QTL sample processing, model training, identification methods, devices and equipment

Abstract

The present application relates to the field of QTL identification technology, and in particular to QTL sample processing, model training, identification methods, devices and equipment. The QTL sample processing method is to sequence and identify a DNA pool constructed by sorting the QTL positioning population according to the target trait phenotype, obtain the SNP data and sample markers therein, and train the RU-net model in sequence to obtain a QTL identification model. The identification model can effectively identify corn plant height QTL, rice plant height QTL, rice flowering period QTL, Wuchang fish intermuscular thorn QTL, etc., and the identification results have lower deviation and signal-to-noise ratio than ΔSNP-index, ED4, G', SmoothLOD and Ridit algorithms, and can effectively identify and identify micro-effect sites with a phenotypic explanation rate as low as 5%.

Description

Technical Field

[0001] The present application relates to the field of QTL identification technology, and in particular to QTL sample processing, model training, identification methods, devices and equipment. Background Art

[0002] QTL stands for quantitative trait locus, which refers to the location in the genome of a gene that controls a quantitative trait. QTL mapping requires the use of genetic markers. By looking for links between genetic markers and the quantitative trait of interest, one or more QTLs can be located next to a genetic marker on the same chromosome. In other words, the marker and QTL are linked.

[0003] Bulked segregant analysis, proposed by Michelmore et al. in 1991, has proven to be an effective method for QTL analysis. To date, various algorithms have been developed for bulked segregant analysis, including the ΔSNP-index method (Takagi et al., 2013), which uses the difference in allele frequencies between high and low pools; the ED4 method (Hill et al., 2013), which uses Euclidean distance; the G' method (Magwene et al., 2011), which uses G-value calculations; the SmoothLOD method (Zhang et al., 2019), which uses LOD calculations; and the Ridit method (Wang et al., 2019), which uses nonparametric tests. With the exception of Ridit analysis, which can be applied to three or more bulked pools, the remaining algorithms are limited to analysis and resolution of two bulked pools. Furthermore, these methods rely on theoretical knowledge to design formulas for detecting loci for target traits, making it difficult to detect loci with minor effects in complex traits and contexts. Summary of the Invention

[0004] In response to the shortcomings of existing algorithms in detecting minor effect sites, the inventors of this application have developed a sample processing method, model training method, QTL identification method, device and equipment for QTL identification model training. The sample processing method sorts, groups and mixes the DNA information of the QTL mapping population to obtain a DNA pool, and uses this to obtain SNP data and sample markers, and uses this to construct training samples to train the model. The resulting QTL identification model can characterize the response relationship between the SNP data of the QTL mapping population and the QTL of the mapping population. The trained QTL identification model can effectively identify maize plant height QTL, rice plant height QTL, rice flowering period QTL, Wuchang fish intermuscular thorn QTL, etc., and the identification results have lower bias and signal-to-noise ratio than the ΔSNP-index, ED4, G', SmoothLOD and Ridit algorithms, and can effectively identify and recognize minor effect sites with a phenotypic explanation rate as low as 5%.

[0005] To this end, in a first aspect, embodiments of the present application disclose a sample processing method for QTL identification model training, which includes:

[0006] Construction of QTL mapping population;

[0007] sorting and grouping individuals in the mapping population according to the phenotype of the QTL mapping population to obtain a plurality of sorted groups;

[0008] Individual DNA samples within each group were mixed to obtain multiple sequenced DNA pools;

[0009] performing fragmentation, sequencing, SNP identification and calculation on the multiple DNA pools to obtain SNP data of the multiple pools;

[0010] The SNP data are marked respectively to obtain the SNP data and sample markers thereof, which are used as samples for training the QTL identification model.

[0011] In certain embodiments, the SNP data includes SNP location information and SNP frequency information, where the SNP frequency is the frequency of occurrence of the SNP in each pooled DNA fragment. The method of labeling the SNP data includes:

[0012] The SNP data with continuously increasing or decreasing SNP frequencies in multiple sequenced DNA pools are marked as 1, otherwise they are marked as 0.

[0013] In certain embodiments, the sample processing method includes a step of filtering the SNP data to remove low-quality SNP sites; the low-quality SNP sites have at least one of the following characteristics:

[0014] The corresponding number of reads from the DNA pool sequencing is lower than the threshold;

[0015] The SNP frequencies in the DNA pool deviate significantly from 0.5 in the same direction; and

[0016] The allele frequency difference between a certain SNP site and the adjacent SNP is greater than 0.1.

