Methods for identifying buffalo breeds and methods and apparatus for constructing predictive models.

Abstract

This invention relates to the field of buffalo breed identification, specifically to a method for buffalo breed identification and a method and apparatus for constructing a prediction model. The prediction model construction method includes: acquiring second-generation NGS sequencing data of a buffalo population; comparing the sequencing data with a reference genome and performing variant detection to obtain variant site data for each buffalo; merging and filtering the variant site data of all buffaloes to obtain a variant site set; and training the buffalo breed prediction model using the variant site set data to obtain the prediction model for the buffalo breed. The prediction model of this invention achieves high accuracy in buffalo breed identification.

Description

Technical Field

[0001] This invention relates to the field of buffalo breed identification, specifically to methods for identifying buffalo breeds and methods and apparatus for constructing predictive models. Background Technology

[0002] Traditional breed identification primarily relies on phenotypic identification. However, DNA molecular marker technology has evolved from early microsatellite and ALFP markers to SNP locus markers, which offer advantages such as wide distribution, high representativeness, genetic stability, and automation. Therefore, single nucleotide polymorphism (SNP) molecular identification, as an SNP-based genetic marker technology, has broad application prospects in fields such as paternity testing and germplasm resource identification. This technology provides unique genetic identity information for individuals by detecting and analyzing SNP differences, facilitating more accurate identification of genetic relationships and management of germplasm resources. In light of this, this invention provides a method for identifying buffalo breeds and a method and apparatus for constructing predictive models. Summary of the Invention

[0003] The technical problem to be solved by this invention is to provide a method for identifying buffalo breeds and a method and apparatus for constructing a predictive model thereon. The aim is to achieve high-accuracy identification of buffalo breeds.

[0004] The technical solution of the present invention to solve the above-mentioned technical problems is as follows: Firstly, a method for constructing a predictive model for a buffalo breed, the method comprising: Genetic data (second-generation NGS sequencing data) of the buffalo population were obtained, the genetic data were compared with a reference genome, and variant detection was performed to obtain variant site data for each buffalo. The mutation site data of all buffaloes were merged and filtered to obtain the mutation site set data; The mutation site set data is filtered based on the RFE (Recursive Feature Elimination) algorithm and the random forest sorting algorithm to obtain modeling site data. The modeling site data and the buffalo breeds corresponding to the modeling site data are used to train a prediction model for the buffalo breeds to obtain the prediction model for the buffalo breeds.

[0005] Furthermore, it also includes processing the genetic data of the buffalo population: the genetic data (Sequenced Reads or Raw Reads) is quality controlled to remove adapters and low-quality reads (sequence fragments) to obtain Clean Reads (high-quality sequence data). The main steps of data quality control are as follows: (1) Remove reads with adapters; (2) Filter reads with N content exceeding 10%; (3) Remove reads with more than 50% of bases with a quality value below 10; then compare the obtained Clean Reads with the reference genome, filter redundant reads according to the location results of the genetic data in the reference genome, and then perform variant detection, including SNP (single base variation) and InDel (insertion / deletion marker).

[0006] Based on the above technical solution, the present invention can be further improved as follows.

[0007] Furthermore, the prediction model for the buffalo breed is trained using the modeling site data and the corresponding buffalo breeds. The specific steps for obtaining the prediction model for the buffalo breed are as follows: The modeling site data is divided into a training set and a test set. The GT genotype of the variant site data in the training set is used as the training matrix, and the breed information of the sample is used as the target value to train the prediction model for the buffalo breed. The prediction model for the buffalo breed is then evaluated by performing 10-fold cross-validation on the training set and outputting the AUC value on the test set, thus obtaining the prediction model for the buffalo breed.

