A method for identifying Jiangshan black-boned chicken based on AI algorithm

Abstract

This invention belongs to the field of Jiangshan Black-boned Chicken identification technology, specifically an AI-based method for identifying Jiangshan Black-boned Chicken. Based on the reference genome GRCg7b, GATK software is used for population SNP identification to obtain the original SNP dataset. After quality control based on allele frequency and deletion rate, over 11 million SNP loci are obtained. Loci with linkage disequilibrium coefficients >0.1 are filtered out, ultimately retaining over 420,000 SNP loci. Jiangshan Black-boned Chicken-specific SNPs are screened using attention weights output from a linear self-attention model, combined with fixed differentiation site scores. Using any SNP and any combination of SNPs as feature variables, the probability of identification as Jiangshan Black-boned Chicken is evaluated using an XGBoost model. This invention has advantages such as high identification accuracy, high computational efficiency, simple operation, and low cost.

Description

Technical Field

[0001] This invention belongs to the field of Jiangshan Black-boned Chicken identification technology, specifically a Jiangshan Black-boned Chicken identification method based on AI algorithms. Background Technology

[0002] Establishing a precise breed identification technology system is an important means of protecting the purebred bloodline of local chickens.

[0003] For a long time, livestock and poultry breed identification has primarily relied on traditional morphological observation. This involves distinguishing physical characteristics such as feather color, body size, crest and wattles, and skin color, combined with production performance. While simple to operate, traditional methods are easily affected by non-genetic factors such as feeding environment and nutritional levels, and struggle to differentiate between similar-looking individuals with mixed bloodlines, resulting in insufficient accuracy. Subsequently, with the development of molecular biology techniques, genomic variations have gradually become "molecular identity cards" for different breeds. Identifying breeds by identifying specific SNP (Single nucleotide polymorphism) sites and calculating detection rates has, to some extent, compensated for the limitations of traditional methods. In recent years, with the widespread adoption and deep application of machine learning algorithms and artificial intelligence in agriculture, AI algorithms have demonstrated significant advantages in feature mining, breed identification, and trait prediction, especially in processing large-scale genomic feature variables, significantly improving computational efficiency and accuracy.

[0004] Compared to traditional methods, SNP-based AI algorithms do not rely on human judgment or prior information, and can automatically mine features potentially associated with the target across the entire genome. Self-attention models are important algorithms in deep learning, the core of which lies in assigning dynamic attention weights to each input element, quantifying the contribution of different features, and thus highlighting key information. However, in practical applications, they face the problem of excessively high computational complexity. Linear self-attention models, as an optimized upgrade of the self-attention mechanism, reduce the computational complexity from O(N...)... 2 The computational complexity is reduced to O(N), resolving the technical pain points of high memory consumption and long computation time in large-scale SNP calculations, while accurately capturing important features associated with the target. Currently, in livestock and poultry breed identification, the industry generally uses traditional statistical methods such as FST and chi-square test, and research on the application of artificial intelligence algorithms such as linear self-attention mechanisms has not been reported. There is still considerable room for improvement in the efficiency and accuracy of breed identification. Summary of the Invention

[0005] In order to overcome the shortcomings of the prior art, at least one technical problem raised in the background art is solved.

[0006] The technical solution adopted by this invention to solve its technical problem is: a method for identifying Jiangshan black-boned chicken based on AI algorithm, comprising the following steps: Step 1: Based on the reference genome GRCg7b, use GATK software to identify population SNPs and obtain the original SNP dataset; Step 2: After quality control based on allele frequency MAF < 0.05 and missing rate > 0.2, more than 11 million SNP loci were obtained; PLINK software was used to filter out loci with linkage disequilibrium coefficient > 0.1, and more than 420,000 SNP loci were finally retained. Step 3: Screen for Jiangshan Black-boned Chicken-specific SNPs by combining the attention weights output by the linear self-attention model with the fixed differentiation site score; Step 4: Using any SNP and any combination of SNPs as feature variables, use the XGBoost model to evaluate the probability of identifying it as Jiangshan Black-boned Chicken.