[0017] In a second aspect, the present application discloses a training method for a QTL identification model, which includes:

[0018] Obtaining a training set, the training set including a plurality of training samples obtained according to the sample processing method, each of the training samples including SNP data and a sample marker of the SNP data, the SNP data being obtained by sequencing a mixed pool of samples from a QTL mapping population;

[0019] The two-dimensional tensor composed of the sample labels is input as the input layer into the residual U-net model. The backpropagation algorithm and the stochastic gradient descent method are used to iteratively update the weights of each layer according to the size of the forward propagation loss value. When the loss value of the model tends to converge, the training is stopped to obtain the QTL recognition model.

[0020] In certain embodiments, the trained QTL identification model characterizes the response relationship between the SNP data of the QTL mapping population and the QTL of the mapping population.

[0021] In some embodiments, the residual U-net model consists of an input layer, an encoder, a decoder, and an output layer; each input of the input layer is a two-dimensional tensor consisting of 64 sites of the sample label, and the encoder is composed of a convolutional neural network and a residual network. After four downsamplings, deep information is obtained, and then after four upsamplings, the deep information is converted into shallow information, and finally passed through a convolution layer with a 1×1 convolution kernel and a sigmoid activation function; the encoder and decoder are both composed of three convolution layers, with convolution kernels of 1×1, 3×3, and 3×3, respectively, and downsampling is completed by maximum pooling, and each upconvolution is followed by a convolution layer; the residual connection acts after every three global convolutions, and there is also a residual connection in the corresponding layers between the encoder and decoder.

[0022] In a third aspect, the present invention discloses a method for identifying QTLs, which includes:

[0023] Obtaining a QTL identification model obtained by the training method described in the second aspect;

[0024] Inputting the sample data to be tested into the QTL identification model to obtain output information, wherein the output information includes the position information of the SNP site on the gene and the confidence level that the SNP site belongs to the QTL interval;

[0025] The QTL interval is identified based on the position information and the confidence level.

[0026] In some embodiments, the identification method is used to identify at least one of the following:

[0027] Identification of QTL for plant height in maize;

[0028] Identification of QTL for rice plant height;

[0029] Identification of QTLs for flowering time in rice; and

[0030] Identification of QTLs for intermuscular spines in Wuchang fish.

[0031] In a fourth aspect, the present application discloses a QTL identification device, comprising:

[0032] An acquisition module is used to acquire a training set, wherein the training set includes SNP data and sample markers of the SNP data, wherein the SNP data is obtained by sequencing mixed pool samples of the QTL mapping population;

[0033] A construction module is used to construct a residual U-net model. The residual U-net model is based on the overall architecture of U-net and is formed by adding residual connections to the encoding part and the decoding part. The encoding part is used to extract low-resolution SNP data, and the decoding part is used to extract high-resolution SNP data.

[0034] A training module is used to train the residual U-net model using the training set to obtain a residual U-net model.

[0035] In a fifth aspect, an embodiment of the present application discloses an electronic device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the computer program, the method described in the second aspect and / or the method described in the second aspect are implemented. BRIEF DESCRIPTION OF THE DRAWINGS

[0036] Figure 1 This is a schematic diagram of the implementation flow of the sample processing method for QTL identification model training provided in an embodiment of the present application.

[0037] Figure 2 This is a schematic diagram of the implementation flow of a training method for a QTL identification model provided in an embodiment of the present application.

[0038] Figure 3This is a schematic diagram of the implementation process of a QTL identification method provided in an embodiment of the present application.

[0039] Figure 4 It is a schematic diagram of a specific method for identifying QTL intervals for corn plant height using a sample processing method, a QTL model training method, and a QTL identification method provided in an embodiment of the present application.

[0040] Figure 5 This is a training set sample for QTL model training provided in an embodiment of the present application.

[0041] Figure 6 This is a signal diagram of the deviation and signal-to-noise ratio calculation using simulated data in this application. The vertical axis is the identification value of each SNP given by the trained residual U-net model, and the horizontal axis is the SNP position.

[0042] Figure 7 This is a diagram of the QTL interval identification results for corn plant height provided in one embodiment of the present application.

[0043] Figure 8 This is a comparison chart of the identification results of the QTL identification method provided by the present application for identifying the QTL interval of corn plant height, provided in one embodiment of the present application, and the identification results of different model methods; the left figure is a comparison of the deviation results of the identification results, and the right figure is a comparison of the signal-to-noise ratio results of the identification results.

[0044] Figure 9 This is a sample data to be tested for rice plant height QTL identification provided in an embodiment of the present application.