[0008] Furthermore, the specific method for mutation detection is as follows: GATK's HaplotypeCaller (Local Haplotype Assembly) algorithm is used to detect SNP and InDel mutations, obtaining mutation site data for each buffalo. The mutation results undergo rigorous filtering to ensure their reliability; the main filtering parameters are as follows: ① Based on the subroutine vcfutils.pl (varFilter-w 5-W 10) in bcftools, SNPs within 5bp of the INDEL and adjacent INDELs within 10bp are filtered out; ②clusterSize 2 clusterWindowSize 5 means that the number of mutations within a 5bp window should not exceed 2; ③QUAL<30, Phred quality value, indicates the possibility of variant variation at this site; quality values ​​below 30 are filtered out. ④ If QD < 2.0, the ratio of the quality of variation to the depth of coverage is the sum of the depths of all samples containing the variant base at this site; samples with QD below 2.0 are filtered out. ⑤MQ < 40, the root mean square of the alignment quality values ​​of all reads aligned to this site; reads with MQ below 40 are filtered out. ⑥FS>60.0, the value converted from the p-value of Fisher's test, describes whether there is obvious positive and negative strand specificity for reads containing only variants and reads containing only reference sequence bases during sequencing or alignment; that is, there will be no strand-specific alignment results, and FS should be close to zero; FS values ​​higher than 60 are filtered out. ⑦ Other variant filtering parameters are processed using the default values ​​specified by GATK. Variants occurring in the CDS region may cause changes in gene function. By searching for genes with non-synonymous SNPs and InDel and SV mutations in the CDS region between the reference genome and the sample, genes that may have functional differences between the sample and the reference genome can be identified.

[0009] The specific method for merging and filtering the mutation site data of all buffaloes is as follows: the mutation site data of all buffaloes are merged using the GenomicsDBImport algorithm of the GATK software package to obtain population site data; the population site data is then filtered using the plink software to obtain the mutation site set data; the filtering conditions are: --geno 0.01–maf 0.05–mind 0.01--indep-pairwise 50 10 0.2.

[0010] Furthermore, the prediction model construction method, while obtaining the prediction model for the buffalo breed, also includes: exporting and storing the prediction model for the buffalo breed to a cluster path, and sorting the importance of each variant site data returned by the prediction model for the buffalo breed, and exporting and storing the importance of each variant site data to a cluster path.

[0011] Secondly, a method for identifying a buffalo breed, the method comprising: The genetic data of the buffalo sample to be identified is compared with the reference genome, and mutation detection is performed to obtain the locus data of the buffalo sample to be identified. The mutation locus data is then filtered to obtain the mutation locus set data of the buffalo sample to be identified. The data of the mutation site set is imported into the prediction model of the buffalo breed constructed by the prediction model construction method to make predictions, thereby obtaining the breed information of the buffalo sample to be identified.

[0012] Thirdly, a predictive model building device for a buffalo breed includes: The data acquisition module is used to acquire the genetic data of the buffalo population, compare the genetic data with a reference genome, and perform mutation detection to obtain the mutation site data of each buffalo. The data processing module is used to merge and filter the mutation site data of all buffaloes to obtain the mutation site set data; The prediction model acquisition module is used to filter the mutation site set data based on the RFE algorithm and the random forest sorting algorithm to obtain modeling site data, and to train the prediction model of the buffalo breed using the modeling site data and the buffalo breed corresponding to the modeling site data, so as to obtain the prediction model of the buffalo breed.

[0013] Fourthly, a breed identification device for buffalo includes: the aforementioned predictive model construction device.

[0014] Fifthly, an electronic device comprising: a processor and a memory; the memory for storing a program, which, when executed by the processor, causes the processor to implement any of the methods for constructing a predictive model for buffalo breeds or the method for identifying buffalo breeds.

[0015] Sixthly, a storage medium comprising a stored program that, when the program is executed, controls the device on which the storage medium is located to execute the method for constructing a predictive model for buffalo breeds, or the method for identifying buffalo breeds.

[0016] Seventhly, a processor for running a program, wherein the program executes the method for constructing a predictive model for a buffalo breed or the method for identifying a buffalo breed.

[0017] This invention completed the mutation detection of 1468 buffalo samples (total data volume: 89T), and constructed and validated random forest and SVM models. After filtering, 41,549,629 mutation sites were obtained, and these sites were ranked by importance to select the best-performing sites for classification, ultimately selecting 8078 feature sites. Performance was evaluated using the ROC function and SVM algorithm. Both algorithms could effectively distinguish 30 different breeds. Except for a few breeds like S21 with an AUC value less than 0.9, the AUC values ​​of other breeds were all greater than 0.9. Furthermore, by adding 10 sites each time to evaluate the minimum subset that the model could distinguish, the analysis showed that the model had good discriminative ability when 200 sites were added, with over 95% of the breeds achieving an AUC value above 0.9.

