Method for improving Crassostrea gigas genome selection predictability based on generic transcriptome information and application
By integrating pantranscriptome information at the population level of Pacific oysters, a classification of conserved, common, and rare transcripts was constructed. By combining a resilient network model to optimize SNP weight allocation, the problem of insufficient prediction accuracy of traditional genomic selection models in Pacific oyster breeding was solved, achieving higher prediction accuracy and genetic analysis capabilities.
Patent Information
- Application Number
- CN202511097585.8
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-08-06
- Publication Date
- 2025-11-18
AI Technical Summary
Traditional genomic selection models are insufficient in capturing non-additive effects in Pacific oyster breeding, and have limited analytical power for functional and rare variations, resulting in insufficient prediction accuracy.
By integrating pantranscriptome information at the population level of Pacific oyster, we construct conserved, common, and rare transcript classifications, and optimize SNP weight allocation using a resilient network model to predict genome selection.
This improved the predictive robustness and genetic analysis capability of the genomic selection model, and enhanced the accuracy of trait prediction and the generalization ability of the model for Pacific oysters.
Smart Images
Figure CN120966970A_ABST
Abstract
Description
Technical Field
[0001] This invention belongs to the field of molecular breeding, specifically relating to a method and its application for improving the predictive power of genome selection in the Pacific oyster based on pantranscriptome information. Background Technology
[0002] The Pacific oyster is one of the most widely farmed species globally, possessing significant economic value. Its economic traits primarily include growth characteristics, nutritional value, and stress resistance. Currently, trait improvement in Pacific oysters mainly relies on traditional selective breeding methods such as family selection and population selection. However, with the development and cost reduction of sequencing technology, molecular breeding, represented by Genomic Selection (GS), is gradually replacing traditional selection methods.
[0003] Genomic selection, a core technology in modern molecular breeding, utilizes genome-wide molecular markers (primarily single nucleotide polymorphisms, SNPs) to construct genome-wide prediction models, enabling early and accurate assessment of the breeding value of individuals or materials that have not yet undergone phenotypic identification. This method significantly shortens the breeding cycle and reduces costs, becoming a key tool for improving selection efficiency and accelerating genetic gain in plant and animal genetic improvement. However, traditional genomic selection models are mainly based on static SNP marker information at the DNA level, and their predictive accuracy is limited by many factors, such as insufficient capture of non-additive genetic effects, neglect of transcriptional expression and regulatory heterogeneity, underutilization of conserved and rare variations, and missed potential important functional genes.
[0004] Pantranscriptomics aims to construct a non-redundant set of all transcripts expressed by all individuals across different populations, subspecies, tissue types, or developmental stages of a target species, thereby capturing the maximum possible transcript diversity that a population can express. By integrating population-level expression diversity information, transcripts can be categorized into conserved, common, and rare transcripts based on their expression frequency. Different types of transcripts directly reflect gene conservation, specificity, and potential novel functions. Integrating them into the genomic selection framework is expected to compensate for the shortcomings of traditional genomic selection models in capturing dynamic genetic effects, thereby improving the prediction accuracy of complex traits. Summary of the Invention
[0005] The technical problem to be solved by this invention is to provide a genome selection optimization model for oysters that integrates population-level pantranscriptome information. This model can solve the problems of insufficient capture of non-additive effects and limited resolution of functional and rare variants in the traditional GS model.
[0006] To solve the above problems, the present invention adopts the following technical solution:
[0007] A method for improving the predictive power of genome selection in the Pacific oyster based on pan-transcriptome information, the steps of which are as follows:
[0008] S1, multi-tissue transcriptome sequencing was performed on several individuals from representative strains of the Pacific oyster;
[0009] S2, using a two-step re-alignment method to construct a unique transcriptome dataset for each sample, merge transcripts and filter out redundant and low-quality transcripts to construct a pan-transcriptome;
[0010] S3 uses the pan-transcriptome constructed in S2 as a reference to quantify the expression of all samples and classifies transcripts into three categories: conserved, common, and rare, based on their expression frequency.
