2OM methylation modification site identification method and system based on feature selection and deep neural network

Abstract

The invention relates to a feature selection and deep neural network-based 2OM methylation modification site identification method and system. The method comprises the following steps: converting a coding mode of an RNA sequence to obtain sequence features; reducing the dimension of the sequence features through an ANOVA feature selection algorithm, carrying out statistics on the magnitude, and carrying out feature selection based on the statistical magnitude; for the RNA sequence of the modification site on the C, inputting the input feature into a first classifier, and outputting a classification result; and for the RNA sequences of the modification sites on A, G and U, inputting the input features into a second classifier, and outputting a classification result. Dimension reduction is carried out on input features through an ANOVA feature selection algorithm, the features are selected based on variance analysis, and different classifiers are adopted for processing 2OM site data at different positions, so that the recognition accuracy of 2OM methylation modification sites can be improved.

Description

Technical Field

[0001] The present invention relates to the technical field of RNA modification site recognition and prediction, and in particular to a 2OM methylation modification site recognition method and system based on feature selection and deep neural network. Background Art

[0002] Ribose 2′-O-methylation is one of the most common and widespread types of RNA modifications, commonly found in ribosomal RNA (rRNA), transfer RNA (tRNA), small nucleolar RNA (snRNA) and messenger RNA (mRNA). This modification includes methylation of the 2′-OH part of ribose (enzymatic transfer of a methyl CH3-group from a methyl donor to an RNA substrate), and can therefore occur in all four nucleotides and other non-classical nucleotides. Since this methylation modification (denoted as "m") can occur on the ribose 2'-OH of any nucleotide residue among A / U / G / C (denoted as "N"), that is, it may be any of 2'-O-methyladenosine (Am), 2'-O-methylguanosine (Gm), 2'-O-methylcytidine (Cm) and 2'-O-methyluridine (Um), it is also called "Nm modification or 2OM modification". From a microscopic perspective, 2OM affects RNA in many ways, as it can increase the hydrophobicity of RNA, protect them from nuclease attacks, stabilize the helical structure, and affect their interactions with proteins or other RNAs. For example, the addition of 2OM modifications can disrupt the tertiary structure of RNA and inhibit RNA-protein interactions through steric effects or by affecting hydrogen bonds. From a macroscopic perspective, 2OM modifications are closely related to diseases, such as intellectual disability, cancer, etc. For example, the enzyme involved in the modification of rRNA 2OM is FTSJ2, which is one of the two paralogs of FTSJ1. FTSJ2 is associated with cell proliferation and is overexpressed in different cancer cell lines, especially in lung cancer cells.

[0003] Existing RNA2OM modification detection methods rely on complex biotechnological means, such as liquid chromatography-mass spectrometry (LC / MS) and two-dimensional thin layer chromatography (2D-TLC) and other classical biochemical methods, but these methods require the use of a lot of human resources, special tools, may damage RNA samples, and consume a relatively large amount of RNA samples. Later traditional machine models can sequence the ac4C sites of mRNA more conveniently and quickly, such as models based on support vector machines (SVMs), which use nucleotide chemical properties and nucleotide composition to encode RNA sequences to identify 2′-O-methylation sites; prediction models based on random forests (RF) and combined multiple encoding schemes; prediction models based on feature selection and extreme gradient boosting classifiers and support vector machines.

[0004] Therefore, most traditional 2OM modification detection methods rely entirely on manual feature extraction and traditional machine learning models, which are expensive, time-consuming, and have low accuracy. Summary of the invention

[0005] Based on this, in order to solve the above technical problems, a 2OM methylation modification site identification method and system based on feature selection and deep neural network are provided, which can improve the accuracy of 2OM methylation modification site identification and reduce costs.

[0006] A method for identifying 2OM methylation modification sites based on feature selection and deep neural network, the method comprising:

[0007] Collecting RNA sequences and converting the encoding mode of the RNA sequences to obtain sequence features;

[0008] The sequence features are spliced ​​to obtain multi-dimensional sequence features, and the multi-dimensional sequence features are input into the ANOVA feature selection algorithm;

[0009] The dimension of the multidimensional sequence feature is reduced by the ANOVA feature selection algorithm, and the statistical value of the variance analysis of the multidimensional sequence feature and the target variable is calculated, and feature selection is performed based on the statistical value to obtain input features;

[0010] For RNA sequences with modification sites on C, the input features are input into the first classifier, activated and pooled by the first classifier, and the classification results are output after being processed by the multi-head self-attention mechanism; for RNA sequences with modification sites on A, G, and U, the input features are input into the second classifier, data processing is performed by the Dropout layer and the linear layer in the second classifier, and the classification results are output after being processed by the Sigmoid layer;

[0011] The 2OM methylation modification site identification result is determined based on the classification result.