[0007] Preferably, in step three, screening for Jiangshan black-bone chicken-specific SNPs includes: SNP1 is located at position 8703393 on chromosome 3, with alleles C / A; SNP2 is located at position 86277759 on chromosome 3, with alleles A / G; SNP3 is located at position 14913140 on chromosome 13, with alleles T / G; SNP4 is located at position 646437 on chromosome 31, with an allele of C / T. SNP5 is located at position 1042461 on chromosome 33, with alleles T / C.

[0008] Preferably, the high-frequency alleles of Jiangshan Black-boned Chicken are: SNP1 is C, SNP2 is G, SNP3 is T, SNP4 is C, and SNP5 is T.

[0009] Preferably, in step three, the linear self-attention model includes a linear self-attention layer and a classification layer; A linear self-attention layer is used to automatically learn the importance weights of each SNP site, with a computational complexity of O(N). The classification layer is used to perform global mean pooling on the features output by the linear self-attention layer, and then outputs the binary classification probability through the fully connected layer.

[0010] Preferably, in step three, the training method for the linear self-attention model includes: Using purebred Jiangshan black-bone chickens as the target population and other chicken breeds as reference populations, based on whole-genome SNP data, the SNP genotype of each individual was encoded into a numerical vector and input into a linear self-attention model.

[0011] Preferably, in step three, the parameters of the linear self-attention model are set as follows: The cross-entropy loss function is used, and the Adam optimizer is used for training with a learning rate of 1e-3 until the loss value is less than 0.1.

[0012] Preferably, in step three, the steps for screening Jiangshan Black-boned Chicken-specific SNPs are as follows: In evaluation mode, the genotype tensor is input, the attention weights output by the linear self-attention layer are extracted, and the average attention weight of each SNP is calculated at the sample level. Calculate a fixed differentiation score for each SNP, which quantifies the degree of genotypic differentiation between the target population and the reference population; By combining attention weights and fixed differentiation scores for evaluation, the top 5 SNP loci were selected as the breed-specific SNP loci of Jiangshan Black-boned Chicken.

[0013] Preferably, in step four, the ratio of the training set to the test set in the XGBoost model is 80%:20%, wherein the ratio of Jiangshan black-bone chicken in the training set, other chicken breeds in the training set, Jiangshan black-bone chicken in the test set, and other chicken breeds in the test set is 8:8:2:2.

[0014] Preferably, the identification method screens out 5 Jiangshan black-bone chicken-specific SNP loci from whole-genome SNP data; the identification accuracy of a single SNP locus is 92.08%-100% as evaluated by the XGBoost machine learning model, the identification accuracy of any combination of 3 or more SNP loci is above 99.6%, and the identification accuracy of all 5 SNP loci combinations is 100%.

[0015] SNP molecular markers for the identification of Jiangshan black-bone chicken, including the SNPs used for screening Jiangshan black-bone chicken-specific SNPs as described above; Each SNP molecular marker includes: SNP1 is located at position 8703393 on chromosome 3, with alleles C / A; SNP2 is located at position 86277759 on chromosome 3, with alleles A / G; SNP3 is located at position 14913140 on chromosome 13, with alleles T / G; SNP4 is located at position 646437 on chromosome 31, with an allele of C / T. SNP5 is located at position 1042461 on chromosome 33, with alleles T / C.

[0016] The beneficial effects of this invention are as follows: This invention utilizes a linear self-attention deep learning algorithm to screen five Jiangshan Black-boned Chicken-specific SNP loci (SNP1-SNP5) from whole-genome SNP data. Evaluation using the XGBoost machine learning model showed that the identification accuracy for a single SNP locus reached 92.08%-100%, the identification accuracy for any combination of three or more SNP loci exceeded 99.6%, and the identification accuracy for all five SNP loci combinations was 100%. In a validation population comprising 499 individuals from 30 chicken breeds, this method reliably and accurately distinguished Jiangshan Black-boned Chicken from other breeds, demonstrating extremely high breed identification reliability.

[0017] This invention employs a linear self-attention mechanism instead of the traditional standard self-attention mechanism, reducing computational complexity from O(N²) to O(N) without requiring the construction of a high-dimensional attention matrix. This improvement significantly reduces memory usage and computation time when processing large-scale whole-genome SNP data (e.g., 420,000 SNP sites), exhibiting good scalability and practicality, and is suitable for high-throughput identification of large batches of samples in germplasm banks.

