Artificial intelligence and high-throughput sequencing-based accurate geographical traceability method for pangolin scales and nail tips thereof and application of accurate geographical traceability method

Abstract

The invention discloses a manis pentadactyla and manis pentadactyla nail tip accurate geographical traceability method based on artificial intelligence and high-throughput sequencing and application of the manis pentadactyla and manis pentadactyla nail tip accurate geographical traceability method. The method comprises the steps that a machine learning method is applied to conduct geographical genetic structure deduction on samples with known geographical coordinates, and a deep neural network prediction model is established; and tracing the sample with unknown geographic coordinates according to the deep neural network model. According to the manis pentadactyla and manis pentadactyla nail tip accurate geographical tracing method based on artificial intelligence and high-throughput sequencing and the application thereof, a large number of genetic markers obtained through original sequencing are simplified into a group of simplified features which cannot be directly compared through dimension reduction, and each feature contains geographical source information of individuals in a sample; on the basis, a deep neural network model is established through simplified feature training, so that the training cost is greatly reduced; and mapping the principal component with the maximum variance to the geographic coordinates of the individual, and achieving the effect of tracing the geographic origin of the individual through a deep neural network model.

Description

Technical Field

[0001] The present invention relates to the technical field of accurate geographical tracing of pangolins and their scales, and in particular to an accurate geographical tracing method of pangolins and their scales based on artificial intelligence and high-throughput sequencing and an application thereof. Background Art

[0002] Traditional methods for tracing the origin of pangolins and their products mainly include morphological identification, stable isotope analysis, haplotype analysis and DNA barcoding technology. These methods can provide information on the source population of pangolins to a certain extent, but they have problems such as low accuracy, complex operation and high cost. Morphological identification is easily affected by individual differences and environmental adaptability, while stable isotope analysis is limited by the quality and availability of samples. Haplotypes can only determine the population affiliation of samples based on existing population background information. Although DNA barcoding technology has certain advantages in species identification, it is still insufficient in terms of accurate geographical tracing. Summary of the invention

[0003] To solve the above problems, the present invention provides a method for precise geographical tracing of pangolins and their scales based on artificial intelligence and high-throughput sequencing and its application. The method uses machine learning methods to deduce the geographical genetic structure of samples with known geographical coordinates and establish a deep neural network prediction model. The method can process a large amount of genetic marker data and convert the raw data into a set of features containing geographical origin information through dimensionality reduction technology, thereby achieving precise tracing of unknown samples.

[0004] To achieve the above objectives, the present invention provides a method for accurate geographical tracing of pangolins and their scales based on artificial intelligence and high-throughput sequencing, which uses KLFDAPC to extract the genomic geographical genetic structure of pangolins or their scales samples with known geographical coordinates and establish a deep neural network model, and uses the deep neural network model to perform geographical tracing of pangolins or their scales with unknown geographical locations; The establishment of a deep neural network model includes the following steps: S1. Perform whole genome resequencing on pangolin or scale samples to obtain whole genome resequencing group sequences of the samples; S2, aligning the whole genome resequencing sequence obtained by sequencing in step S1 to the reference genome and detecting mutations, merging the mutation detection files of each sample and setting parameter filtering to obtain high-quality mutation sites; S3, using 012 coding genotypes, performing KLFDAPC analysis on the individual × genotype matrix to obtain the geographic genetic structure of pangolins or their scale samples, and building a deep neural network model based on the background information of pangolins or their scale samples; S4. The five-fold cross-validation method was used to test the prediction performance of the deep neural network model, and the automatic parameter optimization method was used to optimize the model parameters to obtain the optimal model.

[0005] Preferably, during the sequencing process of step S1, tissues that are difficult to grind and have low nucleic acid content are removed from the sample, and after the tissues are removed, they are quickly frozen in liquid nitrogen, stored at -80°C, and transported on dry ice. Repeated freezing and thawing should be avoided during sample delivery.

[0006] Preferably, during the filtering process of step S2, filtering is performed with minimum allele frequency MAF>0.05 and deletion rate ≤0.2.

[0007] Preferably, in step S3, 0, 1, and 2 represent homozygous, heterozygous, and homozygous minor alleles of the locus, respectively.

