population, and external phenotypic data were obtained using an external phenotypic recognition method. The external phenotypic data included at least gill area, and further included pectoral fin area, anterior pelvic fin height, anterior pelvic fin width, fork length and height, anterior pectoral fin area, anterior pectoral fin height, anterior pectoral fin width, and head length.

[0028] Specifically, such as Figure 1As shown, an image recognition method based on YOLOv8 was used to identify the external phenotype of the oval pomfret. Each sample was placed in a transparent grid sampling frame, and an external phenotype photograph was taken using a fixed camera. A total of 100 oval pomfret images were collected for annotation and model training. PyCharm 2023 and OpenCV were used to uniformly convert the images to 1080×720 pixels and save them as JPG format. Labelme was used to annotate the 100 oval pomfret images, and the annotated parts included the eyes, dorsal fin, pectoral fin, caudal fin, and outline.

[0029] This embodiment uses 500 images, divided into training, validation, and test sets in an 8:1:1 ratio. All training was performed on an Ubuntu 20.04.6 LTS operating system with an AMD EPYC 7763 64-core CPU and an NVIDIA GeForce RTX 3090 graphics card (24 GB VRAM). The model was trained using the YOLOv8 base model with an initial learning rate of 0.001, weight decay of 0.0005, an optimal training epoch of 300, and a batch size of 8. Recognition accuracy was evaluated using the Pearson correlation coefficient between the hand-measured data and the recognition data. The results showed that the correlation between the hand-measured data for total length and height and the recognition data was greater than 0.93.

[0030] This method identified 43 external phenotypes and further calculated proportional traits such as gill weight ratio, gill head ratio, gill body length area ratio, gill body length ratio, body length to height ratio, weight to length to height ratio, and condition factor to construct a phenotype database.

[0031] As a preferred implementation method, correlation analysis and machine learning screening are first performed on each external phenotype and disease resistance phenotype to obtain a set of candidate external phenotypes. Specifically, the Spearman rank correlation coefficient is calculated using the `cor` function in R language for correlation analysis. When the number of individuals exceeds 200, a Spearman rank correlation coefficient greater than 0.27 is used as the screening threshold for relevance. The results show that nine phenotypic traits with a correlation greater than 0.27 with the three disease resistance traits are identified: pectoral fin area, anterior pelvic fin height, anterior pelvic fin width, fork length and height, anterior pectoral fin area, anterior pectoral fin height, anterior pectoral fin width, head length, and gill area.

[0032] Machine learning methods were employed to perform association analysis between extrinsic phenotypes and binary disease resistance phenotypes. Models used included decision trees (DF), random forests (RF), and support vector machines (SVM), with a training, validation, and test set ratio of 8:1:1. StandardScaler was used for standardization during training, and model performance was evaluated using cross-validation, learning curves, and validation curves. Figure 2As shown, the results indicate that BPEFW, HEADW, PF, and GILLA are strongly associated with disease resistance traits, with the SVM model performing better in classifying BPEFW and GILLA.

[0033] Example 3: Genotype Acquisition and Genome-Wide Association Analysis Genomic DNA is extracted from each individual in the reference population and resequencing is performed to obtain raw sequence data. The raw sequence data undergoes quality control, alignment, variant detection, and filtering to obtain high-quality SNP data. Preferably, the genotype data is obtained through low-depth sequencing combined with genotype imputation.

[0034] Specifically, in the stage of constructing the reference haplotype database, paired-end sequencing libraries were constructed using the MGIEAsy library preparation kit manufactured by BGI Genomics, and PE150 sequencing was performed on the DNBSEQ-T7 platform, with a sequencing depth of 20× for each sample; in the stage of constructing the anti-A. ocellatumm breeding system, 1× low-depth resequencing was performed on the reference population and the validation population, and genotype filling was performed in conjunction with the aforementioned reference haplotype database.

[0035] The raw data quality control was performed using SOAPnuke with the following filtering parameters: "-n 0.01 -l 20 -q 0.3 --adaMR 0.25--ada_trim --polyX 50 --minReadLen 150". These parameters included: removing sequences containing more than 1% N, removing sequences less than 20 bp in length, removing sequences with a quality value less than 0.3, removing connector contamination, removing sequences containing more than 50 consecutive polyXs, and retaining sequences with a length greater than or equal to 150 bp.