[0018] This invention innovatively integrates linear self-attention weights and fixed differentiation site scores to jointly screen for breed-specific SNP loci. The attention weights reflect the contribution of each SNP automatically learned by the model to the classification decision, while the fixed differentiation scores quantify the degree of genotypic differentiation between the target population and the reference population. This combination ensures the biological rationality of the screening results and enhances the interpretability of the model, providing a referable technical approach for the mining of specific molecular markers in other livestock and poultry breeds.

[0019] After screening for five specific SNP loci, this invention eliminates the need for whole-genome sequencing or complex deep learning model training. PCR is performed on the target population and other populations, followed by analysis using a trained XGBoost model to determine the probability of the breed being Jiangshan chicken. This method is simple to operate, low in cost, requires minimal experimental equipment, and is easily promoted and used in grassroots breeding farms, inspection and quarantine agencies, and market supervision departments. Attached Figure Description

[0020] The invention will now be further described with reference to the accompanying drawings.

[0021] Figure 1 This is a flowchart of an AI algorithm-based identification method for Jiangshan black-bone chicken according to the present invention. Detailed Implementation

[0022] To make the technical means, creative features, objectives and effects of this invention easier to understand, the invention will be further described below in conjunction with specific embodiments.

[0023] Example 1 like Figure 1 As shown, this embodiment aims to mine breed-specific SNP sites of Jiangshan Black-boned Chicken based on the whole genome SNP data of Jiangshan Black-boned Chicken using a linear self-attention deep learning algorithm. This enables efficient and accurate identification of Jiangshan Black-boned Chicken from other chicken breeds, providing technical support for germplasm identification, breed tracing, and intellectual property protection of Jiangshan Black-boned Chicken. The AI-based identification method for Jiangshan Black-boned Chicken described in this embodiment includes the following steps: Materials Methods Experimental population: The target population was purebred Jiangshan Black-boned Chicken (n=57), with a clear pedigree and no hybrid bloodlines. 1 mL of blood was collected from the venous vein under the wing and stored at -20℃.

[0024] Genome sequencing: Genomic DNA was extracted from blood using traditional methods. After quality control, whole-genome resequencing technology was used to sequence Jiangshan Silkie chickens, with a sequencing depth of no less than 10×. Resequencing data from 21 publicly available chicken breeds were integrated. Based on the reference genome GRCg7b, population SNP identification was performed using GATK software to obtain the original SNP dataset. After quality control based on allele frequency (MAF < 0.05) and deletion rate (missing > 0.2), over 11 million SNP loci were obtained. PLINK software was further used to filter out loci with linkage disequilibrium coefficients > 0.1 (--indep-pairwise 50 5 0.1), ultimately retaining 423,511 high-quality, linkage-disequilibrium-free SNP loci.

[0025] Data preprocessing: Construct a grouping file and record individual IDs, where Jiangshan black-bone chicken is marked as "target" and all other breeds are marked as "reference".

[0026] Linear Self-Attention Model: The linear self-attention model used in this embodiment reduces the computational complexity from O(N²) of the traditional standard self-attention model to O(N), eliminating the need to construct a high-dimensional attention matrix and making it suitable for large-scale SNP data processing. The specific structure includes two core modules: a linear self-attention layer responsible for automatically learning the importance weights of each SNP site. Its key design difference from traditional attention mechanisms lies in that it does not use three-dimensional matrix multiplication to calculate attention scores, but instead achieves linear complexity attention weight learning through feature projection and global mean calculation; and a classification layer that performs global mean pooling on the feature tensor output by the linear self-attention layer to obtain the global features of each sample (dimension: number of samples × 16). The linear classification layer (nn.Linear(16, 2)) outputs the classification probability of the sample (dimension: number of samples × 2), achieving binary classification between "Jiangshan Black-boned Chicken (target)" and "Other Chicken Breeds (reference)". The parameters are set as follows: Adam optimizer is used, learning rate is 1e-3, cross-entropy loss function (nn.CrossEntropyLoss()) is used to measure the model classification error, training is performed for 15 epochs, and the classification loss is calculated until the model converges (loss < 0.1).