[0008] Preferably, in step S3, the specific steps of determining the geographic genetic structure of pangolins or their scale samples and constructing a deep neural network include: S31, use PCA to reduce the dimension of the individual × genotype matrix to obtain the principal components PCs containing the geographical location information in the genotype; S32, KLFDA analysis is performed on the first 20 to 50 standardized PCs to further extract the nonlinear characteristics in the principal components. At this time, the principal component KLFDAPCs not only contains the decisive geographical location information, but also retains the linear and nonlinear associations between samples; S33, use Purkinje analysis to calculate the correlation between the first two principal components KLFDAPC1 and KLFDAPC2 and the geographic coordinates of the sampling points, compare and calculate the geometric similarity between KLFDAPCs and the real geographic locations of the samples through Purkinje transformation, and draw the first two principal component axes after rotation; S34. Perform linear regression on KLFDAPC1 and KLFDAPC2 with the longitude and latitude of the samples to establish the corresponding relationship between the dimensionality reduction features and the longitude and latitude, that is, the deep neural network model.

[0009] Preferably, in step S4, the first 10 dimension reduction features of KLFDAPC are regressed with the geographic coordinates of the sample, and the prediction performance of the deep neural network model is tested using a five-fold cross validation method.

[0010] Preferably, the process of tracing the source of a sample with unknown geographic coordinates includes: The KLFDAPC1 and KLFDAPC2 of the unknown geographic coordinate samples are input into the deep neural network model. The output of the decoded deep neural network model is the geographic coordinates corresponding to KLFDAPC1 and KLFDAPC2. The geographic coordinates can be mapped to the map for geographic tracing.

[0011] The present invention provides a method for accurate geographical tracing of pangolins and their scales based on artificial intelligence and high-throughput sequencing and its application, which has the following beneficial effects: (1) Through dimensionality reduction, the large number of genetic markers obtained from the original sequencing are simplified into a set of genetic characteristics that cannot be directly compared. At the same time, each characteristic contains the geographical origin information of the individuals in the sample; (2) Establishing a deep neural network model through genetic feature training greatly reduces the training cost; (3) Map the genetic characteristics with the largest variance to the individual's geographic coordinates, and use a deep neural network model to trace the individual's geographic origin.

[0012] The technical solution of the present invention is further described in detail below through the accompanying drawings and embodiments. BRIEF DESCRIPTION OF THE DRAWINGS

[0013] Figure 1 This is a flow chart of a method for accurate geographical tracing of pangolins and their scales based on artificial intelligence and high-throughput sequencing according to the present invention; Figure 2 is the correlation between KLFDAPC1-2 and longitude and latitude after Pu-type transformation; where a is the relationship between KLFDAPC1 and longitude on the projection map and its gradient distribution on the geographic coordinates, and b is the relationship between KLFDAPC2 and latitude on the projection map and its gradient distribution on the geographic coordinates; Figure 3 is the performance of the model, that is, the correlation between the predicted longitude and latitude and the true longitude and latitude; where a is the correlation between the sample longitude predicted by the neural network model and the true longitude value; b is the correlation between the sample latitude predicted by the neural network model and the true latitude value. DETAILED DESCRIPTION

[0014] In order to make the purpose, technical scheme and advantages disclosed in the embodiments of the present invention more clearly understood, the embodiments of the present invention are further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the embodiments of the present invention and are not intended to limit the embodiments of the present invention. Based on the embodiments in this application, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of this application.

[0015] Embodiment 1 A method for accurate geographical tracing of pangolins and their scales based on artificial intelligence and high-throughput sequencing is used to infer the geographical origin of pangolins. In this embodiment, a total of 110 pangolin samples were used for sequencing.

[0016] The 110 sample collections were conducted with permission from the forestry department and in compliance with China's National Wildlife Protection Law.

[0017] The flowchart of the geographical tracing method is as follows Figure 1 As shown, the specific steps include: S1. KLFDAPC was used to extract the genetic features of 110 pangolin samples and establish a deep neural network model. The establishment process is as follows: S11. First, sequence the pangolin samples and remove tissues that are difficult to grind and have low nucleic acid content. After the tissues are removed, quickly freeze them in liquid nitrogen, store them at -80℃, and transport them on dry ice. Repeated freezing and thawing should be avoided during the process.

[0018] S12. Align the whole gene resequencing group sequence of the sequenced sample to the reference genome and detect SNP single nucleotide polymorphism variation. Merge the variation detection files of each sample and filter them with the minimum allele frequency <0.05 and the missing rate <0.2 to obtain high-quality SNP sites.

[0019] S13, using 012 to encode genotypes, representing homozygous, heterozygous and homozygous minor alleles of the locus, to obtain the resequencing genotype matrix, such as Figure 1 As shown in a. KLFDAPC analysis was performed on the resequencing genotype matrix to obtain the geographical genetic structure, as shown in Figure 1 As shown in b. Using the background information of the sample, a deep neural network is constructed to decode and parse the geographic information of the sample, as shown in Figure 1 As shown in c. Specifically include: S131. Use PCA to reduce the dimension of the resequencing genotype matrix to obtain principal components PCs containing geographic location information in the genotype.

