Crop mitochondrial editing site prediction method, system and storage medium

Abstract

The application discloses a crop mitochondrial editing site prediction method and system and a storage medium. Mitochondrial genome sequences of multiple species and corresponding C-to-U RNA editing site annotation information are obtained. Fragments of a set length are taken from the upstream and downstream of the mitochondrial genome sequence and the corresponding C-to-U RNA editing site annotation information sequence as positive samples with the editing site as the center. Any site without editing site annotation is taken as the center, and fragments of a set length are taken from the upstream and downstream of the mitochondrial genome sequence and the corresponding C-to-U RNA editing site annotation information as negative samples. Part of the positive samples and the negative samples corresponding to each species are used to construct a training set of each species, the training set of each species is used as the input of a convolutional neural network, and the optimal prediction model corresponding to each species is obtained. The application has high prediction accuracy and can better fit the potential encoding mechanism of C-to-U RNA editing of different species.

Description

Technical Field

[0001] This invention relates to editing site prediction technology, and in particular to a method, system and storage medium for predicting crop mitochondrial editing sites. Background Technology

[0002] RNA editing is a predetermined modification that causes codon changes through nucleotide alterations at the RNA level. It is achieved through the substitution, insertion, or deletion of nucleotides and is a necessary processing step for mitochondrial genes to produce functional proteins. It is also one of the important ways in which the cell nucleus regulates mitochondrial gene expression. Since its first discovery in the 20th century, the mechanism of RNA editing has received high attention.

[0003] C-to-U RNA editing is the most common type in crops. Efficient, rapid, and economical identification of C-to-U RNA editing sites remains a significant challenge. Traditional RNA editing site identification procedures are complex, involve numerous steps, and are costly. While the results are relatively accurate, they often only identify single genes within the coding region. In fact, RNA editing in non-coding regions can also regulate gene expression. Furthermore, the entire experimental process is susceptible to many potential interferences, such as PCR errors, DNA editing interference from SNPs and mutants, and the impact of different cell types' gene expression levels on transcripts. Therefore, a series of techniques are needed to reduce the impact of erroneous results during identification.

[0004] Thanks to the development of machine learning, new research ideas have been provided for high-precision prediction of editing sites. Many scholars have made extensive attempts to predict RNA editing sites using traditional machine learning methods, such as decision trees, random forests, and support vector machines, but the prediction accuracy is generally low, and they rarely consider differences between species. Until 2022, Qin et al., through various feature extraction modeling methods, speculated that the C-to-U RNA editing mechanisms of closely related species may have greater similarities, providing a new research approach for predicting C-to-U RNA editing across editing sites.

[0005] Manually defined feature extraction methods have a significant impact on the accuracy of traditional machine learning models. Deep learning's autonomous feature learning mode, while producing highly abstract features that complicate feature interpretation, avoids the limitations of human thinking, maximizing the reconstruction of patterns and improving the model's fitting ability. In 1986, Rumelhart et al. proposed a Back Propagation Network (BPN) for learning representations by back-propagating errors, and deep learning gradually gained attention from researchers. As a type of deep learning model, the Convolutional Neural Network (CNN) possesses characteristics such as local connectivity, weight sharing, pooling operations, and multi-layered structures. Its powerful automatic feature learning capabilities make it the most widely used in image and text recognition.

[0006] Genome data is essentially a text sequence composed of bases. Current technologies have not explored the interspecies connections of C-to-URNA editing mechanisms, resulting in limited prediction accuracy. Summary of the Invention

[0007] The technical problem to be solved by this invention is to provide a method, system and storage medium for predicting crop mitochondrial editing sites, thereby improving the accuracy of editing site prediction, in order to address the shortcomings of existing technologies.