[0036] The alignment and mutation detection process is as follows: Clean reads were aligned to the reference genome of the oval pomfret using bwa-mem (v0.7.17); SAMtools (v0.1.19) was used for SAM / BAM conversion and sorting; PCR duplicates were removed and the index was rebuilt using MarkDuplicates in GATK4 (v4.3.0.0); GVCF files were generated using HaplotypeCaller in GATK4, and combined with CombineGVCFs and GenotypeGVCFs to obtain the final genotype data.

[0037] The SNP filtering conditions were as follows: initial screening was performed using VariantFiltration in GATK 4.3 with parameters QUAL < 30.0, QD < 2.0, FS > 60.0, SOR > 4.0, MQRankSum < -12.5, and ReadPosRankSum < -8.0; subsequently, VCFtools (v0.1.16) was used to remove multi-allelic loci, MAF < 0.01 loci, and genotypic loci (max-missing = 1); and functional annotation was performed using SnpEff (v5.1).

[0038] As a preferred implementation, the genotype imputation was performed using a reference haplotype database constructed with Shapeit and Glimpse 2.0. Experimental results showed that the reference database constructed with Shapeit had higher imputation accuracy at various sequencing depths.

[0039] Genome-wide association studies (GWAS) were conducted targeting disease resistance and external phenotypes, respectively. Specifically, GWAS analysis was performed using a mixed linear model (MLM) with GEMMA software (v0.98.3), and the population structure was corrected using the phylogenetic matrix and PCA results. Correction was performed using the first 14 principal components and the phylogenetic matrix. The significance threshold was adjusted using Bonferroni correction (P=1 / N), and the suggestive threshold was set to −log10(P)=5. The genome expansion factor λ was also calculated and used to correct for the GWAS results.

[0040] In the GWAS analysis of the resistance to *A. ocellatumm*, 1,495,683 SNPs were retained for subsequent analysis after imputation and filtering. After correction using MLM combined with the first 14 PCA and kinship, GWAS analysis of 1,440,331 SNPs from 509 individuals was performed for the three resistance phenotypes. The results showed that the expansion factor for the binary trait was 1.072, for the grading trait it was 1.104, and for the time-to-death trait it was 1.084. Among these, 109 significant loci and 327 suggestive loci were identified for the time-to-death trait.

[0041] Example 4: Genetic correlation analysis and Mendelian randomization analysis Using the disease resistance-related SNP loci and extrinsic phenotype-related SNP loci obtained in Implementation Method 3, genetic correlation analysis was performed on candidate extrinsic phenotypes and disease resistance phenotypes to screen for extrinsic phenotypes with high genetic correlation.

[0042] Specifically, heritability and genetic correlation analyses were performed using GCTA software based on the REML algorithm and the BLUP linear model. First, a genetic relationship matrix (GRM) was constructed, and PCA was used to obtain population structure information. The first 10 principal components were included as covariates in the heritability estimation and genetic correlation analysis of all traits. For the A. ocellatumm resistance trait and the nine external phenotypes, univariate heritability and bivariate genetic correlations were calculated, respectively.

[0043] The results showed that the genetic correlation between the grading trait and time of death was 0.877, P = 2.74 × 10⁻⁶. -14 The genetic correlation between gill area (GILLA) and time of death was 0.972 (P=0.0036), showing a significant positive correlation.

[0044] As a preferred implementation, external phenotype-related SNPs are used as candidate instrumental variables, and disease resistance-related SNPs are used as the source of information for outcome variables. Mendelian randomization analysis is then performed on the candidate external phenotypes and disease resistance phenotypes.

[0045] Specifically, Mendelian randomization analysis was implemented using the TwoSampleMR package in the R language. First, the extrinsic GWAS data was read using the read_exposure_data function to ensure that information such as SNP effect value beta and standard error se were correctly imported. Then, the GWAS data for the resistance to A. ocellatumm was read in the same way as the outcome data. The SNP sites in the exposure data and the SNP sites in the outcome data were merged, and then the harmonise_data function was used to perform allele orientation alignment.

[0046] Preferably, the inverse variance weighted method (IVW) is used as the main analysis method, combined with the MR Egger method and the weighted median method for auxiliary verification; at the same time, the mr_heterogeneity function is used for heterogeneity test, the mr_pleiotropy_test function is used for pleiotropy test, and the results are visualized using mr_scatter_plot, mr_funnel_plot and mr_leaveoneout_plot.