[0020] S132. Perform KLFDA analysis on the first 20 to 50 standardized PCs to further extract nonlinear features in the principal components. At this time, the principal components (KLFDAPCs) not only contain the decisive geographic location information, but also retain the linear and nonlinear associations between samples.

[0021] S133. Pursuit analysis was used to calculate the correlation between the first two principal components (KLFDAPC1 and KLFDAPC2) and the geographic coordinates of the sampling points. Pursuit transformation was used to compare and calculate the geometric similarity between KLFDAPCs and the true geographic locations of the samples, and the first two principal component axes after rotation were plotted.

[0022] S134, perform linear regression on KLFDAPC1 and KLFDAPC2 with the sample longitude and latitude, establish the corresponding relationship between the dimensionality reduction features and the longitude and latitude, and thus realize the mapping of PC on the real map.

[0023] Figure 2The linear correlation between the KLFDAPCs after the Pu-type transformation and the geographic longitude and latitude coordinates is shown, Y=aX+b, Y is the longitude and latitude coordinates, and X is the KLFDAPC genetic characteristic value. The correlation coefficients (R 2 ) reached 42% and 89% respectively. The results are as follows Figure 2 As shown in a and b, a is longitude and b is latitude. This shows that the principal components transformed by the Pu-type transformation have a high geometric correlation with longitude and latitude, and also have a certain linear correlation. The Pu-type similarity reaches 66%. Considering that the actual level of traceability is generally carried out at the large-scale level of the country or region, a Pu-type similarity greater than 50% can effectively ensure the application of traceability in actual scenarios.

[0024] The above results show that the geographic location information contained in the principal components can be used to trace the geographic origin of wild animals.

[0025] S14. By constructing a multilayer perceptron neural network model, the first 10 dimensionality reduction features of KLFDAPC (geographic genetic structure) were regressed with the geographic coordinates (latitude and longitude) of the samples, and the prediction performance of the deep neural network model was obtained using the five-fold cross-validation method.

[0026] The geographical genetic structure dataset of pangolins obtained through KLFDPAC was randomly and evenly divided into five subsets. Four of the subsets were selected as training sets (80% training sets) and the remaining subset was selected as the test set (20% test set). The neural network model was trained and evaluated on the test set. This process was repeated 5 times to ensure that each subset was used as a test set once. For each test, the correlation coefficient R between the predicted value and the true value of the geographical coordinates was used. 2 , to evaluate the performance of the model. Finally, the performance measurement results of the five tests were averaged to obtain the final performance evaluation of the model. In this process, the automatic parameter optimization method was used to optimize the model parameters to obtain the optimal model.

[0027] The performance results of the deep neural network model are as follows Figure 3 As shown. Among them, a is the correlation between the sample longitude predicted by the neural network model and the true longitude value; b is the correlation between the sample latitude predicted by the neural network model and the true latitude value.

[0028] The results obtained by resampling statistics show that the machine learning model constructed by deep neural network and KLFDAPC features can achieve R in predicting the true geographic coordinates of samples. 2 >85% accuracy.

[0029] In particular, for the KLFDAPCs results, the geometric similarity between the first two dimensionally reduced features and the geographic coordinates reaches 0.66, which is higher than 0.63 of the PCs method.

[0030] S2. Apply the deep neural network model trained above for verification.

[0031] The model constructed by the above KLFDAPC results covers the geographic genetic structure of the population sampling site. This process is to input the whole genome resequencing data of the samples with known background information into the KLFDAPC model, and the geographic genetic structure generated by the model is transformed and linearly projected onto the map, that is, the geographic location information of each sample is presented.

[0032] For unknown samples obtained, after whole genome resequencing, the sequencing data is input into the KLFDAPC model to predict the geographical genetic characteristic value. The output value of the genetic characteristic decoding deep neural network model is the geographical coordinates corresponding to the genetic characteristic. It is mapped on the map for tracing. For example, 10 individuals from different sources were obtained to obtain their whole genome resequencing data, and the 10 resequencing data were input into KLFDAPC to obtain the predicted genetic geographical genetic characteristic value. This characteristic value was predicted by the previously constructed linear model to obtain the individual's geographical coordinates. The predicted longitude and latitude of the unknown sample and the actual longitude and latitude are shown in Table 1 below: Table 1 Predicted longitude and latitude of unknown samples and actual longitude and latitude ; Comparing the predicted longitude and latitude with the general geographical origin of the examples, it can be found that they are not much different.