[0012] In one embodiment, the method further comprises:

[0013] Sequences containing 2OM sites were collected as data sets;

[0014] Based on the data set, a nucleotide window size of 20 bp was selected from the upstream and downstream of each 2OM site as a positive sample;

[0015] Based on the dataset, a nucleotide window size of 41 bp was randomly selected from the distal end of each 2OM site as a negative sample;

[0016] The positive samples and negative samples are converted into proportional data sets by downsampling, and the proportional data sets are divided into a training data set and a test data set.

[0017] In one embodiment, the method further comprises:

[0018] According to the different modified bases at the 2OM site, the ratio dataset was divided into four subsets: A, U, C, and G;

[0019] The RNA sequences whose modification sites are on C are the C subset; the RNA sequences whose modification sites are on A, G, and U are the A, G, and U subsets, respectively.

[0020] In one embodiment, the encoding mode of the RNA sequence is converted to obtain sequence features, including:

[0021] The RNA sequence text is converted into ANF, NCP, PseKNC, PseEIIP, PS2, DPCP_2, K-mer, and ENAC encoding to obtain the feature vector.

[0022] In one embodiment, calculating the statistical value of variance analysis of the multidimensional sequence features and the target variable by the ANOVA feature selection algorithm includes:

[0023] Dividing the multidimensional sequence features into a plurality of feature groups by the ANOVA feature selection algorithm;

[0024] Calculate the inter-group mean square and the intra-group mean square of each feature group;

[0025] According to the between-group mean square and the within-group mean square, a statistical value is calculated based on the target variable.

[0026] In one embodiment, for an RNA sequence with a modification site on C, the input feature is input into a first classifier, activated and pooled by the first classifier, and processed by a multi-head self-attention mechanism to output a classification result, including:

[0027] For the RNA sequence with the modification site on C, the input feature is input into the first classifier to enter the convolution layer, and the convolved data is output through the output channel;

[0028] The convolved data is processed by ReLU activation and maximum pooling to output a tensor;

[0029] The output tensor is input into the multi-head self-attention mechanism and processed through the fully connected layer to obtain the classification result.

[0030] In one embodiment, for RNA sequences with modification sites on A, G, and U, the input features are input into a second classifier, data is processed by a Dropout layer and a linear layer in the second classifier, and a classification result is output after processing by a Sigmoid layer, including:

[0031] For RNA sequences with modification sites on A, G, and U, the input features are input into a second classifier, and a discard ratio is set in a Dropout layer in the second classifier;

[0032] The input features pass through the Dropout layer, the input features are processed based on the discard ratio, and the processed features are input to the linear layer and the Sigmoid layer for processing to obtain the classification results.

[0033] In one embodiment, determining the 2OM methylation modification site identification result based on the classification result includes:

[0034] Obtaining a score threshold, and comparing the classification result with the score threshold to obtain a comparison result;

[0035] If the comparison result is that the classification result is higher than the score threshold, the 2OM methylation modification site recognition result is determined to be positive;

[0036] If the comparison result is that the classification result is not higher than the classification threshold, the 2OM methylation modification site identification result is determined to be negative.

[0037] A 2OM methylation modification site identification system based on feature selection and deep neural network, the system comprising:

[0038] A feature conversion module is used to collect RNA sequences and convert the encoding method of the RNA sequences to obtain sequence features;

[0039] A feature splicing module is used to splice the sequence features to obtain multi-dimensional sequence features, and input the multi-dimensional sequence features into the ANOVA feature selection algorithm;

[0040] A feature selection module, used to reduce the dimension of the multidimensional sequence features by using the ANOVA feature selection algorithm, calculate the statistical value of the variance analysis of the multidimensional sequence features and the target variable, perform feature selection based on the statistical value, and obtain input features;

[0041] A classification module, for RNA sequences with modification sites on C, inputting the input features into a first classifier, performing activation and pooling processing through the first classifier, and outputting classification results after processing through a multi-head self-attention mechanism; for RNA sequences with modification sites on A, G, and U, inputting the input features into a second classifier, performing data processing through a Dropout layer and a linear layer in the second classifier, and outputting classification results after processing through a Sigmoid layer;

[0042] An identification module is used to determine the 2OM methylation modification site identification result based on the classification result.