[0011] S4, perform phenotyping and genome resequencing on the candidate population to obtain phenotypic data and SNP genotyping data;
[0012] S5. Extract SNPs within the three types of transcripts and their upstream and downstream ranges of 2kb as prior information for genomic selection to estimate the genomic breeding value (GEBV).
[0013] S6 uses the 10-fold cross-validation method to perform correlation analysis between the estimated breeding values and the actual phenotypic values, thereby evaluating the predictive accuracy of the model.
[0014] Further, step S1 specifically involves selecting no fewer than 10 individuals from each representative strain of Pacific oyster, extracting total RNA from the major tissues (gills, adductor muscle, mantle, gonads, digestive glands, and labial flaps) of each individual, constructing a strand-specific transcriptome library, and performing sequencing using next-generation sequencing technology. The sequencing data volume should be no less than 6G / sample.
[0015] Further, step S2 specifically involves: constructing the oyster genome (GCF_902806645.1) index and performing the first alignment to detect splicing sites in all samples; then integrating the newly discovered splicing sites for a second genome alignment; constructing a transcriptome dataset for each sample based on the alignment results; merging the newly identified datasets and the original reference transcriptomes; removing redundancy and filtering low-quality transcripts to construct a pantranscriptome.
[0016] Furthermore, step S3 specifically involves: transcripts expressed in ≥95% of all samples are considered conserved transcripts, transcripts expressed in 1%-95% are considered common transcripts, and transcripts expressed in ≤1% are considered rare transcripts.
[0017] Furthermore, step S4 specifically involves: the number of candidates should be no less than 1000 individuals; phenotypic data of all individuals should be measured; total DNA should be extracted; a resequencing library should be constructed; whole-genome resequencing and SNP genotyping should be performed using next-generation sequencing technology; and the sequencing depth should be no less than 10×.
[0018] Further, step S5 specifically involves: extracting SNPs within 2kb of the three transcripts and their upstream and downstream regions as three different prior information subsets for genomic selection analysis; and using a resilient network model to optimize the weight allocation of the three transcript-related SNPs by balancing sparsity and relevance, wherein the calculation formula is as follows:
[0019]
[0020] In the formula y i For typographic data, x ij Let w be the genotype in the j-th SNP sample i. j The weight information is determined by the prior type of label j, β j Let λ represent the effect size of label j, and λ represent the overall regularization strength.
[0021] The breeding value of an individual is calculated by summing the effect values of the SNPs. The formula is as follows:
[0022]
[0023] In the formula, p is the total number of SNPs, x ij β is the genotype value of marker j for individual i. j Let j be the effect value.
[0024] Furthermore, step S6 specifically involves: randomly dividing all samples (n) into 10 mutually exclusive subsets, each subset containing approximately n / 10 individuals, followed by iterative training and validation.
[0025] Furthermore, the iterative training method performs k rounds of training. In each training round, the remaining 9 subsets (excluding the k subsets) are merged and removed to form the training set, and the kth subset is used as the validation set. A breeding value prediction model is fitted on the training set, and the fitted model is used to predict and estimate the GEBV of individuals in the validation set. The accuracy of the model prediction is evaluated by calculating the Pearson correlation coefficient between GEBV and the actual measured value.
[0026] The present invention also provides the application of the method in oyster breeding.
[0027] Compared with existing technologies, the advantages of this invention are as follows:
[0028] This invention systematically integrates transcriptional expression frequency information at the population level of Pacific oysters, and performs functional stratification and weighting of gene functional units and their regulatory regions according to the degree of conservation of different transcripts, thereby enhancing the predictive robustness and genetic analysis capability of the genome selection model.
[0029] The markers for different conserved transcript types defined in this invention are applicable across major strains of Crassula longifolia, facilitating the integration of multi-population data and enhancing the model's generalization ability. Attached Figure Description
[0030] Figure 1 A flowchart of the genome prediction method provided by this invention;
[0031] Figure 2 The lengths and numbers of new and old transcripts in the pantranscriptome assembled in this invention;
[0032] Figure 3 This is a classification of different transcripts in the pantranscriptome assembled in this invention. Specific implementation methods
[0033] The technical solution of the present invention will be further explained below through embodiments, but the scope of protection of the present invention is not limited in any way by the embodiments.