[0027] Using the same SNP data, the training of the conventional method (standard self-attention model) and the model used in this invention (linear self-attention model) were compared. The specific performance test data is shown in Table 1 below: Table 1 Performance Tests of Standard Self-Attention Model and Linear Self-Attention Model It should be noted that the training time was calculated based on a specific server configuration with 1TB of memory and dual Intel Xeon Gold 6230R processors (52 cores / 104 threads, 2.1GHz). FLOPs represent floating-point computations.

[0028] Identification of Jiangshan Black-boned Chicken-specific SNPs: After the model training was completed, the attention weights output by the linear self-attention layer were combined with the fixed differentiation site scores to screen for Jiangshan Black-boned Chicken-specific SNPs.

[0029] The specific operations are as follows: 1. Attention weight extraction: After model training, in evaluation mode (torch.no_grad()), the genotype tensor is input, and the attention weights output by the linear self-attention layer are extracted. The mean of the weights for each SNP is calculated at the sample level to obtain the average attention weight for each SNP. The higher the weight value, the greater the contribution of the SNP to breed identification and the stronger its specificity. 2. Fixed differentiation site scoring: To further strengthen the loci that are "homozygous fixed in Jiangshan Black-boned Chicken and conversely homozygous in other chicken breeds", the fixed differentiation score of each SNP is calculated. This score quantifies the degree of genotypic differentiation between the target population and the reference population. The higher the score, the stronger the fixedness and specificity of the locus. Finally, the attention weights and fixed differentiation scores are combined for evaluation, and the top 5 SNP loci are selected as the breed-specific SNP loci for Jiangshan Black-boned Chicken.

[0030] This embodiment uses a linear self-attention deep learning algorithm to assign different weights to the SNP attention score and frequency distribution score to obtain Jiangshan Black-boned Chicken-specific SNPs. A total of 5 Jiangshan Black-boned Chicken SNP loci were obtained, as shown in Table 2. The 21 chicken breeds involved in this embodiment are 16 local chicken breeds, including Huiyang Bearded Chicken, Xishuangbanna Fighting Chicken, Dulong Chicken, Dehua Black Chicken, Turpan Fighting Chicken, Xingyi Dwarf Chicken, Black-boned Chicken, Shuanglian Chicken, Tianfu Broiler Chicken, Hetian Chicken, Chishui Black-boned Chicken, Lingkun Chicken, Wuhua Chicken, Xichuan Black-boned Chicken, Qiandongnan Small Fragrant Chicken, and Weining Chicken, as well as 5 foreign chicken breeds, including White Rock Chicken, White Leghorn Chicken, Loch Ness Red Chicken, Red Junglefowl, and Fayoumi Chicken.

[0031] Table 2. SNP locus statistics of Jiangshan black-bone chicken It should be explained that the five sequences in the sequence listing are the upstream and downstream sequences of the SNP1-SNP5 sites, used to determine primer positions and amplify fragments containing the target SNPs. This invention actually detects the genotype of each SNP site; each site can be used individually or in combination for the identification of Jiangshan Black-boned Chicken.

[0032] SNP1 corresponds to the first sequence in the sequence listing, namely GRCg7b reference genome 3:8702993-8703793, totaling 801 bp; SNP1 (3:8703393) is located at position 401 of this sequence. The upstream primer 5′-GCTGCTATCAGCACCCAGAT-3′ is located at positions 193-212 of the sequence, corresponding to genome 3:8703185-8703204; the downstream primer 5′-AGCGCATGGTATTGGTGAGT-3′ binds complementary to positions 599-618 of the sequence, corresponding to genome 3:8703591-8703610. The amplification range of the two primers is 3:8703185-8703610, with an expected product of 426 bp and the target site located at position 209 of the amplified product. The genotypic information of this locus can be used as one of the identification characteristics of Jiangshan Black-boned Chicken, and can be combined with the other 4 SNP loci for joint identification of Jiangshan Black-boned Chicken.