[0043] The above-mentioned 2OM methylation modification site identification method and system based on feature selection and deep neural network reduces the dimension of the input features through the ANOVA feature selection algorithm, selects features based on variance analysis, and uses different classifiers to process the 2OM site data at different positions, thereby improving the identification accuracy of 2OM methylation modification sites. BRIEF DESCRIPTION OF THE DRAWINGS

[0044] Figure 1 This is an application environment diagram of a 2OM methylation modification site identification method based on feature selection and deep neural network in one embodiment;

[0045] Figure 2 It is a schematic diagram of a process of identifying 2OM methylation modification sites based on feature selection and deep neural network in one embodiment;

[0046] Figure 3 A schematic diagram of a process for identifying 2OM methylation modification sites based on feature selection and deep neural network in another embodiment;

[0047] Figure 4 A structural diagram of a 2OM prediction model based on feature selection and deep neural network in one embodiment;

[0048] Figure 5 It is a structural block diagram of a 2OM methylation modification site identification system based on feature selection and deep neural network in one embodiment;

[0049] Figure 6 It is a structural block diagram of a 2OM methylation modification site identification system based on feature selection and deep neural network in another embodiment;

[0050] Figure 7 A box plot of the average ACC performance of three methods on four data sets in one embodiment;

[0051] Figure 8 FIG. 4 is a diagram showing the internal structure of a computer device in one embodiment. DETAILED DESCRIPTION

[0052] In order to make the purpose, technical solution and advantages of the present application more clearly understood, the present application is 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 present application and are not used to limit the present application.

[0053] It is understood that the terms "first", "second", etc. used in this application may be used herein to describe classifiers, but these classifiers are not limited by these terms. These terms are only used to distinguish a first classifier from another classifier. For example, without departing from the scope of this application, a first classifier may be referred to as a second classifier, and similarly, a second classifier may be referred to as a first classifier. Both the first classifier and the second classifier are classifiers, but they are not the same classifier.

[0054] The 2OM methylation modification site identification method based on feature selection and deep neural network provided in the embodiment of the present application can be applied to Figure 1 In the application environment shown. Figure 1 As shown, the application environment includes a computer device 110. The computer device 110 can collect RNA sequences and convert the encoding method of the RNA sequences to obtain sequence features; the computer device 110 can splice the sequence features to obtain multi-dimensional sequence features, and input the multi-dimensional sequence features into the ANOVA feature selection algorithm; the computer device 110 can reduce the dimension of the multi-dimensional sequence features through the ANOVA feature selection algorithm, and calculate the statistical value of the variance analysis of the multi-dimensional sequence features and the target variable, and select features based on the statistical value to obtain input features; the computer device 110 can input the input features into the first classifier for RNA sequences with modification sites on C, activate and pool them through the first classifier, and output the classification results after processing by the multi-head self-attention mechanism; for RNA sequences with modification sites on A, G, and U, input the input features into the second classifier, perform data processing through the Dropout layer and the linear layer in the second classifier, and output the classification results after processing by the Sigmoid layer; the computer device 110 can determine the 2OM methylation modification site recognition results based on the classification results. The computer device 110 may be, but is not limited to, various personal computers, laptop computers, smart phones, robots, and other devices.

[0055] In one embodiment, Figure 2 As shown, a method for identifying 2OM methylation modification sites based on feature selection and deep neural network is provided, comprising the following steps:

[0056] Step 202: collect RNA sequences and convert the encoding method of the RNA sequences to obtain sequence features.

[0057] Among them, the collected RNA sequence contains ribose 2′-O-methylation modification sites, namely RNA 2OM modification sites. Before identifying 2OM methylation modification sites, the RNA sequence text can be converted into ANF, NCP, PseKNC, PseEIIP, PS2, DPCP_2, K-mer and ENAC encoding methods to obtain sequence features.

[0058] In one embodiment, a method for identifying 2OM methylation modification sites based on feature selection and deep neural network is provided, which may also include a process of obtaining a data set, and the specific process includes: collecting a sequence containing a 2OM site as a data set; based on the data set, selecting a nucleotide window size of 20bp from the upstream and downstream of each 2OM site as a positive sample; based on the data set, randomly selecting a nucleotide window size of 41bp from the far end of each 2OM site as a negative sample; converting the positive samples and negative samples into a proportional data set by downsampling, and dividing the proportional data set into a training data set and a test data set.