[0018] The beneficial effects of this invention are as follows: Traditional methods for distinguishing buffalo breeds mainly rely on observing physical characteristics, geographical distribution, and production uses. This method is intuitive but highly subjective and difficult to accurately identify individuals with similar appearances or hybrids. In contrast, the genomic method of this invention, through direct analysis of DNA sequence variations, can perform absolutely accurate identification at the molecular level. It can not only distinguish breeds with highly similar genetic backgrounds with an accuracy rate of nearly 99%, but also reveal the evolutionary history and kinship of the population, and locate key genes affecting important economic traits such as milk production and disease resistance. This provides strong and objective technical support for scientific genetic resource protection, breed certification, and molecular breeding, and plays a decisive role in solving the problems of "synonyms" and "synonyms" among buffalo. Attached Figure Description

[0019] Figure 1 This is a flowchart of the mutation detection process of the present invention; Figure 2 This is a statistical chart showing the effective data volume of each sample in this invention; Figure 3 This is a statistical graph showing the sequencing quality value Q30 and GC content of each sample in this invention; Figure 4 This is a statistical graph showing the alignment information between the sample and the reference genome in this invention; Figure 5 A ranking chart showing the importance of the variant site data in this invention; Figure 6 This is the evaluation graph of the minimum subset of feature sites in the model of this invention; Figure 7 This is a flowchart of a method for constructing a predictive model for a buffalo breed according to the present invention. Detailed Implementation

[0020] The principles and features of this invention are described below. The examples given are for illustrative purposes only and are not intended to limit the scope of the invention. Where specific techniques or conditions are not specified in the embodiments, they should be performed according to the techniques or conditions described in the literature in this field, or according to the product instructions. Reagents or instruments whose manufacturers are not specified are all conventional products that can be purchased through legitimate channels.

[0021] Example This embodiment relates to a method for constructing a predictive model for a buffalo breed, the method comprising ( Figure 7 ): Genetic data (second-generation NGS sequencing data) of the buffalo population were obtained, the genetic data were compared with a reference genome, and variant detection was performed to obtain variant site data for each buffalo. The mutation site data of all buffaloes were merged and filtered to obtain the mutation site set data; The mutation site set data is filtered based on the RFE (Recursive Feature Elimination) algorithm and the random forest sorting algorithm to obtain modeling site data. The modeling site data and the buffalo breeds corresponding to the modeling site data are used to train a prediction model for the buffalo breeds to obtain the prediction model for the buffalo breeds.

[0022] Furthermore, the process also includes processing the genetic data of the buffalo population: The genetic data (Sequenced Reads or Raw Reads) undergoes quality control to remove adapters and low-quality reads (sequence fragments) to obtain Clean Reads (high-quality sequence data). The main steps of data quality control are as follows: (1) removing reads with adapters; (2) filtering reads with an N content exceeding 10%; (3) removing reads with more than 50% of bases having a quality value below 10. The obtained Clean Reads are then compared with a reference genome. Based on the location of the genetic data in the reference genome, redundant reads are filtered, and then variant detection is performed. Variance detection includes SNPs (single-base variants) and InDels (insertion / deletion markers).

[0023] Preferably, in this embodiment, the prediction model for a buffalo breed is trained using the modeling site data and the buffalo breed corresponding to the modeling site data. The specific steps for obtaining the prediction model for the buffalo breed are as follows: The modeling site data is divided into a training set and a test set. The GT genotype of the variant site data in the training set is used as the training matrix, and the breed information of the sample is used as the target value to train the prediction model for the buffalo breed. The prediction model for the buffalo breed is then evaluated by performing 10-fold cross-validation on the training set and outputting the AUC value on the test set, thus obtaining the prediction model for the buffalo breed.

[0024] Preferably, the specific method for mutation detection in this embodiment is as follows: using GATK's HaplotypeCaller (Local Haplotype Assembly) algorithm to perform SNP and InDel mutation detection to obtain mutation site data for each buffalo; wherein, the mutation results are strictly filtered to ensure the reliability of the mutation results.