[0008] To solve the above-mentioned technical problems, the technical solution adopted by the present invention is: a method for predicting mitochondrial editing sites in crops, comprising the following steps:

[0009] S1. Obtain mitochondrial genome sequences and corresponding C-to-U RNA editing site annotation information from multiple species; take fragments of a set length upstream and downstream of the mitochondrial genome sequence and the corresponding C-to-U RNA editing site annotation information as positive samples, centered on the editing site; take fragments of a set length upstream and downstream of the mitochondrial genome sequence and the corresponding C-to-U RNA editing site annotation information as negative samples, centered on any site without editing site annotation.

[0010] S2. Construct a training set for each species using some positive and negative samples corresponding to each species. Use the training set of each species as the input of a convolutional neural network. Use the probability that the center point of each sample is an edit point as the output of the convolutional neural network. Train the convolutional neural network to obtain the optimal prediction model for each species.

[0011] The convolutional neural network includes an input layer, an embedding layer, a first convolutional layer, a second convolutional layer, a downsampling layer, and an output layer connected in sequence.

[0012] This invention is the first to use deep learning to predict mitochondrial C-to-U RNA editing events in crops using genome sequences. Compared to traditional machine learning modeling, this model has the advantage of autonomously learning features, avoiding the limitations of manually designed feature extraction, and can restore the regularity of C-to-U RNA editing phenomena in biology to the greatest extent, thus improving the accuracy of editing site prediction.

[0013] Interspecific analysis of the model prediction results revealed that closely related species have more similar C-to-U RNA editing mechanisms. This suggests that this finding should be used as a screening criterion when predicting new species to make the prediction results more reliable.

[0014] In this invention, the learning rate is set to 0.001 during the training of the convolutional neural network.

[0015] In this invention, the sizes of the first convolutional layer and the second convolutional layer are set to 16 to 30.

[0016] In this invention, the number of positive samples and the number of negative samples are equal. This is to minimize the risk of suboptimal parameter optimization (i.e., poor optimization) and evaluation distortion.

[0017] To further improve the reliability of prediction, in step S1 of this invention:

[0018] 1) Determine whether the genera of the multiple species are the same, and merge the positive and negative samples of all species of the same genera to obtain the first fused sample;

[0019] 2) Determine whether the families of species from other genera are the same, and merge the positive and negative samples of all species from the same family to obtain the second fused sample;

[0020] 3) Determine whether the orders of species from other families are the same, and merge the positive and negative samples of all species with the same order to obtain the third fused sample;

[0021] 4) Determine whether the classes of the other different target species are the same, and merge the positive and negative samples of all species in the same class to obtain the fourth fused sample;

[0022] The training set acquisition process in step S2 corresponds to the following:

[0023] A training set is constructed using the first fusion sample, the second fusion sample, the third fusion sample, and the fourth fusion sample.

[0024] The method of the present invention further includes:

[0025] S3. For newly acquired species, use the optimal prediction model corresponding to that species to predict C-to-U RNA editing sites.

[0026] To further improve the reliability of the prediction results, in step S3 above, for the species to be predicted, the optimal prediction model closest to the species to be predicted is obtained in sequence according to the genus, family, order and class of the species to be predicted, and the optimal prediction model is used to predict the C-to-U RNA editing site of the species to be predicted.

[0027] As an inventive concept, the present invention also provides a crop mitochondrial editing site prediction system, which includes:

[0028] One or more processors;

[0029] A memory having stored one or more programs that, when executed by one or more processors, cause the one or more processors to implement the steps of the method described above.

[0030] As an inventive concept, the present invention also provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the steps of the method described above.

[0031] Compared with the prior art, the beneficial effects of the present invention are: the present invention has high prediction accuracy and can better fit the potential coding mechanism of C-to-U RNA editing in different species. Attached Figure Description

[0032] Figure 1 The range of hyperparameters for the model in this embodiment of the invention;

[0033] Figure 2 This is the model structure of an embodiment of the present invention;

[0034] Figure 3 This invention presents the interspecific relationships of 13 crops using an Euclidean distance model.

[0035] Figure 4 Interspecific relationships in the Euclidean distance model for dicotyledonous plants, as described in this embodiment of the invention;

[0036] Figure 5 This invention provides interspecific prediction and interspecific ANI relationships for 13 crops.

