Peking duck variety identification method based on machine learning

Abstract

A Beijing duck variety identification method based on machine learning is characterized by comprising the following steps: (1) extracting genome DNA of a duck to be detected; (2) detecting the genotypes of 200 SNP loci in the duck genome DNA to be detected; and (3) processing the SNP locus genotype information obtained in the step (2) by using a machine learning model to obtain a judgment result. According to the method, the difficulty of Beijing duck variety identification is preliminarily solved, the efficiency of Beijing duck variety identification can be greatly improved, the experimental scheme can be used for construction of molecular identity cards of other varieties, and a great promotion effect is brought to protection and utilization of germplasm resources.

Description

Technical Field

[0001] The present application belongs to the field of poultry breed identification. Specifically, the present application provides a method for identifying Beijing duck breeds based on machine learning. Background Art

[0002] In recent years, the protection and utilization of germplasm resources has become a hot topic. At present, with the increasing global exchanges, the blood exchanges of livestock and poultry breeds are also increasing, which has also led to the protection and utilization of genetic resources becoming the focus and difficulty of germplasm resource protection and utilization. Sequencing technology can easily obtain a large number of SNP variations. At the same time, the information content of SNPs can be analyzed by using population genetic parameters, and SNP sites with high information content can be found for analysis, which can better complete the work of variety classification. In order to better protect and utilize these germplasm resources, it is necessary to establish a complete, feasible and highly accurate variety allocation process and method as soon as possible.

[0003] At present, machine learning algorithms have been widely used due to their high accuracy and strong generalization. At present, there are some related methods for cattle, sheep, pigs and other animals, using different population genetic parameters to screen high-information SNP sites, and using machine learning methods to establish models and predict breed classification. However, there are still few related studies on duck breed allocation. The molecular information of many duck breeds is unclear, and the identification methods are relatively simple, resulting in insufficient identification and breeding technology for new breeds. We used sequencing technology, population genetic parameter analysis and machine learning algorithms to establish a set of Beijing duck breed identification processes and models, and verified the feasibility of this method, which can provide technical support for the subsequent classification and protection of other duck breeds, and play an important role in promoting the revitalization of the seed industry. Summary of the invention

[0004] On the one hand, the present application provides a method for identifying the variety of Beijing ducks based on machine learning, characterized in that the method comprises: (1) extracting the genomic DNA of the duck to be tested; (2) detecting the genotype of the SNP site in the genomic DNA of the duck to be tested; (3) using a machine learning model to process the SNP site genotype information obtained in (2) to obtain a judgment result; the SNP site is shown in Table 1.

[0005] Samples for extracting genomes may be blood, plasma, muscle, feathers, feces and other commonly used poultry test samples, with blood samples being preferred.

[0006] Furthermore, (2) uses whole genome resequencing, SNP gene chip method, mass spectrometry, SNaPshot method, KASP method, ARMS method, CAPS method, molecular beacon method, melting curve method, Taqman probe or direct sequencing method to detect the SNP site genotype in the duck genome DNA to be tested.

[0007] Whole genome resequencing, gene chip method, mass spectrometry, SNaPshot method, KASP method, ARMS method, CAPS method, molecular beacon method, melting curve method, Taqman probe, direct sequencing method and other methods can be used to obtain SNP data in this application. Due to the large number of SNP sites, whole genome resequencing and SNP gene chip method are preferred, and SNP gene chip method is particularly preferred. Other methods can also be applied after appropriate adjustments, grouping and other improvements.

[0008] In the construction of SNP gene chips, those skilled in the art can perform operations such as selection of solid phase supports, probe design and fixed connection according to conventional knowledge in the field, or entrust the processing of commercially available products.

[0009] Furthermore, the machine learning model in (3) is a support vector machine model.

[0010] The machine learning model of the present application preferably uses the support vector machine algorithm (SVM). Support vector machine is a binary classification model. Its basic model is a linear classifier with the largest interval defined in the feature space. It is one of the most commonly used machine learning algorithms. Support vector machine performs well in both high-dimensional data and low-dimensional data. At the same time, it also has the advantages of good robustness, strong generalization ability, and relatively intuitive model. Support vector machine is widely used in variety classification due to its high efficiency and strong generalization in classification work. The basic principle of support vector machine can be simply summarized as: by finding an optimal decision boundary, separate different categories of data and maximize the "interval" between different categories. The maximum interval makes it different from the perceptron; SVM also includes kernel skills, which makes it a substantially nonlinear classifier. In variety classification work, it can learn the characteristics of SNPs of different varieties, and then use this information to build a model to complete the variety classification work.