[0033] Finally, it should be noted that the above embodiments are only used to illustrate the technical solution of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that they can still modify or replace the technical solution of the present invention with equivalents, and these modifications or equivalent replacements cannot cause the modified technical solution to deviate from the spirit and scope of the technical solution of the present invention.

Claims

1. A method for accurate geographical tracing of pangolins and their scales based on artificial intelligence and high-throughput sequencing, characterized in that: KLFDAPC was used to extract the genomic geographic genetic structure of pangolins or their scales with known geographic coordinates and to establish a deep neural network model. The deep neural network model was used to trace the geographical origin of pangolins or their scales with unknown geographic locations. The establishment of a deep neural network model includes the following steps: S1. Perform whole genome resequencing on pangolin or scale samples to obtain whole genome resequencing group sequences of the samples; S2, aligning the whole genome resequencing sequence obtained by sequencing in step S1 to the reference genome and detecting mutations, merging the mutation detection files of each sample and setting parameter filtering to obtain high-quality mutation sites; S3, using 012 coding genotypes, performing KLFDAPC analysis on the individual × genotype matrix to obtain the geographic genetic structure of pangolins or their scale samples, and building a deep neural network model based on the background information of pangolins or their scale samples; S4. The five-fold cross-validation method was used to test the prediction performance of the deep neural network model, and the automatic parameter optimization method was used to optimize the model parameters to obtain the optimal model.

2. The method for accurate geographical tracing of pangolins and their scales based on artificial intelligence and high-throughput sequencing according to claim 1, characterized in that: During the sequencing process of step S1, tissues that are difficult to grind and have low nucleic acid content are removed from the sample. After the tissue is removed, it is quickly frozen in liquid nitrogen, stored at -80°C, and transported on dry ice. Repeated freezing and thawing should be avoided during sample delivery.

3. The method for accurate geographical tracing of pangolins and their scales based on artificial intelligence and high-throughput sequencing according to claim 1, characterized in that: In the filtering process of step S2, filtering was performed with minimum allele frequency MAF>0.05 and missing rate ≤0.

2.

4. The method for accurate geographical tracing of pangolins and their scales based on artificial intelligence and high-throughput sequencing according to claim 1, characterized in that: In step S3, 0, 1, and 2 represent homozygous, heterozygous, and homozygous minor alleles of the locus, respectively.

5. The method for accurate geographical tracing of pangolins and their scales based on artificial intelligence and high-throughput sequencing according to claim 1, characterized in that: In step S3, the specific steps of determining the geographic genetic structure of pangolin or its scale samples and constructing a deep neural network include: S31, use PCA to reduce the dimension of the individual × genotype matrix to obtain the principal components PCs containing the geographical location information in the genotype; S32, perform KLFDA analysis on the first 20 to 50 standardized PCs to further extract the nonlinear characteristics in the principal components. At this time, the principal component KLFDAPCs not only contains the decisive geographical location information, but also retains the linear and nonlinear associations between samples; S33, use Purkinje analysis to calculate the correlation between the first two principal components KLFDAPC1 and KLFDAPC2 and the geographic coordinates of the sampling points, compare and calculate the geometric similarity between KLFDAPCs and the real geographic locations of the samples through Purkinje transformation, and draw the first two principal component axes after rotation; S34. Perform linear regression on KLFDAPC1 and KLFDAPC2 with the longitude and latitude of the samples to establish the corresponding relationship between the dimensionality reduction features and the longitude and latitude, that is, the deep neural network model.

6. The method for accurate geographical tracing of pangolins and their scales based on artificial intelligence and high-throughput sequencing according to claim 5, characterized in that: In step S4, the first 10 dimension reduction features of KLFDAPC are regressed with the geographic coordinates of the samples, and the prediction performance of the deep neural network model is tested using the five-fold cross validation method.

7. The method for accurate geographical tracing of pangolins and their scales based on artificial intelligence and high-throughput sequencing according to claim 1, characterized in that: The process of tracing the origin of samples with unknown geographic coordinates includes: The KLFDAPC1 and KLFDAPC2 of the unknown geographic coordinate samples are input into the deep neural network model. The output of the decoded deep neural network model is the geographic coordinates corresponding to KLFDAPC1 and KLFDAPC2. The geographic coordinates can be mapped to the map for geographic tracing.

8. An application of the method for accurate geographical tracing of pangolins and their scales based on artificial intelligence and high-throughput sequencing as claimed in any one of claims 1 to 7, characterized in that: Apply KLFDAPC to the geographical traceability of endangered wild animals and their products, with pangolins as a typical example.

Images