[0025] The specific method for merging and filtering the mutation site data of all buffaloes is as follows: the mutation site data of all buffaloes are merged using the GenomicsDBImport algorithm of the GATK software package to obtain population site data; the population site data is then filtered using the plink software to obtain the mutation site set data; the filtering conditions are: --geno 0.01–maf 0.05–mind 0.01--indep-pairwise 50 10 0.2.

[0026] Preferably, the prediction model construction method described in this embodiment, while obtaining the prediction model of the buffalo breed, also includes: exporting and storing the prediction model of the buffalo breed to a cluster path, and sorting the importance of each variant site data returned by the prediction model of the buffalo breed, and exporting and storing the importance of each variant site data to a cluster path.

[0027] For example, by using a random forest or SVM model, the importance of each variant site data can be calculated to obtain the reduction data of the model accuracy or GINI coefficient for each variant site data; the feature site set can then be selected based on the reduction data of the model accuracy or GINI coefficient.

[0028] Specifically, this embodiment describes a method for constructing a predictive model for a buffalo breed, which includes the following specific steps: (1) Variation detection and quality control: First, the data was quality controlled to remove adapters and low-quality reads. Then, BWA alignment software was used to align with the reference genome (https: / / ngdc.cncb.ac.cn / gwh / Assembly / 262 / show). GATK was used for variant detection, and the GenomicsDBImport algorithm in GATK was used to merge large-sample VCFs (VCFs are text files describing SNPs (single-base variations), INDELs (insertion / deletion markers), and SVs (structural variation sites). Finally, Plink software was used to filter variants to obtain the final site information, followed by molecular marker site screening. The specific process is as follows: Figure 1 As shown.

[0029] (1-1) Sequencing data quality control analysis: The raw sequencing reads (Sequenced Reads or Raw Reads) obtained from sequencing contain low-quality reads with adapters. To ensure the quality of information analysis, the Raw Reads are filtered to obtain Clean Reads (high-quality sequence data) for subsequent information analysis. The main steps of data filtering are as follows: (1) Remove reads with adapters; (2) Filter reads with an N content exceeding 10%; (3) Remove reads with more than 50% of bases having a quality value below 10. The evaluation results of the sequencing output data for each sample are shown in [link to relevant documentation]. Figure 2 .

[0030] In next-generation sequencing (NGS), each base sequenced generates a corresponding quality value. This quality value is a key indicator of sequencing accuracy. A higher percentage of data with a quality value of Q30 or higher indicates higher sequencing accuracy. Q30 represents the quality value of a single base and also represents the error rate percentage for that base. Specifically, Q30 means a correct accuracy of 99.9% and an error rate of 0.1%. The GC content and Q30 quality values ​​of 1468 valid samples were statistically analyzed using FASTP software; the statistical chart is shown below. Figure 3 .