[0037] Figure 6 This invention compares the predicted mitochondrial full-edit site of C. pepo with the NCBI-annotated site. Figure 6 In the diagram, the blue dots represent C-to-U RNA editing sites annotated by NCBI, and the green dots represent DEEPReditor-CMG prediction results. The darker the color, the more sites are located near this region. Detailed Implementation

[0038] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0039] Example 1

[0040] This embodiment provides a method for predicting mitochondrial editing sites in crops.

[0041] In this embodiment, the mitochondrial genomes and their annotation information for 13 crops (A. thaliana, Arabidopsis thaliana ecotype Col-0, A. thaliana ecotype Col-0, B. napus, Capsella bursa-pastoris, C. bursa-pastoris, Citrullus lanatus, C. lanatus, Cucurbita pepo, C. pepo, Lactuca sativa var. Capitata, L. sativa var. Capitata, Nelumbo nucifera, N. nucifera, O. sativa, Raphanus sativus, R. sattivus, Solanum tuberosum, S. tuberosum, Triticum aestivum, T. aestivum, and Zea mays, Z. mays) were obtained from the National Center for Biotechnology. Information (NCBI) download (Table 1). Since the mitochondrial reference genome NC_036024.1 of T. aestivum does not have edit site annotation information, but edit site annotation information is available for 11 mitochondrial genome fragments: X57965, X57968, Y13920, AB022060, AJ295996, X69720, X75036, X79609, X69205, AF082025, and AJ535507, the T. aestivum samples were derived from these 11 fragments. Samples from other species were derived from the whole mitochondrial genome (with edit site annotations).

[0042] Table 1. Classification information and sample size of thirteen crops

[0043]

[0044] In this embodiment, based on the mitochondrial genome sequences of 13 species and the corresponding C-to-U RNA editing site annotation information, positive samples were taken with the editing site C as the center, with 250bp upstream and downstream (Du P, He T, Li Y. Prediction of C-to-U RNA editing sites in higher plant mitochondria using only nucleotide sequence features. Biochem Biophys Res Commun. 2007, 358:336-341. and Qin S, Fan Y, Hu S, et al. iPReditor-CMG: Improving a predictive RNAeditor for crop mitochondrial genomes using genomic sequence features and an optimal support vector). machine.Phytochemistry.2022,200:113222.); while negative samples are centered on C without edit site annotation, with 250bp taken upstream and downstream in the same way; among them, the sequence of G annotated with C-to-U edit site is complemented (according to the central dogma, the sequence complementation is achieved by Python programming: A→T, C→G, T→A, G→C), the number of negative samples is equal to the number of positive samples and the extraction process is random.

[0045] The training, validation, and test sets each consist of equal numbers of positive and negative samples, with a training-to-independent-test-set ratio and a validation-to-training-set ratio of 8:2. The training and independent-test sets are randomly split five times using different random factors. The training dataset in this embodiment is balanced (data balanced: the ratio of positive to negative samples in the training set is 1:1, and the ratio of positive to negative samples in the test set is also 1:1), which minimizes the risk of poor parameter optimization and evaluation distortion (Krawczyk, B. Learning from imbalanced data: open challenges and future directions. Prog Artif Intell. 2016, 5:221-232.).

[0046] In this embodiment, it is based on Tensorflow (Goldiegadde.TensorFlow 2.0.0). https: / / github.com / tensorflow / tensorflow / releases / tag / v2.0.0.) and Keras(Keras). https: / / github.com / keras-team / keras A CNN model is built with the following structure: input layer + embedding layer + two convolutional layers and pooling layer + regularization layer (30%) + output layer. The batch size is 50, the optimizer is Adam, and the loss function is the binary linear cross-entropy function (see structure). Figure 2 The evaluation index is ACC (Formula 1).

[0047] In this embodiment, the hyperparameter optimization process mainly involves range search. A preliminary range can be defined first through random search, followed by grid search to optimize the hyperparameters. This approach has the advantage of quickly narrowing the hyperparameter range, significantly reducing computational consumption, saving time and cost, and making it more suitable for rapid model training on portable hosts.