[0033] SNP2 corresponds to the second sequence in the sequence listing, namely GRCg7b reference genome 3:86277259-86278259, totaling 1001 bp; SNP2 (3:86277759) is located at position 501 of this sequence. The upstream primer 5′-GCAGTGTTGGGAATTGGACC-3′ is located at positions 297-316 of the sequence, corresponding to genome 3:86277555-86277574; the downstream primer 5′-CATGACCCAGGTGCAAGACT-3′ binds complementary to positions 737-756 of the sequence, corresponding to genome 3:86277995-86278014. The amplification range of the two primers is 3:86277555-86278014, with an expected product of 460 bp and the target site located at position 205 of the amplified product. The genotypic information of this locus can be used as one of the identification characteristics of Jiangshan Black-boned Chicken, and can be combined with the other 4 SNP loci for joint identification of Jiangshan Black-boned Chicken.

[0034] SNP3 corresponds to the third sequence in the sequence listing, namely GRCg7b reference genome 13:14912740-14913540, totaling 801 bp; SNP3 (13:14913140) is located at position 401 of this sequence. The upstream primer 5′-GACCCTGGGAAGAAAAGGGC-3′ is located at positions 115-134 of the sequence, corresponding to genome 13:14912854-14912873; the downstream primer 5′-TTCTGTGCCAGGAAGATGGG-3′ binds complementary to positions 571-590 of the sequence, corresponding to genome 13:14913310-14913329. The amplification range of the two primers is 13:14912854-14913329, with an expected product of 476 bp and the target site located at position 287 of the amplified product. The genotypic information of this locus can be used as one of the identification characteristics of Jiangshan Black-boned Chicken, and can be combined with the other 4 SNP loci for joint identification of Jiangshan Black-boned Chicken.

[0035] SNP4 corresponds to the 4th sequence in the sequence listing, namely GRCg7b reference genome 31:645937-646937, totaling 1001 bp; SNP4 (31:646437) is located at position 501 of this sequence. The upstream primer 5′-TGGGACCACTGTAATTCCTTGTT-3′ is located at positions 216-238 of the sequence, corresponding to genome 31:646152-646174; the downstream primer 5′-CTTTTCTCCACCCCCACATCTGTA-3′ binds complementary to positions 657-680 of the sequence, corresponding to genome 31:646593-646616. The amplification range of the two primers is 31:646152-646616, with an expected product of 465 bp, and the target site is located at position 286 of the amplified product. The genotypic information of this locus can be used as one of the identification characteristics of Jiangshan Black-boned Chicken, and can be combined with the other 4 SNP loci for joint identification of Jiangshan Black-boned Chicken.

[0036] SNP5 corresponds to the 5th sequence in the sequence listing, namely GRCg7b reference genome 33:1041961-1042961, totaling 1001 bp; SNP5 (33:1042461) is located at position 501 of this sequence. The upstream primer 5′-TTAACGTCCCCTCCAACCGA-3′ is located at positions 341-360 of the sequence, corresponding to genome 33:1042301-1042320; the downstream primer 5′-TCTCAGCCTGGTATTGTGCC-3′ binds complementary to positions 868-887 of the sequence, corresponding to genome 33:1042828-1042847. The amplification range of the two primers is 33:1042301-1042847, with an expected product of 547 bp, and the target site is located at position 161 of the amplified product. The genotypic information of this locus can be used as one of the identification characteristics of Jiangshan Black-boned Chicken, and can be combined with the other 4 SNP loci for joint identification of Jiangshan Black-boned Chicken.

[0037] Based on the sequencing results, the genotypes of five SNP loci were determined. The detection rates of different loci in 22 chicken breeds, including Jiangshan Silkie chicken, are shown in Table 3. Table 3. Detection frequency of Jiangshan Black-boned Chicken-specific alleles in different breeds It should be explained that: the high-frequency alleles of Jiangshan black-bone chicken in the third column refer to the higher frequency of the indicator allele at the SNP locus in Jiangshan black-bone chicken, and the specific genotype can be homozygous or heterozygous carrying the allele; the detection frequency is the frequency of the indicator allele in the third column in different breeds, and the detection frequency of non-genotypes.