[0034] Example 1
[0035] like Figure 1 As shown, a method for improving the predictive power of genome selection in the Pacific oyster based on pan-transcriptome information is described, and the steps of the method are as follows:
[0036] S1, multi-tissue transcriptome sequencing was performed on several individuals from representative strains of the Pacific oyster;
[0037] S2, using a two-step re-alignment method to construct a unique transcriptome dataset for each sample, merge transcripts and filter out redundant and low-quality transcripts to construct a pan-transcriptome;
[0038] S3 uses the pan-transcriptome constructed in S2 as a reference to quantify the expression of all samples and classifies transcripts into three categories: conserved, common, and rare, based on their expression frequency.
[0039] S4, perform phenotyping and genome resequencing on the candidate population to obtain phenotypic data and SNP genotyping data;
[0040] S5. Extract SNPs within the three types of transcripts and their upstream and downstream ranges of 2kb as prior information for genomic selection to estimate the genomic breeding value (GEBV).
[0041] S6 uses the 10-fold cross-validation method to perform correlation analysis between the estimated breeding values and the actual phenotypic values, thereby evaluating the predictive accuracy of the model.
[0042] As one implementation method, step S1 specifically involves selecting no fewer than 10 individuals from a representative strain of Pacific oyster, extracting total RNA from the major organs or tissues of each individual, constructing a strand-specific transcriptome library, and performing sequencing using next-generation sequencing technology. The sequencing data volume should be no less than 6G / sample.
[0043] In one implementation method, step S2 specifically involves: constructing an index of the oyster genome (GCF_902806645.1) and performing a first alignment to detect splicing sites in all samples; then integrating newly discovered splicing sites for a second genome alignment; constructing a transcriptome dataset for each sample based on the alignment results; merging the newly identified dataset and the original reference transcriptome, removing redundancy, and filtering low-quality transcripts to construct a pantranscriptome.
[0044] As one implementation method, step S3 specifically involves: transcripts expressed in 95% or more of all samples are considered conserved transcripts, transcripts expressed in 1%-95% are considered common transcripts, and transcripts expressed in 1% or less are considered rare transcripts.
[0045] As one implementation method, step S4 specifically involves: the number of candidates should be no less than 1000 individuals, phenotypic data of all individuals are measured, total DNA is extracted, a resequencing library is constructed, whole-genome resequencing is performed using next-generation sequencing technology and SNP genotyping is performed, and the sequencing depth should be no less than 10×.
[0046] In one implementation method, step S5 specifically involves: extracting SNPs within 2kb of the three transcripts and their upstream and downstream regions as three different prior information subsets for genomic selection analysis; and using a resilient network model to optimize the weight allocation of the three transcript-related SNPs by balancing sparsity and correlation, wherein the calculation formula is as follows:
[0047]
[0048] In the formula y i For typographic data, x ij Let w be the genotype in the j-th SNP sample i. j The weight information is determined by the prior type of label j, β j Let λ represent the effect size of label j, and λ represent the overall regularization strength.
[0049] The breeding value of an individual is calculated by summing the effect values of the SNPs. The formula is as follows:
[0050]
[0051] In the formula, p is the total number of SNPs, x ijβ is the genotype value of marker j for individual i. j Let j be the effect value.
[0052] As one implementation method, step S6 specifically involves: randomly dividing all samples (n) into 10 mutually exclusive subsets, each subset containing approximately n / 10 individuals, followed by iterative training and verification.
[0053] In a more preferred embodiment, the iterative training method performs k rounds of training. In each training round, the remaining 9 subsets (excluding the k subsets) are merged and removed to form the training set, and the kth subset is used as the validation set. A breeding value prediction model is fitted on the training set, and the fitted model is used to predict the GEBV of individuals in the validation set. The accuracy of the model prediction is evaluated by calculating the Pearson correlation coefficient between the individual's GEBV and the actual measured value.
[0054] Example 2: A specific application of a method for improving the predictive power of genome selection in Pacific oysters based on pan-transcriptome information.
[0055] 1. Multi-tissue transcriptome sequencing of a representative population of Pacific oyster.