[0048] In this embodiment, the effectiveness of the proposed scheme is verified using the following interspecific relationship indicators. The interspecific relationship indicators are as follows:

[0049] (1) Interspecific ANI values

[0050] Average nucleotide identity (ANI) is a method at the nucleic acid level that replaces DNA-DNA hybridization (DDH) by mimicking the experimental process of DDH to calculate the similarity of all orthologous protein-coding genes between any two genomes (Goris J, Konstantinidis KT, Klappenbach JA, et al. DNA-DNA hybridization values ​​and their relationship to whole-genome sequence similarities. Int J Syst Evol Microbiol. 2007, 57:81-91.). The Basic local alignment search tool (Blast) is an algorithm for calculating ANI (Altschul SF, Gish W, Miller W, et al. Basic local alignment search tool. J MolBiol. 1990, 215:403-410.). Due to its fast alignment speed and high accuracy, it is widely used in sequence alignment research. ANI analysis was performed on the whole mitochondrial genomes of 13 species (T. aestivum served as the mitochondrial reference genome, ID: NC_036024.1). Using the BLAST method, pairwise alignments were performed on each of the 13 species, yielding a total of 78 alignment results. These values ​​are referred to as "interspecific ANI values" in this embodiment. A higher "interspecific ANI value" indicates a closer relationship between two species, while a lower value indicates a more distant relationship.

[0051] (2) Interspecific predicted values

[0052] In the model building section, after selecting the optimal CNN model for each of the 13 species, interspecific predictions were performed on the 13 species. For example, if the model for species A predicts the value of species B, it is recorded as "interspecific prediction value +", and conversely, if the model for species B predicts the value of species A, it is recorded as "interspecific prediction value -". All samples of each species were used for prediction, and the evaluation metric was ACC, resulting in a total of 2*78 sets of interspecific prediction results. The larger the "interspecific prediction value" between two species, the closer the two species are, and vice versa.

[0053] (3) Model Euclidean distance

[0054] In the interspecific prediction design steps, the last sigmoid layer of the optimal CNN model for each species is removed, and the penultimate layer is used as the output layer. Assuming each model's output is a vector of length 256, this vector best represents the differences between the models. Then, positive and negative samples from each of the 13 species are used as the input layers for their respective models, resulting in 13 positive and 13 negative sample vectors (to ensure vector stability and reduce error, the input layer consists of all samples from each species, and the resulting vector matrix is ​​then averaged longitudinally). Finally, the Euclidean distance (Formula 2) is calculated between the 13 vectors of the positive and negative samples, resulting in a total of 2*78 sets of Euclidean distances. The smaller the "model Euclidean distance" between two species, the closer the two species are; conversely, the larger the distance, the more distant the relationship.

[0055]

[0056]

[0057] TP: True positive, FN: False negative, TN: True negative, FP: False positive, d: Euclidean distance, a i : The i-th position of vector A, b i : The i-th element of vector B.

[0058] Taking the species C. pepo (randomly selected) as an example, a single-layer convolutional network model was built with learning rates of 0.1, 0.001, and 0.0001. The kernel size in the convolutional layer was 10-26, with a stride of 2. The size of the embedding layer changed synchronously with the kernel size. The filter size was 32-256, which doubled. The size of the fully connected layer was 32-256, which doubled. Thus, 135 single-layer convolutional models were set for each learning rate.

[0059] When the learning rate is 0.01, 65 out of 135 models fail to fit (val_ACC = 0.5). However, models with learning rates of 0.001 and 0.0001 show stronger and more stable fits, with the 0.001 learning rate significantly outperforming the 0.0001 learning rate. Figure 1 (a) In the diagram, the kernel size does not significantly affect the accuracy of the validation set, but there is a general positive correlation. When the kernel size is less than 14, the model results are poor, and when it is 26, the accuracy of the validation set is optimal. Figure 1 (b)); while a larger filter size significantly improves the accuracy of the validation set. Figure 1 (c) The size of the fully connected layer has little impact on the accuracy of the validation set; a size of 256 yields better results. Figure 1 (d) in the middle.