[0045] Figure 10 This is a comparison chart of the identification results of the rice plant height QTL interval identification method provided by the present application and the identification results of different model methods; the left figure is a comparison of the deviation results of the identification results, and the right figure is a comparison of the signal-to-noise ratio results of the identification results.

[0046] Figure 11 This is a sample data to be tested for rice flowering period QTL identification provided in an embodiment of the present application.

[0047] Figure 12 This is a comparison chart of the identification results of the rice flowering period QTL interval identified by the QTL identification method provided by the present application in an embodiment of the present application and the identification results of different model methods; the left figure is a comparison of the deviation results of the identification results, and the right figure is a comparison of the signal-to-noise ratio results of the identification results.

[0048] Figure 13 This is a sample data to be tested for the identification of intermuscular spines QTL in Wuchang fish provided in an embodiment of the present application.

[0049] Figure 14 This is a comparison chart of the identification results of the method provided by the present application when the QTL identification method provided by the present application is used to identify the QTL interval of the intermuscular spines of Wuchang fish, provided in one embodiment of the present application; the left figure is a comparison of the deviation results of the identification results, and the right figure is a comparison of the signal-to-noise ratio results of the identification results. DETAILED DESCRIPTION

[0050] In order to make the purpose, technical solutions and advantages of this application more clearly understood, the present application is further described in detail below with reference to the following examples. It should be understood that the specific examples described herein are merely for the purpose of explaining this application and are not intended to limit this application. Reagents not described in detail in this application are all conventional reagents and can be obtained from commercial channels; methods not specifically described in detail are all conventional experimental methods and can be obtained from the prior art.

[0051] It should be noted that the terms "first", "second", etc. in the description and claims of the present invention and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence, nor do they play a substantial limiting role on the subsequent technical features. It should be understood that the numbers used in this way can be interchanged where appropriate, so that the embodiments of the present invention described herein can be implemented in an order other than those illustrated or described herein. In addition, the terms "including" and "having" and any variations thereof are intended to cover non-exclusive inclusions. For example, a process, method, system, product or device that includes a series of steps or units is not necessarily limited to those steps or units clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.

[0052] Recently, methods based on convolutional neural networks have begun to be applied to the field of rapid magnetic resonance imaging. This method utilizes a large amount of prior information to learn and train the convolutional neural network, obtaining optimized network parameters. The trained convolutional neural network can then rapidly reconstruct high-quality MRI images, making it a promising rapid MRI imaging method. The residual U-net network is a type of convolutional neural network with a relatively simple structure, few training parameters, and short training time. It addresses the performance degradation of deep convolutional neural networks at extremely deep depths. Currently, existing approximate applications of the residual U-net are primarily focused on image recognition processes, which are related to the convolution results and feature extraction of each layer. Applications in other fields present difficulties in processing the target object and extracting features.

[0053] In order to address the deficiencies of the prior art, the present application discloses a sample processing method for QTL identification model training and a training method for the QTL identification model, and sequentially constructs a deep learning (DL) QTL identification model based on the residual U-net model, which characterizes the response relationship between the SNP data of the QTL positioning group and the QTL of the positioning group. For example, the response relationship includes: the QTL of the positioning group responds to "the sample mark of the SNP data on the positioning group genome is continuously 1", for example, the sample mark of the positioning group based on the SNP data on the group of each point is 1, and the sample marks of the SNP data at the 5 to 10 bp positions before and after are all 1, then this interval is the identified QTL interval.

[0054] Figure 1 This is a sample processing method for QTL identification model training provided in one embodiment of the present application. The samples obtained according to the sample processing method include SNP data and sample markers. The sample markers are used to construct a two-dimensional tensor suitable for training a residual U-net model. For example, the sample processing method includes:

[0055] S100, construction of QTL mapping population;

[0056] S200, sorting and grouping individuals in the QTL mapping population according to the phenotype of the QTL mapping population to obtain a plurality of sorted groups;

[0057] S300, mixing individual DNA samples in each group to obtain multiple sequenced DNA pools;

[0058] S400, fragmenting, sequencing, identifying and calculating SNPs on the multiple DNA pools to obtain multiple SNP data;

[0059] S500 , respectively label the SNP data to obtain the SNP data and sample labels thereof, to serve as samples for training the QTL identification model.

[0060] In some embodiments of S100, constructing a QTL mapping population includes selecting two parents with significant differences in target traits, performing one or more generations of hybridization, selfing, or backcrossing to construct populations such as F2, RIL, and NIL, which are the mapping populations. Obtaining DNA samples includes planting the constructed mapping population in the field, and when the plants reach the six-leaf stage, using a borer to remove approximately 0.1 g of fresh leaves from each plant, extracting leaf DNA using the Trizol method, and measuring the concentration using a spectrophotometer.