[0056] 1.1 Sample Selection
[0057] Five representative groups of Pacific oysters were collected: “Haida No. 1”, “Haida No. 2”, “Haida No. 3”, Vibrio-resistant strains, and wild populations. Ten individuals were collected from each group, for a total of 50 individuals.
[0058] 1.2 Tissue Sampling
[0059] Six tissues were collected from each individual: gills, adductor muscle, mantle, gonads, labia, and digestive glands, totaling 300 samples. After the tissues were removed, they were flash-frozen in liquid nitrogen and then stored at -80°C for subsequent RNA extraction.
[0060] 1.3 RNA Extraction
[0061] The tissues were placed in a high-speed low-temperature grinder and ground at 4°C. Total RNA was extracted using the Trizol method, the concentration was measured using a Qubit 4.0 fluorescence quantitative quantitation instrument, the OD value was measured using a NanoDrop 2000 instrument, and the integrity was detected by agarose gel electrophoresis.
[0062] 1.4 Library Construction and High-Throughput Sequencing
[0063] mRNA was captured from total RNA samples using mRNA Capture Base. The enriched mRNA was randomly fragmented into 250bp–450bp segments. The mRNA was reverse transcribed into cDNA, and sequencing adapters were added, followed by 10–15 cycles of library amplification. To construct strand-specific libraries, dUTP was added during second-strand synthesis. The amplified library was sorted and purified using magnetic beads to ensure a final insert peak of approximately 300bp. Library fragment analysis was performed using an Agilent 2100. For qualified libraries, sequencing was performed using "pari-end 150" on an Illumina high-throughput sequencing platform, yielding 6GB of sequencing data per sample.
[0064] 2. Pan-transcriptome construction
[0065] 2.1 Sequencing data quality control
[0066] Clean Reads were obtained by applying the default filtering parameters to the raw sequencing data using the Fastp (V0.23.4) software.
[0067] 2.2 Data Comparison
[0068] The oyster genome (GCF_902806645.1) index was constructed using STAR (v2.7.8a) software with the parameters "--runMode genomeGenerate--genomeFastaFiles genome.fa". STAR was used to perform the first data alignment of clean reads with the indexed genome. All splice sites generated from the first alignment of samples were merged (SJ.out.tab). Based on these splice sites, STAR was used with the parameter "--sjdbFileChrStartEndSJ.out.tab" to construct the genome index a second time. STAR was used to perform the second data alignment of clean reads with the newly indexed genome, and the output BAM files were sorted.
[0069] 2.3 Pan-transcriptome construction
[0070] Transcript datasets were constructed independently using two software programs, Stringtie (v2.2.3) and Scallop (v0.10.5). The above datasets and the original reference transcripts were then merged using RTDmaker software, and low-quality transcripts and redundant data were filtered out to construct a pantranscriptome. The filtering parameters were "--SJ-reads 102--tpm 12--fragment-len 0.7--antisense-len 0.5".
[0071] 3. Transcript Quantification and Classification
[0072] 3.1 Quantitative
[0073] Using STAR software with the pantranscriptome as a reference, all clean reads of the samples were re-aligned to the reference genome. Trinity (v2.15.1) software was used to count the number of reads that matched the reference transcriptome, generate the expression matrix, and calculate the relative expression level FPKM value.
[0074] 3.2 Classification
[0075] Transcripts with FPKM < 1 were filtered out, and the remaining transcripts were divided into three types according to their expression frequency: transcripts expressed in 95% or more of the samples were defined as conserved transcripts, transcripts expressed in 1%-95% of the samples were defined as common transcripts, and transcripts expressed in less than 1% of the samples were defined as rare transcripts.
[0076] 4. Obtaining candidate population phenotypes and genotypes
[0077] 4.1 Phenotypic Measurement
[0078] One hundred and 1000 individuals from a representative population of Pacific oysters were randomly selected and their phenotypes were measured. The growth phenotypes were shell height, shell length, and shell width, and the stress resistance phenotypes were Vibrio resistance traits (survival time and survival status after resistance to Vibrio infection).