[0038] The analysis of the table above shows that the five SNP loci have significant frequency differences between Jiangshan Black-boned Chicken and other breeds. Jiangshan Black-boned Chicken has high frequency alleles at these five SNP loci, with a detection frequency of 76.23%-94.74%. Among 21 breeds including Black-boned Chicken, the detection frequency of SNP1, SNP2, and SNP3 does not exceed 15.00%. The detection frequency of SNP4 is relatively high in Dulong Chicken and Lingkun Chicken, ranging from 29.17% to 31.25%, while the detection frequency does not exceed 10% in other breeds. The frequency of SNP5 is relatively high in White Leghorn Chicken and White Locker Chicken, ranging from 20.31% to 32.69%, while the detection frequency in other breeds does not exceed 15%.

[0039] Example 2 To further verify the breed identification efficiency of high-frequency loci of Jiangshan black-bone chicken, this embodiment supplements several breeds, such as Lushi chicken, which are different from those in Embodiment 1, for preliminary verification of detection frequency.

[0040] Materials Methods Experimental population: In this embodiment, blood was collected from the subwing veins of Lushi chickens, Xiushui yellow chickens, Daweishan miniature chickens, Yao chickens, Luxi fighting chickens, Xiaoshan chickens, recessive white rock chickens, and commercial broiler chickens. The blood was placed in anticoagulant tubes, shaken well, and stored at -20°C.

[0041] DNA extraction and PCR amplification: DNA was extracted from 141 blood samples of 8 chicken breeds, including Lushi chicken, using the phenol-chloroform method. Specific amplification primers for different loci were designed using Oligo software. Using the DNA as a template, SNP1, SNP2, SNP3, SNP4, and SNP5 were amplified by PCR and sequenced, and gene frequencies were statistically analyzed. The PCR reaction mixture consisted of: 1.5 μl template DNA, 1.0 μl 10 pmol / μl upstream primer, 1.0 μl 10 pmol / μl downstream primer, 12.5 μl 2*Master mix, and 9.0 μl ddH2O. The PCR reaction conditions were as follows: pre-denaturation temperature 94℃, time 10 min; denaturation temperature 94℃, time 30 s; annealing temperature 60℃, time 30 s; extension temperature 72℃, time 50 s; for a total of 32 cycles; final extension temperature 72℃, time 5 min.

[0042] This embodiment uses PCR amplification to verify the detection frequency of the specific loci detected in Example 1 in a supplementary population. Based on the sequencing results, the genotype and allele frequencies of the five SNP loci in different breeds were statistically analyzed. The detection rates of different loci in eight chicken breeds, including Lushi chicken, are shown in Table 4. Table 4. Detection frequency of Jiangshan Black-boned Chicken-specific alleles in supplementary breeds. It should be explained that: the high-frequency alleles of Jiangshan black-bone chicken in the third column refer to the higher frequency of the indicator allele at the SNP locus in Jiangshan black-bone chicken, and the specific genotype can be homozygous or heterozygous carrying the allele; the detection frequency is the frequency of the indicator allele in the third column in different breeds, and the detection frequency of non-genotypes.

[0043] The analysis of the table above shows that the five SNP loci have significant frequency differences between Jiangshan Black-boned Chicken and other breeds. The detection frequency of the five SNP loci in Jiangshan Black-boned Chicken is 76.32%-94.74%. Among the eight breeds including Lushi Chicken, the detection frequencies of SNP1, SNP2, SNP3 and SNP5 are less than 10%. SNP4 has a relatively high detection frequency in Lushi Chicken and Xiushui Yellow Chicken, ranging from 20.00% to 25.00%, but it still differs significantly from the allele frequency of Jiangshan Black-boned Chicken.

[0044] Example 3 Examples 1 and 2 have identified five specific alleles that are present at high frequencies in Jiangshan Black-boned Chicken. In this example, machine learning algorithms are used to verify different loci and combinations to evaluate the accuracy of Jiangshan Black-boned Chicken identification.

[0045] Materials and methods: Experimental population: 499 chickens of 30 breeds, including Jiangshan black-bone chicken, involved in Examples 1 and 2.