[0031] (1-2) Statistical analysis of variation results: The sequencing reads obtained from the second-generation sequencing need to be repositioned to the reference genome (https: / / ngdc.cncb.ac.cn / gwh / Assembly / 262 / show). Only samples with an alignment rate greater than 90% can be used for subsequent variant analysis. The bwa[1] software is mainly used to align short sequences obtained from second-generation high-throughput sequencing (such as Illumina sequencing platforms) with the reference genome. By aligning and locating the Clean Reads on the reference genome, the sequencing depth, genome coverage and other information of each sample are statistically analyzed, and variant detection is performed. The detection of SNP (Single Nucleotide Polymorphism) and small InDel (small Insertion and Deletion) is mainly achieved using the GATK[2] software toolkit. Based on the location results of the Clean Reads on the reference genome, samtools (v1.9)[3] is used to filter redundant reads to ensure the accuracy of the detection results. Then, GATK's HaplotypeCaller (Local Haplotype Assembly) algorithm is used to detect SNP and InDel mutations. Each sample first generates its own VCF, then performs population joint-genotype; finally, filtering is performed to obtain the final set of variant sites. Figure 4The mutation results undergo rigorous filtering to ensure their reliability. The main filtering parameters are as follows: ① The subroutine vcfutils.pl (varFilter-w 5-W 10) in bcftools filters out SNPs within 5bp of an INDEL and adjacent INDELs within 10bp. ② clusterSize 2 clusterWindowSize 5 means that the number of mutations within a 5bp window should not exceed 2; ③ QUAL < 30, Phred quality value, indicates the likelihood of variant variation at that site. Values ​​below 30 are filtered out. ④ For QD < 2.0, the ratio is the quality of variation divided by the depth of coverage, which is the sum of the coverage depths of all samples containing the variant base at this site. Samples with a QD below 2.0 are filtered out. ⑤ If MQ < 40, it represents the root mean square of the alignment quality values ​​of all reads aligned to this site. Reads with MQ below 40 are filtered out. ⑥ FS > 60.0, the value converted from the p-value of the Fisher test, describes whether there is significant positive or negative strand specificity for reads containing only variants and reads containing only reference sequence bases during sequencing or alignment. In other words, for alignments without strand specificity, the FS should be close to zero. Reads with FS higher than 60 are filtered out. ⑦ Other variant filtering parameters were processed using the default values ​​specified by GATK. Variants occurring in the CDS region may cause changes in gene function. By searching for genes with non-synonymous SNPs and InDel and SV mutations in the CDS region between the sample and the reference genome, genes with potential functional differences between the sample and the reference genome were identified. The variants between each sample and the reference genome are shown in Table 1 below; note that due to the large number of samples, Table 1 only shows the relevant information for the first 20 samples. For details, please refer to the statistical information in the results file.

[0032] Table 1. Statistics on the number of mutations in the sample. (1-3) Sample VCF merging: A high missing NP rate may indicate poor data quality, possibly due to technical reasons such as sequencing errors or improper sample processing. Filtering out these SNPs can improve the accuracy and reliability of subsequent analyses. Low-frequency alleles may be rare or recent variants in the population; their impact on phenotype may be small, or their effects may be difficult to estimate accurately due to limited sample size. Setting a MAF threshold can help reduce noise and focus on variants that are more likely to be biologically significant. Similar to genotype missing rates, a high individual missing rate may reflect poor data quality. Filtering out these individuals ensures the robustness of the analysis results and avoids bias caused by data quality issues.

[0033] In the genome, linkage disequilibrium may exist between adjacent SNP sites, meaning that the association between alleles at these sites is not random. This association can lead to redundant information provided by some SNP sites because they are highly correlated with neighboring SNP sites. By filtering out these highly correlated SNP sites, redundant information in the data can be reduced, thereby improving the efficiency and accuracy of subsequent analyses.

[0034] To ensure the accuracy of the variant locus data, the variant information of qualified individuals was merged using the GenomicsDBImport algorithm from the GATK software package. Population VCF loci were then screened using the Plink software with the following filtering conditions: `--geno 0.01 –maf 0.05 –mind 0.01 --indep-pairwise 50 10 0.2`. The final filtered dataset contained 41,549,629 variant loci for subsequent molecular marker screening.

[0035] (2) Model building: Using the obtained set of variant sites, the accuracy of the variety identification results was evaluated by combining the SVM model with random forest and using cross-validation analysis, ultimately achieving high-accuracy variety identification.

[0036] (2-1) Random Forest Model Construction and Validation: The filtered set of variant sites was split into a 6:4 data set, with 60% used as the training set and 40% as the test set. The variant site set GT (GT refers to the genotype of an individual at a specific SNP site, such as AA, AB, or BB, where A and B represent different alleles) of the training set was used as the training matrix. To obtain the optimal model parameters and model stability, 10-fold cross-validation was performed.

[0037] The importance of variables is calculated by recursively eliminating the corresponding variables and then calculating the decrease in model accuracy or GINI coefficient; the feature site set is then selected based on the decrease in model accuracy or GINI coefficient. Figure 5 ), Figure 5 In the graph, 0 indicates the impact of variable substitution on data classified as 0; 1 indicates the impact of variable substitution on data classified as 0; MeanDecreaseAccuracy indicates the decrease in accuracy after variable substitution; MeanDecreaseGini indicates the decrease in the GINI coefficient after variable substitution. Larger values ​​indicate more important variables. The importance of the 41,549,629 mutation sites after quality control was ranked, and the sites with the best classification were selected, ultimately resulting in 8,078 characteristic sites.