[0079] 4.2 Construction of resequencing libraries
[0080] Total DNA was extracted from individuals using the phenol-chloroform method and subjected to quality testing. DNA samples that passed the quality test were randomly fragmented into 300-500 bp fragments using transposase, and sequencing adapters were ligated to the enzyme digestion products. After purification with magnetic beads, PCR amplification, and library sorting, sequencing libraries were constructed. The constructed libraries underwent quality control using an Agilent 2100. Libraries that passed quality control were then used for whole-genome resequencing on an Illumina high-throughput sequencing platform using a "pair-end 150" sequencing strategy.
[0081] 4.3 Genotype Acquisition
[0082] SNP genotyping was performed using GATK (v4.1.9.0). The HaplotypeCaller module generated GVCF files for each individual, and the GenotypeGVFs module was used for co-analysis of the GVCFs to improve genotyping accuracy. Subsequently, VCFtools (v0.1.13) was used for filtering to obtain high-quality SNPs, with the filtering criteria being allele frequencies greater than 0.05 and deletion rates less than 0.01.
[0083] 5. Genomic selection analysis based on transcript conservation
[0084] 5.1 Prior Information Site Extraction
[0085] Extract the classification results and location information of different transcripts from step 3.2, and extend the upstream and downstream of each transcript by 2kb to capture its potential regulatory regions. Use the plink 1.9.0 software parameter "--extract range" to extract the SNP sites in these regions as a priori information dataset, with the dataset prefixes bld1, bld2, and bld3.
[0086] 5.2 Genome Selection Using Resilient Network Models
[0087] The elastic network model was implemented using LDAK (v6.1) software, initially based on a genome selection model using prior information. Weighted matrices LDAK.tagging and LDAK.matrix were constructed using the parameters "--power-0.25--annotation-number 3--anotation-prefix bld--calc-tagging LDAK" based on three different types of prior information. Subsequently, the LDAK elastic network model was used to estimate the GEBV of growth and stress resistance traits based on the LDAK.tagging file, with the parameter "--elastic".
[0088] The conventional genomic selection model, which does not consider prior information, was implemented using GCTA (v1.94.1) and the R package BGLR. The G matrix was constructed using the GCTA parameter "--make-grm-alg 1", and BGLR was used to read the G matrix and perform association analysis with growth and stress resistance phenotypes and estimate GEBV.
[0089] 6. Accuracy Assessment
[0090] The candidate population (1000 samples) was randomly divided into 10 mutually exclusive subsets, each containing approximately 100 individuals. Iterative training and validation were then performed. A total of 10 training rounds were conducted. In each round, the 9 subsets excluding the corresponding subset were merged and removed to form the training set, which was then used as the validation set. A breeding value prediction model was fitted to the training set, and the fitted model was used to predict the GEBV of individuals in the validation set. The Pearson correlation coefficient between GEBV and actual measurements was calculated using the `cor` function in R to evaluate the model's prediction accuracy.
[0091] 7. Results
[0092] 7.1 Pan-transcriptome assembly
[0093] After removing redundant and low-quality transcripts, 271,749 non-redundant new transcripts were finally constructed, with an average of 905 new transcripts identified per sample. The new transcripts and the original reference transcripts were combined to obtain a pan-transcriptome of *Orychophragmus violaceus* consisting of 345,045 transcripts, in which the length of the new transcripts was generally longer than that of the original reference transcripts. Figure 2 ).
[0094] Transcripts are classified into three categories according to their expression frequency, such as Figure 3 As shown, conserved transcripts accounted for 22.34%, common transcripts accounted for 52.12%, and rare transcripts accounted for 25.54%.
[0095] 7.2 Improved Genome Selection Efficiency
[0096] After filtering, the original SNP genotyping data yielded 15,456,923 high-quality SNPs. Based on the transcript classification results, SNPs within 2kb upstream and downstream were extracted to obtain three prior information sets: a conservative prior information set of 2,473,107 SNPs, a conventional prior information set of 6,646,476 SNPs, and a rare prior information set of 1,854,830 SNPs.