[0060] Based on the performance patterns of single-layer convolutional neural networks, the learning rate was set to 0.001, the kernel size for each layer was set to 10-26, and the stride was 2. The first layer "filter" had 256 kernels, the second layer had 128 and 256 kernels, and the third layer had 64 and 256 kernels. The size of the fully connected layer was 256. A total of three convolutional neural network models with different layer sizes were created, each with 18 different parameters. Figure 1 (e)). Two and seven models failed to fit in the two-layer and three-layer convolutional networks, respectively, with a validation set accuracy of 0.5. The two-layer convolutional structure has higher validation set accuracy than the single-layer convolutional structure and better stability than the three-layer convolutional structure.

[0061] Ultimately, the two-layer convolutional network structure was selected as the final model structure. Figure 2 The convolutional kernel size ranges from 16 to 30, with a stride of 2. The filters are 128 and 256, the size of the fully connected layer is 256, the learning rate is 0.001, the optimizer is Adam, the loss function is a binary linear cross-entropy function, the evaluation metric is ACC, and the number of iterations is dynamically adjusted for optimal results. In this embodiment, the strictly range-tuned CNN model is called DEEPReditor-CMG.

[0062] Among the reference species A. thaliana, B. napus, and O. sativa, the independent prediction results of DEEPReditor-CMG in this embodiment are superior to those of previous methods. Based on the PREP-Mt method, the test_ACC is improved by a maximum of 13.41%, 6.90%, and 14.46%, respectively; based on the Du-SVM method, the test_ACC is improved by a maximum of 9.41%, 9.41%, and 14.46%, respectively; and based on the iPReditor-CMG method, the test_ACC is improved by a maximum of 2.20%, 10.71%, and 17.28%, respectively (Table 2). Compared to the iPReditor-CMG method, which uses 10% of the dataset as the independent test set, the new method DEEPReditor-CMG in this embodiment uses 20%, and the training set and independent test set are randomly split five times using different random factors, making the results more convincing. Furthermore, among the other 13 species, the DEEPReditor-CMG model, which features strict range-tuning, also outperformed the two-layer convolutional CNN (Empirical Hyperparameters CNN, EHCNN) model in this embodiment compared to the independent prediction results of Supersampler+BPN (SSBPN) and empirical hyperparameters (Chen J, Qian X, He Y, et al. An artificial neural network model based on DNA damage response genes to predict outcomes of lower-grade glioma patients. Brief Bioinform. 2021, 22(6): bbab190.) (Table 3). This demonstrates that range-tuning of hyperparameters in this embodiment is necessary and effective. Therefore, the model in this embodiment can better fit the potential coding mechanism of C-to-U RNA editing in different species.

[0063] Table 2 Comparison of ACC between DEEPReditor-CMG and the Reference Method

[0064]

[0065] Note: In DEEPReditor-CMG, the training set and independent test set are randomly divided five times using different random factors.

[0066] Table 3. ACC results of four methods for different species

[0067]

[0068] Note: In DEEPReditor-CMG, the training set and independent test set are randomly divided five times using different random factors.

[0069] Among the 13 species (10 dicotyledons and 3 monocotyledons): the interspecific ANI value and the model Euclidean distance were significantly negatively correlated (edited: R = -0.31, p = 0.006; unedited: R = -0.35, p = 0.002). Figure 3 Moreover, this negative correlation was elevated to a highly significant negative correlation among the 10 dicotyledonous plants (edited: R = -0.66, p = 9.68e-7; unedited: R = -0.64, p = 2.74e-6). Figure 4 Since there are only three monocotyledonous plants, although their "interspecific ANI value" and "model Euclidean distance" are strongly negatively correlated, the reliability is low. The final results show that the closer the two species are, the smaller the "model Euclidean distance" is, indicating that the C-to-U RNA editing coding mechanism of the two species is more similar. Conversely, when the two species are not closely related, the difference in the C-to-U RNA editing coding mechanism between the two species is greater.