[0061] In some embodiments of S200, the phenotypic traits of each individual plant are measured, and all individual plants are sorted according to the phenotypes (if the population consists of different families, the individuals within each family are sorted separately). For example, when the target trait is plant height, the individuals in the population can be sorted in descending order of their plant heights.

[0062] In some embodiments of S300, considering the variation range of phenotypes and the experimental cost, all individual plants are evenly divided into N (2 < N < 11) groups. The DNA of the individual plants in each group is mixed in equal amounts, and finally N DNA pools are obtained.

[0063] In some embodiments of S400, after fragmenting the DNA of the DNA pools by sonication, a sequencing library with an insert size of 400 - 500 bp is constructed according to the protocol of the KAPAHyper Prep Kit ( platform). Then, the DNA library of each pool is loaded into a lane using the Illumina Hiseq2500 system, and sequencing is performed using IlluminaHiSeq Xten to generate 150 bp paired-end reads.

[0064] In some embodiments of S500, the SNP data includes SNP position information and SNP frequency information. The SNP frequency is the frequency of its occurrence in each DNA pool fragment (i.e., for the calculation step). The method of marking the SNP data includes: marking the SNP data with continuously increasing or continuously decreasing SNP frequencies in multiple sorted DNA pools as 1, otherwise marking as 0. For example, observing the information of all pools at 128 consecutive SNP sites around the SNP locus, determining whether it is the target locus interval, and then marking the 64 middle sites, that is, the positive and negative judgments of the current small interval utilize the surrounding information; marking the SNPs with SNP frequency changes conforming to the increasing or decreasing trend as positive samples 1, and those not conforming as negative samples 0.

[0065] In some embodiments, the sample processing method includes a step of filtering the SNP data to remove low-quality SNP sites; the low-quality SNP sites have at least one of the following characteristics:

[0066] The number of reads for the corresponding DNA pool sequencing is lower than the threshold; this threshold is usually 1 / 3 of the sequencing depth;

[0067] The SNP frequency in the DNA pool significantly deviates from 0.5 in the same direction; and

[0068] The allele frequency difference between a certain SNP locus and the adjacent SNP is greater than 0.1.

[0069] In order to address the deficiencies of the prior art, the present invention discloses a training method for a QTL identification model. Figure 2 This is a flowchart of a training method for QTL identification model provided by one embodiment of the present application. The method in this embodiment can be executed by an electronic device. The electronic device includes but is not limited to a computer, tablet computer, server, mobile phone, camera or wearable device, etc. Among them, the server includes but is not limited to an independent server or a cloud server, etc. Figure 2 As shown, the method for training a residual U-net model for identifying maize plant height QTLs may include steps S101 and S102. The method for training a residual U-net model for identifying maize plant height QTLs can effectively detect minor QTL information with an explanation rate as low as 5% in complex traits and complex backgrounds.

[0070] The QTL model training method includes:

[0071] S101. Obtain a training set, the training set including multiple training samples, each of the training samples including SNP data and a sample label for the SNP data, the SNP data being obtained by sequencing a pooled sample of a QTL mapping population, and the sample labeling method including: labeling SNP data with a continuously increasing or continuously decreasing SNP frequency in multiple sequenced pooled DNAs as 1, and otherwise labeling SNP data as 0;

[0072] S102. Input the two-dimensional tensor composed of the sample labels as the input layer into the residual U-net model, and use the backpropagation algorithm and stochastic gradient descent method to iteratively update the weights of each layer according to the size of the forward propagation loss value. When the loss value of the model tends to converge, stop training and obtain the QTL recognition model.

[0073] In some embodiments of S102, the residual U-net model is composed of an input layer, an encoder, a decoder, and an output layer; each input of the input layer is a two-dimensional tensor composed of 64 marked sites in the training set, and the encoder is composed of a convolutional neural network and a residual network. After four downsamplings, deep information is obtained, and then after four upsamplings, the deep information is converted into shallow information, and finally passed through a convolution layer with a 1×1 convolution kernel and a sigmoid activation function; the encoder and decoder are both composed of three convolution layers, and the convolution kernels are 1×1, 3×3, and 3×3 respectively, and downsampling is completed by maximum pooling, and each upconvolution will also pass through a convolution layer; the residual connection acts after every three global convolutions, and there are also residual connections in the corresponding layers between the encoder and decoder.