[0097] Using SNP datasets with different levels of conservation as prior information, the optimized model provided in this invention improves the prediction accuracy of growth and stress resistance traits of Pacific oysters by 0.235% to 2.064% compared with the conventional GBLUP model (Table 1).
[0098] Table 1 shows the prediction accuracy of the two GS models.
[0099]
Claims
1. A method for improving the predictive power of genome selection in the Pacific oyster based on pan-transcriptome information, characterized in that, The steps of the method are as follows: S1, multi-tissue transcriptome sequencing was performed on several individuals from representative strains of the Pacific oyster; S2, using a two-step re-alignment method to construct a unique transcriptome dataset for each sample, merge transcripts and filter out redundant and low-quality transcripts to construct a pan-transcriptome; S3 uses the pan-transcriptome constructed in S2 as a reference to quantify the expression of all samples and classifies transcripts into three categories: conserved, common, and rare, based on their expression frequency. S4, perform phenotyping and genome resequencing on the candidate population to obtain phenotypic data and SNP genotyping data; S5. Extract SNPs within three types of transcripts and their upstream and downstream 2kb ranges as prior information for genomic selection to estimate genomic breeding values. S6 uses the 10-fold cross-validation method to perform correlation analysis between the estimated breeding values and the actual phenotypic values, thereby evaluating the predictive accuracy of the model.
2. The method according to claim 1, characterized in that, Step S1 is as follows: Select no fewer than 10 individuals from each representative strain of Pacific oyster, and extract total RNA from the tissues of each individual, including gills, adductor muscle, mantle, gonads, digestive glands, and labial flaps, construct a strand-specific transcriptome library, and sequence it using next-generation sequencing technology. The sequencing data volume should be no less than 6G / sample.
3. The method according to claim 1, characterized in that, Step S2 specifically involves: constructing the oyster genome (GCF_902806645.1) index and performing the first alignment to detect splicing sites in all samples. Then, the newly discovered splicing sites are integrated for the second genome alignment. Based on the alignment results, a transcriptome dataset for each sample is constructed. After merging the newly identified dataset and the original reference transcriptome, redundancy is removed and low-quality transcripts are filtered to construct a pantranscriptome.
4. The method according to claim 1, characterized in that, Step S3 specifically involves defining transcripts expressed in 95% or more of all samples as conserved transcripts, transcripts expressed in 1%-95% as common transcripts, and transcripts expressed in 1% or less as rare transcripts.
5. The method according to claim 1, characterized in that, Step S4 specifically involves: the number of candidates should be no less than 1000 individuals, measuring the phenotypic data of all individuals, extracting total DNA, constructing a resequencing library, performing whole-genome resequencing and SNP genotyping using next-generation sequencing technology, and the sequencing depth should be no less than 10×.
6. The method according to claim 1, characterized in that, Step S5 specifically involves: extracting SNPs within 2kb of the three transcripts and their upstream and downstream regions as three distinct prior information subsets for genomic selection analysis; and using a resilient network model to optimize the weight allocation of the three transcript-related SNPs by balancing sparsity and correlation, with the calculation formula being: In the formula y i For typographic data, x ij Let w be the genotype in the j-th SNP sample i. j The weight information is determined by the prior type of label j, β j Let λ represent the effect size of label j, and λ be the overall regularization strength. The breeding value of an individual is calculated by summing the effect values of the SNPs. The formula is as follows: In the formula, p is the total number of SNPs, x ij β is the genotype value of marker j for individual i. j Let j be the effect value.
7. The method according to claim 1, characterized in that, Step S6 specifically involves randomly dividing all samples n into 10 mutually exclusive subsets, each subset containing approximately n / 10 individuals, followed by iterative training and validation.
8. The method according to claim 7, characterized in that, The iterative training method performs k rounds of training. In each training round, the nine subsets excluding the kth subset are merged and removed to form the training set, and the kth subset is used as the validation set. A breeding value prediction model is fitted on the training set, and the fitted model is used to predict and estimate the GEBV of individuals in the validation set. The accuracy of the model prediction is evaluated by calculating the Pearson correlation coefficient between GEBV and the actual measured value.
9. The application of the method according to any one of claims 1-8 in oyster breeding.