[0059] Specifically, in this embodiment, the dataset can retrieve the original 2OM position information in RMBase and experimental data from Nm-seq, integrate 4795 2OM sites in the human genome with 2802 new 2OM sites in the Nm-seq experiment, and obtain a total of 7597 2OM sites. These positions are widely distributed in CDS, 3'-UTR, 5'-UTR, introns, exons and intergenic regions, covering almost all types of RNA, including tRNAs, rRNAs, scRNAs, scaRNAs, SnRNAs, snoRNAs, lincRNAs, protein-coding genes and pseudogenes, etc.

[0060] Among them, in order to be consistent with the format of RMBase, a 20bp nucleotide window size can be selected from the upstream and downstream of each 2OM site as a positive sample, and a negative sample of the same length (41bp) can be randomly selected from the distal end of each 2OM site at a ratio of approximately 1:3. Through this random sampling process, bias against specific regions of RNA is avoided. Specifically, considering that the redundancy of sequence data may affect the accuracy of the model, in this embodiment, the CD-HIT tool can be used to cluster the sequences, and the threshold is set to 80%; finally, a total of 6091 positive samples and 21520 negative samples were obtained, and converted to a 1:1 ratio data set by downsampling.

[0061] In one embodiment, a 2OM methylation modification site identification method based on feature selection and deep neural network is provided, which may also include a process of dividing a data set, and the specific process includes: according to the different modified bases of the 2OM site, the proportional data set is divided into four subsets of A, U, C, and G; the RNA sequence with the modification site on C is the C subset; the RNA sequences with the modification sites on A, G, and U are the A, G, and U subsets respectively.

[0062] That is, in this embodiment, according to the different modified bases at the 2OM site, the data set is divided into four subsets: A, U, C, and G. Finally, each subset is randomly divided into a training data set and a test data set in a ratio of 7:3.

[0063] In one embodiment, a 2OM methylation modification site identification method based on feature selection and deep neural network is provided, which may also include a process of converting the encoding method, and the specific process includes converting the RNA sequence text into the encoding methods of ANF, NCP, PseKNC, PseEIIP, PS2, DPCP_2, K-mer, and ENAC to obtain a feature vector.

[0064] Specifically, in this embodiment, the sequence is converted into PS2 encoding: position specificity of two nucleotides (PS2), PS2 is based on one-hot encoding of two nucleotides, that is, second-order features, and all adjacent nucleotide pairs are regarded as features, such as AA, AC, AG, etc. Therefore, there are 4×4=16 pairs of adjacent nucleotides. For the convenience of representation, the nucleotide pairs of the sequence are encoded as the number of corresponding nucleotide pairs in the sequence: for example, for the RNA sequence AACGUUU, through this encoding method, AA is encoded as 1, AC is encoded as 1, CG is encoded as 1, GU is encoded as 1, and UU is encoded as 2, but in fact, there are 16 nucleotide pairs combined in pairs, and finally a 16-dimensional feature vector will be obtained. Assuming RNA sequence x, it can be expressed by the formula: PS2(x)=(Count(AA), Count(AC), ..., Count(UU)), where Count() is used to count the number of nucleotide pairs in the RNA sequence.

[0065] Among them, the sequence is converted into PseKNC encoding: PseKNC (Pseudo K-tuple Nucleotide Composition) is a feature extraction method for DNA, RNA or protein sequences, mainly used to describe the distribution of k-mers in biological sequences in order to better capture the local structural information of the sequence. First, for a given RNA sequence S, which contains L bases, it is divided into non-overlapping subsequences of length k, called k-mers. K-tuple Nucleotide Composition (KNC) is a method to describe the combination and frequency of these k-mers. For N types of k-mers, count the number of times it appears in the sequence and construct a frequency distribution vector, denoted as: {X1, X2, ..., X N}; Calculate the frequency of each k-mer in the sequence and get the frequency vector: Among them, Pi is the frequency of the i-th k-mer; secondly, PseKNC improves the expressiveness of sequence features by introducing "pseudo-components", including two aspects of pseudo-components: one is obtained by the weight factor ωi of each k-mer frequency: The second is the correlation vector obtained by the correlation between different k-mers: Finally, the frequency vector, weight factor and correlation factor of KNC are combined to obtain the PseKNC feature vector:

[0066] Convert the sequence into ANF code: ANF (Accumulated Nucleotide Frequency) is a method used to describe the characteristics of nucleic acid sequences. It means that in a given sequence set, for each base at a position, the cumulative frequency of different bases at that position is calculated. Assume d i is the nucleotide n at position i i The digital mapping can be expressed as follows: Among them, |N i | means from the first position to the i-th position.