[0074] In some embodiments, "training" includes pre-training and tuning; during the tuning process, the last 10 layers of the model are trained and the parameters of other layers are frozen.

[0075] In some embodiments, the SNP data in the training set is n×64×m, where n represents the batch size, 64 represents 64 sites, and m represents the frequency information of m mixed pools.

[0076] like Figure 3 This is a schematic diagram of the implementation process of a QTL identification method provided in one embodiment of the present application. The method in this embodiment can be executed by an electronic device. The electronic device includes but is not limited to a computer, a tablet computer, a server, a mobile phone, a camera or a wearable device. Among them, the server includes but is not limited to an independent server or a cloud server. Figure 3 The QTL identification method includes steps S101 to S102, and further includes:

[0077] S103: Input the sample data to be tested into the QTL identification model to obtain output information, which includes the location information of the SNP site on the gene and the confidence level that the SNP site belongs to the QTL interval;

[0078] S104: Identify QTL intervals based on location information and confidence.

[0079] In some embodiments, the position information of the SNP site on the gene and the confidence that the SNP site belongs to the QTL interval are 64×1 tensors, and the segmentation effect is achieved by identifying each SNP site.

[0080] In some embodiments, to make the recognition result more effective, all points are subjected to a cubic kernel regression smoothing process, which can effectively improve the signal-to-noise ratio and reduce the impact of noise.

[0081] To further elaborate on the QTL interval identification for maize plant height using the above sample processing method, QTL model training method and QTL identification method, Figure 4 The implementation process of this embodiment is shown in the following figure:

[0082] Figure 4Figure a shows the implementation process of S100 to S300 of the sample processing method for QTL identification model training. In the embodiment of this step, the classic corn inbred lines Huang Zao Si and 1462 were selected, and the shorter Huang Zao Si was used as the female parent and the taller 1462 was used as the male parent for hybridization to obtain F1, and the F1 was self-pollinated to obtain F2, and the F2 population was the positioning population. In the embodiment of this step, the target property was corn plant height. The F2 population was sown in the fields of Beijing, and a total of 7160 individual plants with a wide range of plant height variations were obtained, including 47 families. When the plant was in the six-leaf stage, about 0.1g of fresh leaves were taken from each individual plant with a hole punch, and the leaf DNA was extracted using the Trizol method and its concentration was measured using a spectrophotometer, which was the DNA sample with the target trait. The individual plants in each family were sorted from low to high in height and divided into 10 equal parts, with the lowest 10% of plant heights as the first part, and 10%-20% as the second part... Then, the corresponding number of individual plant DNA samples from each family were mixed in equal amounts to obtain 10 DNA pools.

[0083] Figure 4 Figure b shows the implementation process of steps S400 to S500 of the sample processing method for QTL identification model training.

[0084] In the embodiment of this step, the DNA in the DNA pool was fragmented by ultrasonic treatment according to the KAPA HyperPrep Kit ( Sequencing libraries with insert sizes of 400-500 bp were constructed using a protocol based on the Illumina platform. Each pool of DNA libraries was then loaded into a single lane on the Illumina HiSeq 2500 system and sequenced using the Illumina HiSeq Xten, generating 150 bp paired-end reads. Each pool generated 200 Gb of sequencing data with a coverage depth of ~100×.

[0085] In the example of this step, the reads generated by sequencing were aligned to the reference genome of version B73V4 using BWA software, and the resulting sam files were converted to bam files using SAMtools software. The bam file contents were then sorted using Picard software and reads generated by PCR duplication were deleted. Whole-genome SNP detection was then performed using the HaplotypeCaller module of GATK software, with all parameters set to the software defaults. Finally, a VCF file containing the variation information for the ten pools was generated.

[0086] To filter out low-quality SNP sites, in the embodiment of this step, S400 also includes filtering out low-quality SNP sites. Low-quality SNP sites include those where the number of reads in the corresponding DNA pool sequencing is lower than a threshold (usually set to 1 / 3 of the sequencing depth), the SNP frequencies in the DNA pool deviate significantly from 0.5 in the same direction (P < 0.01), and the allele frequency difference between a certain SNP site and an adjacent SNP is greater than 0.1.