[0011] In addition to SVM, models such as linear regression, clustering, label propagation, and Q-learning can be used.

[0012] On the other hand, the present application provides a method for constructing a Beijing duck breed identification model based on machine learning, characterized in that the method comprises:

[0013] (A) Obtain SNP data of different duck breeds and perform quality control;

[0014] (B) Genetic parameter calculation of the quality-controlled SNP data obtained in (A);

[0015] (C) Some SNP loci with higher European genetic distance were selected to construct a Beijing duck breed identification model based on machine learning.

[0016] Furthermore, in (A), SNP data of different duck breeds are obtained by one or more methods selected from database download, whole genome resequencing, and SNP gene chip sequencing.

[0017] Furthermore, in (A), whole genome resequencing or gene chip sequencing is performed, and the SNP data of different duck breeds are obtained by comparison of the sequencing data using the gatk process and variation comparison.

[0018] Furthermore, in (B), the Euclidean genetic distance of the quality-controlled SNP data is calculated and sorted accordingly.

[0019] On the other hand, the present application provides a SNP gene chip, characterized in that the SNP gene chip comprises reagents for detecting the SNP sites shown in the following Table 1.

[0020] On the other hand, the present application provides the use of the above-mentioned SNP gene chip in the preparation of Beijing duck breed identification reagents.

[0021] The present invention proposes a combination of 200 SNP sites and constructs a machine learning model, which provides a method for the identification of Beijing ducks and non-Beijing ducks, preliminarily solves the problem of Beijing duck breed identification, and can greatly improve the efficiency of Beijing duck breed identification. The experimental scheme can be used for the construction of molecular identity cards of other breeds, which has greatly promoted the protection and utilization of germplasm resources.

[0022] The present invention obtains for the first time the SNP combination and machine learning model for Beijing duck breed identification, which can be applied to Beijing duck breed identification, facilitates the identification of Beijing duck breeds, facilitates the subsequent use of germplasm resources, and also provides an important reference for breed identification. BRIEF DESCRIPTION OF THE DRAWINGS

[0023] Figure 1 Displays the distribution of SNP data used in the experiment on chromosomes. DETAILED DESCRIPTION

[0024] Example 1 Beijing duck breed distribution feature selection and model construction

[0025] Blood samples were collected from local duck breeds across the country and some data were downloaded from NCBI. There are 16 breeds in total, namely, spot-billed duck, Jinding duck, Jingxi duck, Congyang duck, naked-necked duck, Rongshui fragrant duck, Shaoxing duck, Xilin mallard, Yulin mallard, Fenghua duck, Gaoyou duck, Yulin black duck, Liancheng white duck, Longsheng green duck, mountain mallard and Beijing duck, with a total of 206 ducks, including 20 Beijing ducks.

[0026] The blood samples of these duck breeds were randomly interrupted using a Covaris ultrasonic crusher, and then the entire library preparation was completed through end repair, A-tailing, sequencing adapter addition, purification, PCR amplification and other steps. The fastq files of all samples were obtained by on-line testing on the DNBseq platform. First, the obtained fastq files were subjected to basic quality control to obtain cleandata, and then the fastq files were aligned, deduplicated, sorted and variant detected using the GATK process. The reference genome used in this study was the latest version of the Mallard reference genome Mallard (ASM874695v1) downloaded from the Ensembl official website, and a VCF file containing SNP variations was obtained. Subsequently, the VCF file was hard-quality controlled using the GATK software to screen out SNP sites with poor quality.

[0027] After obtaining high-quality VCF files, this study conducted soft quality control on high-quality VCF files. The main quality control steps are: first, only retain the SNP sites on the double allele and autosomes, and then remove the SNP sites with a missing rate greater than 5% and the SNP sites with a minor allele frequency less than 5% to obtain the final VCF file. Next, the mean AED between each group is calculated, and the formula is as follows:

[0028]

[0029] Where P Aj is the frequency of allele A in the jth variety, and n is the number of varieties. The AED values ​​between each population were calculated, and the average was calculated and sorted from large to small.