[0070] Since the "interspecific predicted value" is directional, but essentially reflects the similarity of RNA editing mechanisms between two species, this embodiment selects the maximum value for the "interspecific predicted value" between two species. Among the 13 species: the "interspecific ANI value" and the "interspecific predicted value" are highly significantly positively correlated (R = 0.45, p = 4e-5); Figure 5 This indicates that the conclusion is consistent with the "model Euclidean distance," which suggests that the closer two species are, the more similar their RNA editing coding mechanisms.

[0071] The "interspecific ANI value" reflects the close relationship between different species at the genomic level, while the "model Euclidean distance" and "interspecific predicted value" are based on a high-precision DL model, reflecting the similarity of the patterns and mechanisms of this editing event between different species at the C-to-U RNA editing level. Between different species: the "interspecific ANI value" is negatively correlated with the "model Euclidean distance," and the "interspecific ANI value" is positively correlated with the "interspecific predicted value" (see Appendix Table 1 for all detailed evaluation values). This demonstrates from multiple perspectives that the closer two species are, the more similar their C-to-U RNA editing mechanisms are, further supporting the inferences in the SVM section. Therefore, when modeling and predicting, it is not appropriate to simply merge datasets from different species; rather, the close relationship between species should be considered. This important conclusion provides a theoretical basis for the following application section.

[0072] In the application process, the 13 species were first classified according to class, order, family, and genus (Table 1). At the genus level, if there were no species of the same genus, the pre-trained model was continued (for example, if there were 3 species of the same genus, the three species could be fused into a model at the genus level; if one genus of the 13 species had only one species, the model at the genus level would continue to use the model in Table 3). Finally, we added 6 fused models, covering all cases at the class, order, family, and genus levels for the 13 species, meaning that as long as the species to be predicted is a dicotyledonous or monocotyledonous plant, a most suitable model can be found for prediction). The datasets of species of the same genus, family, order, and class were fused sequentially, and the DEEPReditor-CMG method was used to construct and train the model (Table 4).

[0073] Table 4. Test_ACC results of the amplified model.

[0074]

[0075] This embodiment of DEEPReditor-CMG can match the model closest to the species to be predicted level by level according to the genus, family, order, and class. This can maximize the reliability of the model prediction results and enhance the personalized prediction function. At the same time, in order to facilitate the use of developers, this embodiment of DEEPReditor-CMG is suitable for predicting user-specified samples and predicting the whole genome of a species.

[0076] Taking the species *C. pepo* as an example, although this species appeared in the model training set, only a very small portion of the whole genome data was used, so whole-genome prediction remains a challenge. The prediction results using the DEEPReditor-CMG in this embodiment are as follows: Figure 6 In this embodiment, 18 of the 22 edit regions were predicted (as indicated by the red box). It was also found that the 0.7×1e6-0.8×1e6 bp interval in the whole genome is very likely to be a high-edit region, which provides guidance for experimental determination.

[0077] Example 2

[0078] Embodiment 2 of the present invention provides a terminal device corresponding to Embodiment 1 above. The terminal device can be a processing device for a client, such as a mobile phone, a laptop, a tablet computer, a desktop computer, etc., to execute the method of the above embodiments.

[0079] The terminal device in this embodiment includes a memory, a processor, and a computer program stored in the memory; the processor executes the computer program in the memory to implement the steps of the method in Embodiment 1 described above.

[0080] In some implementations, the memory may be high-speed random access memory (RAM), and may also include non-volatile memory, such as at least one disk storage device.

[0081] In other implementations, the processor can be any type of general-purpose processor, such as a central processing unit (CPU) or a digital signal processor (DSP), and there is no limitation here.

[0082] Example 3

[0083] Embodiment 3 of the present invention provides a computer-readable storage medium corresponding to Embodiment 1 above, on which a computer program / instructions are stored. When the computer program / instructions are executed by a processor, they implement the steps of the method of Embodiment 1 above.