[0087] In the embodiment of this step, positive and negative samples are marked, and a training set and a validation set are divided. Based on the VCF file obtained above, combined with the positions of some known plant height genes, some positions are selected on the whole genome for positive and negative sample marking. The frequencies of ten pools of 128 consecutive SNP sites at the observation position are observed. If the allele frequency change conforms to increasing or decreasing, the middle 64 sites are marked as positive sample 1; if the allele frequency change fluctuates around 0.5, the middle 64 sites are marked as negative sample 0. Figure 4 b shows the frequency of samples marked as 1 (Relevant) in each pool and the frequency of samples marked as 0 (Irrelevant) in each pool. The labeled SNP data were divided into training set, validation set and test set according to the ratio of 6:3:1. Figure 5 are SNP data examples for training set samples, which respectively show the SNP sites at different positions on chromosome 1 and the allele frequencies of the SNP sites in 10 mixed pools.

[0088] Figure 4 c shows a schematic diagram of using the training set to train the constructed RU-net and using the validation set for verification. Figure 4 The left figure is a schematic diagram of the RU-Net structure provided in this embodiment, and the right figure is the validation result of the trained model using a validation set consisting of 10 DNA pools. As can be seen from the results, the AUC validation result demonstrates that the recognition authenticity of its detection method is very high.

[0089] Figure 4 d shows the results of identification using the trained RU-net model. In this embodiment, the built RU-net model is used to scan SNPs across the entire genome to identify the confidence level of each SNP associated with the phenotype. Figure 7 As shown in the figure, the SNP position is used as the horizontal axis and the SNP confidence is used as the vertical axis to draw a graph, and a LOWESS fitting is performed on it. The peak of the fitting curve is the identified target site, as shown in the figure. Figure 7 As shown on the right, QTL intervals were identified at chromosomes 1, 2, 3, and 6.

[0090] In addition, 10 QTL sites were randomly set on the genome, and the generation of maize F2 population was simulated. The mixed pool sequencing as shown above was performed to obtain VCF files of the variation information of the ten pools. The highest pool and the lowest pool were selected for SNP filtering and labeling (some methods are only applicable to two mixed pools). G', ED4, K, SmoothLOD, Ridit and SNP-index methods were used for identification, and the identified QTL signals and the real QTL signals were mapped (as shown in Figure 2). Figure 6 ), calculate the deviation and signal-to-noise ratio, and statistically analyze the deviation and signal-to-noise ratio differences between these models and algorithms and the RU-net model (deep learning model DL) improved by the embodiment of the present application. The results are as follows Figure 8 As shown, the recognition results of the RU-net model (deep learning model DL) are the best in terms of bias (left figure) and signal-to-noise ratio (right figure), which illustrates the accuracy of the training method for the RU-net model and the identification of corn QTL information provided in this application.

[0091] In one embodiment, to further describe the use of the above sample processing method, QTL model training method and QTL identification method to identify QTL intervals for rice plant height, refer to Figures 1 to 5 In the implementation process shown in FIG, the rice plant height QTL mapping population was processed using the sample processing method described above, and the sample data to be tested (such as Figure 9 As shown), it is input into the above training to obtain the QTL recognition model. The recognition result of the RU-net model (deep learning model DL) is the best in terms of deviation and signal-to-noise ratio. The output result is as follows Figure 10 As shown, Figure 10 a is the LOWESS fitting curve, Figure 10 b is the identified QTL interval result, which shows that the training method for the RU-net model and the RU-net model provided in this application can effectively identify rice QTL information with a very high accuracy.

[0092] In one embodiment, referring to Figures 1 to 5 In the implementation process shown in the figure, the rice flowering period QTL mapping population is processed by the sample processing method as described above, and the individuals in the mapping population are grouped according to the order of flowering time to obtain multiple DNA pools, and then fragmented, sequenced, SNP identified and calculated based on the pools to obtain multiple SNP data, which are marked to obtain the sample data to be tested (such as Figure 11 As shown), the sample data to be tested is trained to obtain the QTL identification model. The output result is as follows Figure 12 As shown, Figure 12 a is the LOWESS fitting curve, Figure 12b is the identified QTL interval result, which shows that the training method for the RU-net model and the RU-net model provided in this application can effectively identify rice QTL information with a very high accuracy.

[0093] In one embodiment, referring to Figures 1 to 5 The implementation process shown in the figure is to use the above-mentioned sample processing method to process the Wuchang fish intermuscular spine QTL (whether there is an intermuscular spine) positioning population, group the individuals in the positioning population according to the order of flowering time, obtain two DNA pools, and perform fragmentation, sequencing, SNP identification and calculation on them to obtain multiple SNP data as the sample data to be tested (such as Figure 13 As shown), input into the above training to obtain the QTL identification model. The output result is as follows Figure 14 As shown, Figure 14 a is the LOWESS fitting curve, Figure 14 b is the identified QTL interval result, which shows that the training method for the RU-net model and the RU-net model provided in this application can effectively identify the QTL information of Wuchang fish with a very high accuracy.