[0030] Then, the first 2000 SNPs were extracted from the largest to the smallest AED values, and the sklearn package in Python was used to perform machine learning feature screening on these SNPs. The model was selected as the linear kernel of the support vector machine, and C = 1.0. The feature selection algorithm was used to obtain 200 SNPs as characteristic loci for breed allocation. Finally, we divided the training set and the test set in a ratio of 7:3, that is, 144 ducks were used as the training set and 62 ducks were used as the test set. The model was built using the training set and tested using the test set. Both the training set and the test set were correctly allocated.

[0031] Table 1 Model parameters of support vector machine

[0032]

[0033]

[0034]

[0035]

[0036]

[0037]

[0038]

[0039]

[0040] Example 2 Use of the Peking Duck Breed Allocation Model

[0041] Blood was collected from 55 Beijing ducks, and preliminary work such as database construction and superior testing was carried out. The obtained VCF file was used to directly extract the 200 SNP loci mentioned above, and the VCF file was converted into a 012 format file using the --recode A parameter in the PLINK software. Missing values ​​were filled with random numbers of 0, 1, and 2, and the SNP information column was extracted to form a SNP feature information file.

[0042] The sklearn package in Python can be used to load the support vector machine model provided in Example 1 or generate a model using the support vector machine parameters provided in Example 1, and then the SNP feature information file is output to the machine learning model. The model will return a prediction result, where 1 represents Beijing duck and 0 represents non-Beijing duck.

[0043] We used 55 preserved Beijing ducks as a validation set to verify the model. The model successfully identified the 55 preserved Beijing ducks as Beijing ducks, proving the reliability of the machine learning allocation model.

[0044] Table 2 Basic information and test results of machine learning training set, test set and validation set

[0045] Training set Test Set Validation set Sample size 144 62 55 Correct allocation number 144 62 55 Allocation accuracy 100％ 100％ 100％

[0046] This study also tested more than 60 non-Peking ducks, and the allocation results could achieve an accuracy of more than 99%.

[0047] The above experimental results show that the SNP site combination and machine learning model proposed in this study can well allocate the Beijing duck breeds. At the same time, the model was tested using an exogenous population and good results were obtained. This shows that the machine learning model has a good ability to allocate Beijing duck breeds, which provides an important basis for its subsequent application in breed identification.

Claims

1. A method for identifying Peking duck breeds based on machine learning, characterized in that: The method comprises: (1) extracting genomic DNA of a duck to be tested; (2) detecting the genotype of a SNP site in the genomic DNA of the duck to be tested; (3) using a machine learning model to process the genotype information of the SNP site obtained in (2) to obtain a judgment result; The SNP sites are the following SNP sites: 。 2. According to the method of claim 1, (2), the SNP site genotype in the duck genome DNA to be tested is detected by using whole genome resequencing, SNP gene chip method, mass spectrometry, SNaPshot method, KASP method, ARMS method, CAPS method, molecular beacon method, melting curve method, Taqman probe or direct sequencing method.

3. According to the method of claim 1, the machine learning model in (3) is a support vector machine model.

4. A method for constructing a Beijing duck breed identification model based on machine learning, characterized in that: The method comprises: (A) Obtain SNP data of different duck breeds and perform quality control; (B) Genetic parameter calculation of the quality-controlled SNP data obtained in (A); (C) Some SNP loci with higher European genetic distance were selected to construct a Beijing duck breed identification model based on machine learning.

5. The method according to claim 4, wherein (A) SNP data of different duck breeds are obtained by one or more methods selected from database downloading, whole genome resequencing, and SNP gene chip sequencing.

6. The method according to claim 5, wherein (A) the whole genome resequencing method or gene chip method is used for sequencing, and the SNP data of different duck breeds are obtained by comparing the sequencing data with the gatk process and comparing the variation.

7. The method according to any one of claims 4 to 6, wherein (B) the Euclidean genetic distance of the quality-controlled SNP data is calculated and sorted accordingly.

8. A SNP gene chip, characterized in that: The SNP gene chip contains reagents for detecting the following SNP sites:

9. Use of the SNP gene chip according to claim 8 in preparing a reagent for identifying the variety of Peking duck.

Images