[0046] Data preprocessing: The three genotypes of the five SNPs were encoded using 0, 1, and 2 respectively, where 0 represents the wild-type homozygous, 1 represents the heterozygous, and 2 represents the mutant homozygous. The `isJS` column was defined as the target variable, with 1 representing Jiangshan Silkie chicken and 0 representing other chicken breeds. The data was then processed to form the input file.

[0047] Machine learning methods were used to evaluate breed identification accuracy: In this embodiment, the XGBoost model was used for training and testing, with a training set to test set ratio of 80%:20%. Due to the significant difference in the number of Jiangshan Black-boned Chickens (n=57) and other breeds (n=442), a stratified sampling method was adopted. 80% of the Jiangshan Black-boned Chickens and 80% of the other breeds were selected for the training set, and 20% of the Jiangshan Black-boned Chickens and 20% of the other breeds were selected for the test set to avoid sampling errors. The optimal hyperparameters were determined using an autotuner on the training set, and after updating the hyperparameters, the breed identification accuracy was evaluated on the test set. This embodiment used any SNP and any combination of SNPs as feature variables, totaling 31 SNPs and SNP combinations. The XGBoost model was used to evaluate the probability of identifying the chicken as a Jiangshan Black-boned Chicken. The logic for using the XGBoost machine learning algorithm to determine if a chicken is a Jiangshan Black-boned Chicken is as follows: For the five candidate SNP loci for identifying Jiangshan Black-boned Chicken, all SNP combinations (31 in total) were generated. First, stratified sampling was performed between Jiangshan Black-boned Chicken and other breeds, dividing the dataset into training and testing sets in an 8:2 ratio. The training set contained 80% of the total number of Jiangshan Black-boned Chickens (n ​​= 46) and 80% of the total number of other breeds (n = 353); the testing set contained 20% of the total number of Jiangshan Black-boned Chickens (n ​​= 11) and 20% of the total number of other breeds (n = 89), ensuring consistent distribution between the two classes. For each SNP combination, an XGBoost binary classification task was constructed. Hyperparameter optimization was performed using grid search combined with 10-fold cross-validation. After training, the classification accuracy was calculated on the testing set, and the final identification accuracy for all combinations was output.

[0048] result Based on any SNP and any combination of SNPs, a total of 31 groups were identified, and the XGBoost model was used to evaluate the breed identification accuracy of Jiangshan Black-boned Chicken. As the number of SNPs used increased from 1 to 5, the average breed identification accuracy gradually increased, from 0.9663 to 1. When using 3 or more SNPs, the identification accuracy reached 99.6%, meaning there was a 99.6% probability that the group could be identified as Jiangshan Black-boned Chicken. The calculation results are shown in Table 5. Table 5. Accuracy of different SNP loci combinations in identifying Jiangshan Black-boned Chicken It should be explained that the average accuracy is the average accuracy under the same number of loci but different combinations of loci.

[0049] Example 4 This embodiment provides a detection kit for identifying Jiangshan black-boned chicken, used to detect specific SNP molecular markers. The components of the kit are shown in Table 6. Table 6. Components of the Reagent Kit The foregoing has shown and described the basic principles, main features, and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The embodiments and descriptions in the specification are merely illustrative of the principles of the invention. Various changes and modifications can be made to the invention without departing from its spirit and scope, and all such changes and modifications fall within the scope of the present invention as claimed. The scope of protection of the present invention is defined by the appended claims and their equivalents.

Claims

1. A method for identifying Jiangshan black-boned chicken based on AI algorithms, characterized in that, Includes the following steps: Step 1: Based on the reference genome GRCg7b, use GATK software to identify population SNPs and obtain the original SNP dataset; Step 2: After quality control based on allele frequency MAF < 0.05 and deletion rate missing > 0.2, more than 11 million SNP loci were obtained; Using PLINK software, sites with linkage disequilibrium coefficients >0.1 were filtered out, ultimately retaining more than 420,000 SNP sites; Step 3: Screen for Jiangshan Black-boned Chicken-specific SNPs by combining the attention weights output by the linear self-attention model with the fixed differentiation site score; Step 4: Using any SNP and any combination of SNPs as feature variables, use the XGBoost model to evaluate the probability of identifying it as Jiangshan Black-boned Chicken.