[0094] To this end, an embodiment of the present application also discloses a structural schematic diagram of a QTL identification device. The device includes an acquisition module, a construction module, a training module and an output module. Among them, the acquisition module is used to obtain a training set, and the training set includes SNP data and sample markers of SNP data, and the SNP data is SNP data obtained by sequencing mixed pool samples of the QTL positioning population. The construction module is used to construct a residual U-net model, and the residual U-net model is based on the overall architecture of U-Net, and residual connections are added to the encoding part and the decoding part respectively. The encoding part is used to extract low-resolution SNP data, and the decoding part is used to extract high-resolution SNP data. The training module is used to train the residual U-net model using the training set to obtain a residual U-net model. The output module is used to identify the QTL information therein using the trained residual U-net model given a DNA sequence.

[0095] The embodiment of the present application also provides an electronic device, which may include one or more processors (only one is shown), a memory, and a computer program stored in the memory and run on one or more processors, for example, a lightweight program of a U-net model. When one or more processors execute the computer program, the various steps in the lightweight method embodiment of the U-net model can be implemented. Alternatively, when one or more processors execute the computer program, the functions of each module / unit in the lightweight device embodiment of the U-net model can be implemented, which is not limited here.

[0096] Those skilled in the art will understand that the electronic device of the present application may include more or fewer components than shown in the figure, or a combination of certain components, or different components. For example, the electronic device may also include input and output devices, network access devices, buses, etc.

[0097] In one embodiment, the processor may be a central processing unit (CPU), or other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), field-programmable gate arrays (FPGA), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor may be a microprocessor or any conventional processor, etc.

[0098] In one embodiment, the memory may be an internal storage unit of an electronic device, such as a hard disk or memory of the electronic device. The memory may also be an external storage device of the electronic device, such as a plug-in hard disk equipped on the electronic device, a smart media card (SMC), a secure digital (SD) card, a flash card, etc. Furthermore, the memory may include both an internal storage unit of the electronic device and an external storage device. The memory is used to store computer programs and other programs and data required by the electronic device. The memory may also be used to temporarily store data that has been output or is about to be output.

[0099] Those skilled in the art can clearly understand that, for the convenience and brevity of description, only the division of the above-mentioned functional units and modules is used as an example for illustration. In actual applications, the above-mentioned functions can be distributed and completed by different functional units and modules as needed, that is, the internal structure of the device can be divided into different functional units or modules to complete all or part of the functions described above. The functional units and modules in the embodiment can be integrated into one processing unit, or each unit can exist physically alone, or two or more units can be integrated into one unit. The above-mentioned integrated unit can be implemented in the form of hardware or in the form of software functional units. In addition, the specific names of the functional units and modules are only for the convenience of distinguishing each other, and are not used to limit the scope of protection of this application. The specific working process of the units and modules in the above-mentioned system can refer to the corresponding process in the aforementioned method embodiment, and will not be repeated here.

[0100] In the embodiments provided in this application, it should be understood that the disclosed devices / electronic devices and methods can be implemented in other ways. For example, the device / electronic device embodiments described above are merely illustrative. For example, the division of modules or units is merely a logical function division. In actual implementation, there may be other division methods, such as multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some interfaces, indirect coupling or communication connection of devices or units, which can be electrical, mechanical or other forms.

[0101] Units described as separate components may or may not be physically separate, and components shown as units may or may not be physical units, that is, they may be located in one place or distributed across multiple network units. Some or all of these units may be selected to achieve the purpose of this embodiment according to actual needs.

[0102] In addition, the functional units in the various embodiments of the present application may be integrated into a single processing unit, or each unit may exist physically separately, or two or more units may be integrated into a single unit. The aforementioned integrated units may be implemented in the form of hardware or software functional units.

[0103] If the integrated module / unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the present application implements all or part of the process in the above-mentioned method embodiment, and the computer program that can be completed by instructing the relevant hardware through a computer program can be stored in a computer-readable storage medium. When the computer program is executed by the processor, it can implement the steps of the above-mentioned various method embodiments. Among them, the computer program includes computer program code, and the computer program code can be in source code form, object code form, executable file or some intermediate form. The computer-readable medium may include: any entity or device that can carry computer program code, recording medium, USB flash drive, mobile hard disk, magnetic disk, optical disk, computer memory, read-only memory (ROM), random access memory (RAM), electric carrier signal, telecommunication signal and software distribution medium. It should be noted that the content contained in the computer-readable medium can be appropriately increased or decreased according to the requirements of legislation and patent practice in the jurisdiction. For example, in some jurisdictions, according to legislation and patent practice, computer-readable media does not include electric carrier signal and telecommunication signal.