[0038] The performance was evaluated using the ROC function; the model was used to predict and plot ROC curves on the test set. The AUC area for 29 breeds was greater than 0.9, indicating that the constructed model can effectively predict and distinguish buffalo breeds. The AUC values ​​for 30 breeds predicted using the random forest algorithm are shown in Table 2 below. Table 2 Performance metrics of the Random Forest algorithm (2) SVM algorithm modeling and verification: SVM (Support Vector Machine) is a classic supervised learning algorithm primarily used to solve classification and regression problems. To test the accuracy of molecular markers, we simultaneously used the SVM algorithm for model building and validation. The AUC values ​​predicted by SVM for 30 varieties are shown in Table 3 below: Table 3 Performance metrics of the SVM algorithm The results in Table 3 show that the SVM algorithm can also distinguish 30 different varieties well. Except for a few varieties such as s21 with an AUC value of less than 0.9, the AUC values ​​of other varieties are all greater than 0.9.

[0039] 3. Minimum subset evaluation of molecular marker sites: For 8078 feature sites, the minimum subset of feature sites required for the model to achieve discriminative performance was evaluated by adding 10 sites at a time. The AUC value was then calculated iteratively for each iteration, as follows: Figure 6 As shown, the results indicate that the model has a good discriminative ability when the number of feature sites reaches 200, and the AUC value of more than 95% of the varieties can reach above 0.9.

[0040] This embodiment also relates to a method for identifying buffalo breeds, the method comprising: The genetic data of the buffalo sample to be identified is compared with the reference genome, and SNP and InDel variant detection is performed to obtain the locus data of the buffalo sample to be identified. The variant locus data is then filtered to obtain the variant locus set data of the buffalo sample to be identified. The data of the mutation site set is imported into the prediction model of the buffalo breed constructed by the prediction model construction method to make predictions, thereby obtaining the breed information of the buffalo sample to be identified.

[0041] This embodiment also relates to a predictive model construction device for buffalo breeds, including: The data acquisition module is used to acquire the genetic data of the buffalo population, compare the genetic data with a reference genome, and perform mutation detection to obtain the mutation site data of each buffalo. The data processing module is used to merge and filter the mutation site data of all buffaloes to obtain the mutation site set data; The prediction model acquisition module is used to filter the mutation site set data based on the RFE algorithm and the random forest sorting algorithm to obtain modeling site data, and to train the prediction model of the buffalo breed using the modeling site data and the buffalo breed corresponding to the modeling site data, so as to obtain the prediction model of the buffalo breed.

[0042] This embodiment also relates to a buffalo breed identification device, including: the aforementioned predictive model construction device.

[0043] This embodiment also relates to an electronic device, which includes a processor and a memory; the memory is used to store a program, which, when executed by the processor, causes the processor to implement any of the methods for constructing a prediction model for buffalo breeds or the method for identifying buffalo breeds.

[0044] This embodiment also relates to a storage medium, which includes a stored program that, when the program is running, controls the device where the storage medium is located to execute the method for constructing a prediction model for buffalo breeds, or the method for identifying buffalo breeds, as described above.

[0045] This embodiment also relates to a processor for running a program, which executes the method for constructing a prediction model for buffalo breeds or the method for identifying buffalo breeds when the program is running.

[0046] The memory may include non-permanent memory in computer-readable media, such as random access memory (RAM) and / or non-volatile memory, such as read-only memory (ROM) or flash RAM, and the memory includes at least one memory chip.

[0047] In summary, this invention completed the mutation detection of 1468 buffalo samples, and conducted analysis including the construction and validation of random forest and SVM models. Through filtering, 41,549,629 mutation sites were obtained, and these sites were ranked by importance to select the best-performing sites, ultimately identifying 8078 feature sites. Performance evaluation was performed using the ROC function and SVM algorithm. Both algorithms effectively distinguished 30 different breeds. Except for a few breeds like S21 with an AUC value less than 0.9, the AUC values ​​of other breeds were all greater than 0.9. Furthermore, we evaluated the minimum subset that the model could distinguish by adding 10 sites each time. Analysis showed that the model had excellent discriminative ability when 200 sites were added, with over 95% of the breeds achieving an AUC value above 0.9.