[0047] The above analysis identified gill area as a key phenotype with a positive causal relationship to resistance to *A. ocellatum* in oval pompanoides. Specifically, 327 SNP loci associated with resistance to *A. ocellatum* were screened from the genome using GWAS, and a combined analysis of 9 extrinsic phenotypes and GWAS loci associated with resistance to *A. ocellatum* was performed, such as... Figure 3As shown, the results indicate that there is a significant genetic correlation and positive causal relationship between gill area and resistance to A. ocellatumm. For every unit increase in gill area, the resistance to A. ocellatumm increases by 0.0915.

[0048] Example 5: Application of key phenotypes in disease resistance breeding The key phenotypic gill area determined in Example 4 was used for screening individuals resistant to the eye-spotted amyloliquefaciens, selecting parents, evaluating families, or predicting breeding values ​​in oval pompano.

[0049] As a preferred implementation, gill area is incorporated as a fixed effect into the genomic selection model. More preferably, GWAS significant loci and gill area are both incorporated as fixed effects into the Bayes Lasso model to improve the accuracy of disease resistance trait prediction.

[0050] Specifically, to construct a genome-wide selection breeding system resistant to *A. ocellatum*, 14 SNP density panels were designed, with SNP numbers of 0.1 K, 0.5 K, 0.8 K, 1 K, 2 K, 3 K, 5 K, 8 K, 10 K, 20 K, 30 K, 50 K, 80 K, and 100 K. Three SNP selection strategies were implemented: random selection, equidistant selection, and GWAS uniform selection. The evaluated GS models included GBLUP, rrBLUP, Ridge Regression Bayes (RRB), Bayes A, Bayes B, Bayes C, Bayes Lasso, as well as random forest, support vector regression, and XGBoost—a total of 10 models. All Bayesian and linear models were implemented using the BGLR package in R software.

[0051] During model training, a 10-fold cross-validation method was used, dividing the dataset into 10 subsets. Nine subsets were selected as the training set and one subset as the validation set in each iteration, repeated 10 times. To reduce evaluation randomness, the number of iterations was set to 7000, and the burn period to 3000. The model's prediction accuracy was evaluated by calculating the Pearson correlation coefficient between the predicted GEBV and the observed phenotypes after adjusting for fixed effects.

[0052] The results show that, under the condition of equidistant point selection and 50 K SNPs, the Bayes Lasso model has the highest prediction accuracy for the categorical traits of *A. ocellatum*, reaching 0.538; based on this, if... Figure 4 As shown, after incorporating gill area as a fixed effect into the model, the prediction accuracy improved by another 23%, eventually reaching 0.753.

[0053] In the validation population, 240 fish were separately infected from the reference population for validation, and disease resistance phenotypic and external phenotypic data were re-collected. Based on the BGLR model training parameters and standardized parameters, GEBV prediction was performed on the validation population. The results showed that the correlation between the predicted GEBV and the actual disease resistance phenotype reached 0.7, indicating that this method has high application value in breeding for resistance to A. ocellatumm.

[0054] As a preferred implementation method, in simulated breeding, the Bayes Lasso model was used to simulate mating of a validation population of 100 A. ocellatum resistant individuals. In each generation, 30% of the superior individuals were selected according to GEBV for the next generation of breeding, and a total of 20 generations were simulated. The results showed that the cumulative genetic gain could be increased by 61.2% by the 4th generation.

[0055] Example 6

[0056] In a preferred embodiment, a reference population is constructed using artificial infection. The disease resistance trait is classified and graded as the main disease resistance phenotype, and gill area is used as the key extrinsic phenotype. Disease resistance-related SNPs and extrinsic phenotype-related SNPs are obtained through genome-wide association analysis. Then, the positive causal relationship between gill area and disease resistance is determined by genetic correlation analysis and Mendelian randomization analysis. Gill area and GWAS significant loci are included as fixed effects in the Bayes Lasso model for breeding value prediction, thereby achieving precise screening and disease-resistant breeding of individuals resistant to Eyespot Amygdaloides egg caseworm.