[0104] The above is only a preferred specific implementation method of the present application, but the scope of protection of the present application is not limited thereto. Any changes or replacements that can be easily thought of by any technician familiar with this technical field within the technical scope disclosed in this application should be covered by the scope of protection of the present application.

Claims

1. A training method for a QTL identification model, comprising: Obtaining a training set, the training set including a plurality of training samples obtained according to the sample processing method, each of the training samples including SNP data and a sample marker of the SNP data, the SNP data being obtained by sequencing a mixed pool of samples from a QTL mapping population; The two-dimensional tensor composed of the sample labels is input as the input layer into the residual U-net model. The back propagation algorithm and the stochastic gradient descent method are used to iteratively update the weights of each layer according to the size of the forward propagation loss value. When the loss value of the model tends to converge, the training is stopped to obtain the QTL recognition model. The sample processing method comprises: Construction of QTL mapping population; sorting and grouping individuals in the mapping population according to the phenotype of the QTL mapping population to obtain a plurality of sorted groups; Individual DNA samples within each group were mixed to obtain multiple sequenced DNA pools; performing fragmentation, sequencing, SNP identification and calculation on the multiple DNA pools to obtain SNP data of the multiple pools; Marking the SNP data respectively to obtain the SNP data and sample markers thereof as samples for training the QTL identification model; The SNP data includes SNP position information and SNP frequency information, wherein the SNP frequency is the frequency of occurrence of the SNP in each DNA pool fragment. The method for marking the SNP data includes: The SNP data with continuously increasing or decreasing SNP frequencies in multiple sequenced DNA pools are marked as 1, otherwise they are marked as 0; The sample processing method further includes a step of filtering the SNP data to remove low-quality SNP sites; the low-quality SNP sites have at least one of the following characteristics: The corresponding number of reads from the DNA pool sequencing is lower than the threshold; The SNP frequencies in the DNA pool deviate significantly from 0.5 in the same direction; and The allele frequency difference between a certain SNP site and the adjacent SNP is greater than 0.

1.

2. The training method according to claim 1, wherein: The trained QTL identification model characterizes the response relationship between the SNP data of the QTL mapping population and the QTL of the mapping population.

3. The training method according to claim 2, wherein: The residual U-net model consists of an input layer, an encoder, a decoder, and an output layer; each input of the input layer is a two-dimensional tensor consisting of 64 sites of the sample label, and the encoder consists of a convolutional neural network and a residual network. After four downsamplings, deep information is obtained, and then after four upsamplings, the deep information is converted into shallow information, and finally passed through a convolution layer with a 1×1 convolution kernel and a sigmoid activation function; the encoder and decoder are both composed of three convolution layers, with convolution kernels of 1×1, 3×3, and 3×3, respectively, and downsampling is completed by maximum pooling, and each upconvolution is followed by a convolution layer; the residual connection acts after every three global convolutions, and there is also a residual connection between the corresponding layers of the encoder and decoder.

4. A method for identifying QTLs, comprising: Obtaining a QTL identification model obtained by the training method according to any one of claims 1 to 3; Inputting the sample data to be tested into the QTL identification model to obtain output information, wherein the output information includes the position information of the SNP site on the gene and the confidence level that the SNP site belongs to the QTL interval; The QTL interval is identified based on the position information and the confidence level.

5. The identification method according to claim 4, used to identify at least one of the following: Identification of QTL for plant height in maize; Identification of QTL for rice plant height; Identification of QTLs for flowering time in rice; and Identification of QTLs for intermuscular spines in Wuchang fish.

6. A QTL identification device, characterized in that: include: an acquisition module, configured to acquire a training set, wherein the training set includes SNP data obtained by the training method according to any one of claims 1 to 3 and sample markers of the SNP data, wherein the SNP data is obtained by sequencing mixed pool samples of a QTL mapping population; A construction module for constructing the residual U-net model according to claim 3, wherein the residual U-net model is based on the overall architecture of U-Net and is formed by adding residual connections to the encoding part and the decoding part, respectively, the encoding part is used to extract low-resolution SNP data, and the decoding part is used to extract high-resolution SNP data; A training module, configured to train the residual U-net model using the training set to obtain a residual U-net model; The output module is used to identify the QTL information therein using the trained residual U-net model given a DNA sequence.

7. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the computer program, the method according to any one of claims 1 to 3 and / or the method according to claim 4 or 5 is implemented.

Images