2. The method for identifying Jiangshan Black-boned Chicken based on AI algorithm according to claim 1, characterized in that, In step three, screening for Jiangshan black-bone chicken-specific SNPs includes: SNP1 is located at position 8703393 on chromosome 3, with alleles C / A; SNP2 is located at position 86277759 on chromosome 3, with alleles A / G; SNP3 is located at position 14913140 on chromosome 13, with alleles T / G; SNP4 is located at position 646437 on chromosome 31, with an allele of C / T. SNP5 is located at position 1042461 on chromosome 33, with alleles T / C.

3. The method for identifying Jiangshan Black-boned Chicken based on AI algorithm according to claim 2, characterized in that, The high-frequency alleles of Jiangshan Black-boned Chicken are: SNP1 is C, SNP2 is G, SNP3 is T, SNP4 is C, and SNP5 is T.

4. The method for identifying Jiangshan Black-boned Chicken based on AI algorithm according to claim 1, characterized in that, In step three, the linear self-attention model includes a linear self-attention layer and a classification layer; A linear self-attention layer is used to automatically learn the importance weights of each SNP site, with a computational complexity of O(N). The classification layer is used to perform global mean pooling on the features output by the linear self-attention layer, and then outputs the binary classification probability through the fully connected layer.

5. The method for identifying Jiangshan Black-boned Chicken based on AI algorithm according to claim 1, characterized in that, In step three, the training methods for the linear self-attention model include: Using purebred Jiangshan black-bone chickens as the target population and other chicken breeds as reference populations, based on whole-genome SNP data, the SNP genotype of each individual was encoded into a numerical vector and input into a linear self-attention model.

6. The method for identifying Jiangshan Black-boned Chicken based on AI algorithm according to claim 1, characterized in that, In step three, the parameters of the linear self-attention model are set as follows: The cross-entropy loss function is used, and the Adam optimizer is used for training with a learning rate of 1e-3 until the loss value is less than 0.

1.

7. The method for identifying Jiangshan Black-boned Chicken based on AI algorithm according to claim 1, characterized in that, In step three, the steps for screening Jiangshan Black-boned Chicken-specific SNPs are as follows: In evaluation mode, the genotype tensor is input, the attention weights output by the linear self-attention layer are extracted, and the average attention weight of each SNP is calculated at the sample level. Calculate a fixed differentiation score for each SNP, which quantifies the degree of genotypic differentiation between the target population and the reference population; By combining attention weights and fixed differentiation scores, the top 5 SNP loci were selected as the breed-specific SNP loci of Jiangshan Black-boned Chicken.

8. The method for identifying Jiangshan Black-boned Chicken based on AI algorithm according to claim 1, characterized in that, In step four, the ratio of training set to test set in the XGBoost model is 80%:20%. The ratio of Jiangshan black-bone chicken, other chicken breeds in the training set, Jiangshan black-bone chicken and other chicken breeds in the test set is 8:8:2:

2.

9. The method for identifying Jiangshan Black-boned Chicken based on AI algorithm according to claim 1, characterized in that, The identification method screened out 5 Jiangshan black-bone chicken-specific SNP loci from whole-genome SNP data. The XGBoost machine learning model evaluated the identification accuracy of a single SNP locus, which was 92.08%-100%. The identification accuracy of any combination of 3 or more SNP loci was over 99.6%, and the identification accuracy of all 5 SNP loci combinations was 100%.

10. SNP molecular markers for the identification of Jiangshan black-boned chicken, characterized in that, SNP molecular markers include the specific SNPs for screening Jiangshan black-bone chicken as described in claim 1; Each SNP molecular marker includes: SNP1 is located at position 8703393 on chromosome 3, with alleles C / A; SNP2 is located at position 86277759 on chromosome 3, with alleles A / G; SNP3 is located at position 14913140 on chromosome 13, with alleles T / G; SNP4 is located at position 646437 on chromosome 31, with an allele of C / T. SNP5 is located at position 1042461 on chromosome 33, with alleles T / C.

Images