[0084] A computer-readable storage medium can be a tangible device that holds and stores instructions for use by an instruction execution device. A computer-readable storage medium can be, for example, but not limited to, an electrical storage device, a magnetic storage device, an optical storage device, an electromagnetic storage device, a semiconductor storage device, or any combination thereof.

[0085] Those skilled in the art will understand that embodiments of this application can be provided as methods, systems, or computer program products. Therefore, this application can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this application can take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code. The solutions in the embodiments of this application can be implemented in various computer languages, such as the object-oriented programming language Java and the interpreted scripting language JavaScript.

[0086] This application is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this application. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart... Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0087] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0088] Although preferred embodiments of this application have been described, those skilled in the art, upon learning the basic inventive concept, can make other changes and modifications to these embodiments. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments as well as all changes and modifications falling within the scope of this application.

[0089] Obviously, those skilled in the art can make various modifications and variations to this application without departing from the spirit and scope of this application. Therefore, if such modifications and variations fall within the scope of the claims of this application and their equivalents, this application also intends to include such modifications and variations.

Claims

1. A method for predicting mitochondrial editing sites in crops, characterized in that, Includes the following steps: S1. Obtain mitochondrial genome sequences and corresponding C-to-U RNA editing site annotation information from multiple species; take fragments of a set length upstream and downstream of the mitochondrial genome sequence and the corresponding C-to-U RNA editing site annotation information sequence as positive samples, centered on the editing site. Centered on any site without edit site annotation, take fragments of a set length upstream and downstream of the mitochondrial genome sequence and the corresponding C-to-U RNA edit site annotation information as negative samples; S2. Construct a training set for each species using some positive and negative samples corresponding to each species. Use the training set of each species as the input of a convolutional neural network. Use the probability that the center point of each sample is an edit point as the output of the convolutional neural network. Train the convolutional neural network to obtain the optimal prediction model for each species. The convolutional neural network includes an input layer, an embedding layer, a first convolutional layer, a second convolutional layer, a downsampling layer, and an output layer connected in sequence. The crop mitochondrial editing site prediction method also includes: S3. For the species to be predicted, use the optimal prediction model corresponding to the species to be predicted to predict the C-to-U RNA editing site. In step S1: 1) Determine whether the genera of the multiple species are the same, and merge the positive and negative samples of all species of the same genera to obtain the first fused sample; 2) Determine whether the families of species from other genera are the same, and merge the positive and negative samples of all species from the same family to obtain the second fused sample; 3) Determine whether the orders of species from other families are the same, and merge the positive and negative samples of all species with the same order to obtain the third fused sample; 4) Determine whether the classes of the other different target species are the same, and merge the positive and negative samples of all species in the same class to obtain the fourth fused sample; The training set acquisition process in step S2 corresponds to the following: A training set is constructed using the first fusion sample, the second fusion sample, the third fusion sample, and the fourth fusion sample; In step S3, for the species to be predicted, the optimal prediction model closest to the species to be predicted is obtained in sequence according to the genus, family, order and class of the species to be predicted, and the optimal prediction model is used to predict the C-to-U RNA editing site of the species to be predicted.

2. The method for predicting crop mitochondrial editing sites according to claim 1, characterized in that, During the training of the convolutional neural network, the learning rate is set to 0.

001.

3. The method for predicting crop mitochondrial editing sites according to claim 1, characterized in that, The sizes of the first and second convolutional layers are set to 16 to 30.

4. The method for predicting crop mitochondrial editing sites according to claim 1, characterized in that, The number of positive samples and the number of negative samples are equal.

5. A crop mitochondrial editing site prediction system, characterized in that, include: One or more processors; A memory having stored one or more programs thereon, which, when executed by the one or more processors, cause the one or more processors to perform the steps of the method according to any one of claims 1 to 4.

6. A computer-readable storage medium, characterized in that, It stores a computer program that, when executed by a processor, implements the steps of the method as described in any one of claims 1 to 4.

Images