[0057] Specifically, this preferred embodiment uses the following parameters: The reference population consisted of 1500 healthy oval pomfret, of which 1000 were used for group challenge to construct a reference population resistant to *A. ocellum*. The infection dose was 9,850 vorticella spores / fish. The infection environment was water temperature 24.4–25.7 °C, salinity 24–26, and pH 7.3–7.5. The first death occurred on the 3rd day post-infection, and observations were recorded every 2 hours after the onset of clinical symptoms. External phenotypic identification was performed using TovNet-Vision, with images uniformly set to 1080×720 pixels. The YOLOv8 basic model was used for training, with 300 epochs, a batch size of 8, and a training, validation, and test set ratio of 8:1:1. Spearman correlation analysis threshold was 0.27, and 9 candidate external phenotypic types were selected. GWAS was performed using GEMMA. The MLM model v0.98.3 was used, and the population structure was corrected by combining the first 14 principal components and the kinship matrix. Mendelian randomization analysis was performed using the TwoSampleMR package in R language, with IVW as the main analysis method, supplemented by the MR Egger method and the weighted median method. The breeding value prediction was performed using the Bayes Lasso model, with equidistant 50 K SNP panels, 10-fold cross-validation, 7000 iterations, and 3000 burning periods. After incorporating gill area and significant GWAS sites as fixed effects, the model prediction accuracy reached 0.753.

Claims

1. A method for screening and breeding key disease-resistant phenotypes in oval pomfret based on Mendelian randomization, characterized in that, The method includes the following steps: S1. Obtain disease resistance phenotype data, external phenotype data and genotype data of the reference population of oval pomfret, wherein the external phenotype data includes at least gill area; S2. Based on the disease resistance phenotype data and the external phenotype data, perform genome-wide association analysis on the genotype data to obtain the set of disease resistance-related SNP sites and the set of external phenotype-related SNP sites. S3. Using the set of SNP sites related to the external phenotype as an instrumental variable and the set of SNP sites related to disease resistance as the source of outcome information, perform Mendelian randomization analysis on the external phenotype and the disease resistance phenotype to screen out the target external phenotype gill area that has a causal relationship with the disease resistance trait from the external phenotype. S4. Based on the gill area determined in step S3, carry out disease-resistant breeding operations against the eye-spotted amyloliquefaciens in the oval pomfret.

2. The method for screening and breeding key disease-resistant phenotypes in oval pomfret based on Mendelian randomization according to claim 1, characterized in that, The positive causal effect is shown as follows: for every unit increase in gill area, the disease resistance increases by 0.0915.

3. The method for screening and breeding key disease-resistant phenotypes in oval pomfret based on Mendelian randomization according to claim 1, characterized in that, In step S1, the disease resistance phenotype data is obtained by artificially infecting the eyedropper amyloides and includes at least one of the following: a binary trait divided by survival / death, a hierarchical trait divided by death time, and a continuous trait recorded by death time.

4. The method for screening and breeding key disease-resistant phenotypes in oval pomfret based on Mendelian randomization according to claim 1, characterized in that, In step S1, the external phenotypic data also includes one or more of the following: pectoral fin area, anterior pelvic fin height, anterior pelvic fin width, fork length and height, anterior pectoral fin area, anterior pectoral fin height, anterior pectoral fin width, and head length.

5. The method for screening and breeding key disease-resistant phenotypes in oval pomfret based on Mendelian randomization according to claim 1, characterized in that, In step S2, the genome-wide association analysis employs a mixed linear model and combines principal component analysis results with a kinship matrix to correct for population structure.

6. The method for screening and breeding key disease-resistant phenotypes in oval pomfret based on Mendelian randomization according to claim 1, characterized in that, In step S3, the Mendelian randomization analysis employs one or more of the following methods: inverse variance weighting, MR Egger method, and weighted median method.

7. The method for screening and breeding key disease-resistant phenotypes in oval pomfret based on Mendelian randomization according to claim 1, characterized in that, In step S4, the disease resistance breeding operation includes: incorporating the gill area as a fixed effect into a predictive model for genomic selection of the disease resistance phenotype, so as to predict breeding values, screen individuals, select parents, or evaluate families.

8. The method for screening and breeding key disease-resistant phenotypes in oval pomfret based on Mendelian randomization according to claim 7, characterized in that, The prediction model is the Bayes Lasso model.

9. The method for screening and breeding key disease-resistant phenotypes in oval pomfret based on Mendelian randomization according to claim 8, characterized in that, When incorporating the gill area as a fixed effect into the prediction model, the significant GWAS SNP sites related to the disease resistance phenotype obtained in step S2 are also incorporated as fixed effects.

10. The method for screening and breeding key disease-resistant phenotypes in oval pomfret based on Mendelian randomization according to claim 9, characterized in that, In step S1, the genotype data is obtained by performing low-depth whole-genome resequencing on individuals in the reference population and then filling in the genotypes using a reference haplotype database.

Images