[0048] Although embodiments of the present invention have been shown and described above, it is understood that the above embodiments are exemplary and should not be construed as limiting the present invention. Those skilled in the art can make changes, modifications, substitutions and variations to the above embodiments within the scope of the present invention.

Claims

1. A method for constructing a predictive model for a buffalo breed, characterized in that, The prediction model construction method includes: Genetic data of the buffalo population is obtained, the genetic data is compared with a reference genome, and mutation detection is performed to obtain the mutation site data of each buffalo. The mutation site data of all buffaloes were merged and filtered to obtain the mutation site set data; The mutation site set data is filtered based on the RFE algorithm and the random forest sorting algorithm to obtain modeling site data. The modeling site data and the buffalo breeds corresponding to the modeling site data are used to train a prediction model for the buffalo breeds to obtain the prediction model for the buffalo breeds.

2. The method for constructing a predictive model for buffalo breeds according to claim 1, characterized in that, The training of a prediction model for a buffalo breed using the modeling site data and the corresponding buffalo breeds includes the following specific steps: The modeling site data is divided into a training set and a test set. The GT genotype of the variant site data in the training set is used as the training matrix, and the breed information of the sample is used as the target value to train the prediction model for the buffalo breed. The prediction model for the buffalo breed is then evaluated by performing 10-fold cross-validation on the training set and outputting the AUC value on the test set, thus obtaining the prediction model for the buffalo breed.

3. The method for constructing a predictive model for buffalo breeds according to claim 1, characterized in that, The specific method for mutation detection is as follows: use GATK's HaplotypeCaller algorithm to detect mutations and obtain mutation site data for each buffalo. The mutation site data of all buffaloes were merged and filtered. Specifically, the mutation site data of all buffaloes were merged using the GenomicsDBImport algorithm of the GATK software package to obtain population site data. The population site data was then filtered using the plink software to obtain the mutation site set data.

4. The method for constructing a predictive model for buffalo breeds according to claim 1, characterized in that, The prediction model construction method, while obtaining the prediction model for the buffalo breed, also includes: exporting and storing the prediction model for the buffalo breed to a cluster path, and sorting the importance of each variant site data returned by the prediction model for the buffalo breed, and exporting and storing the importance of each variant site data to a cluster path.

5. A method for identifying a buffalo breed, characterized in that, The identification method includes: Obtain the set of variant sites in the buffalo samples to be identified; The data of the set of variant sites is imported into the prediction model of buffalo breed constructed by the prediction model construction method of any one of claims 1 to 4 for prediction, thereby obtaining the breed information of the buffalo sample to be identified.

6. A device for constructing a predictive model for a buffalo breed, characterized in that, include: The data acquisition module is used to acquire the genetic data of the buffalo population, compare the genetic data with a reference genome, and perform mutation detection to obtain the mutation site data of each buffalo. The data processing module is used to merge and filter the mutation site data of all buffaloes to obtain the mutation site set data; The prediction model acquisition module is used to filter the mutation site set data based on the RFE algorithm and the random forest sorting algorithm to obtain modeling site data, and to train the prediction model of the buffalo breed using the modeling site data and the buffalo breed corresponding to the modeling site data, so as to obtain the prediction model of the buffalo breed.

7. A breed identification device for buffalo, characterized in that, include: The predictive model construction apparatus according to claim 6.

8. An electronic device, characterized in that, The electronic device includes a processor and a memory; the memory is used to store a program, which, when executed by the processor, causes the processor to implement the method for constructing a predictive model for buffalo breeds as described in any one of claims 1 to 4, or the method for identifying buffalo breeds as described in claim 5.

9. A storage medium, characterized in that, The storage medium includes a stored program that, when the program is executed, controls the device where the storage medium is located to perform the method for constructing a predictive model for buffalo breeds as described in any one of claims 1 to 4, or the method for identifying buffalo breeds as described in claim 5.

10. A processor, characterized in that, The processor is used to run a program, which executes the method for constructing a predictive model for buffalo breeds as described in any one of claims 1 to 4, or the method for identifying buffalo breeds as described in claim 5.

Images