[0067] Convert the sequence into NCP code: NCP code means that in the RNA sequence fragment, the four nucleotides can be divided into three categories according to their properties, namely ring structure, functional group and hydrogen bond, as shown in the following table:

[0068]

[0069] Specifically, A and G have a double ring structure in chemical structure, while C and U have a single ring structure; A and C belong to amino groups, while G and U belong to keto groups; A and U form a weak hydrogen bond, while C and G can form a strong hydrogen bond. Therefore, according to these chemical properties, A, U, C and G can be represented by coordinates (1,1,1), (0,0,1), (0,1,0) and (1,0,0), respectively, and each input sample sequence can be encoded into a vector of length 41×3=123.

[0070] Converting sequences into PseEIIP code: The method using electron-ion interaction pseudopotential (EIIP) values ​​is used to encode RNA, DNA or protein sequences, which involves evaluating the energy of delocalized electrons in a nucleotide or amino acid sequence. When applied to RNA sequences, specific EIIP values ​​are assigned: A has a value of 0.1260, C has a value of 0.1340, G has a value of 0.0806, and U has a value of 0.1335. Based on the EIIP code, the PseEIIP code can be expressed as: PseEIIP = [(EIIP xyz ×f(NA xyz )) (i) ,…,(EIIP xyz ×f(NA xyz )) (T) ]; EIIP xyz =EIIP x +EIIP y +EIIP z ; Among them, f(NA xyz ) represents the normalized frequency of trinucleotides, NA xyz ∈{AAA, AAC, AAG, ..., UUU}, i = 1, 2, 3, ..., T, and T represents the total number of trinucleotides.

[0071] Convert the sequence into ENAC (Enhanced Nucleotide Acid Composition) encoding: Based on the frequency of nucleotide composition, a sliding window method is used to capture local sequence information. Assuming the RNA sequence x, the calculation formula of ENAC is expressed as: Among them, N i represents one of the four nucleotides, and W represents the window size.

[0072] Convert the sequence into DPCP_2 code: DPCP_2 code is an improved version of DPCP_1 code. It is a method to represent the characteristics of RNA based on the frequency of occurrence of two nucleotide combinations, and use this as the feature vector of the RNA sequence. Assuming that the length of the RNA sequence is L, the calculation formula of DPCP_2 can be expressed as: Among them, NAx , NA y ∈{AA,AC,AG,...,UU},m=1,2,...,(L-1),n=(m+1),...,L, represents the ith physicochemical property of the nucleotide pair.

[0073] Convert the sequence into K-mer encoding: The basic idea of ​​K-mer encoding is to divide the RNA sequence into subsequences of length k, called k-mers. For each k-mer, calculate its frequency of occurrence in the RNA sequence, and use these frequencies as elements of the feature vector, so that the RNA sequence can be represented as a feature vector of fixed length for training and prediction of machine learning models. Specifically, the RNA sequence x can first be divided into non-overlapping subsequences of length k, namely k-mers; for each k-mer, calculate its number of occurrences or frequency in the RNA sequence; use the number of occurrences or frequency of each k-mer as an element of the feature vector; if necessary, the feature vector can be normalized or other preprocessing operations can be performed; assuming that the length of the RNA sequence x is L, the feature vector of the k-mer encoding can be expressed as: Where K-mer(x) is the k-mer encoding of RNA sequence x, f i is the frequency of occurrence of the i-th k-mer in the RNA sequence. In this example, k can be set to 3, 4, 5, and 4 3 +4 4 +4 5 =1344-dimensional feature vector.

[0074] Step 204 , concatenate the sequence features to obtain multi-dimensional sequence features, and input the multi-dimensional sequence features into the ANOVA feature selection algorithm.

[0075] After converting the sequence into 8 features, we can concatenate them to get data with a feature dimension of 1706, which can be directly input into the feature selection algorithm. Among them, the ANOVA feature selection algorithm is based on the method of analysis of variance (Analysis of Variance). In the one-way analysis of variance, we calculate whether the variances between different groups are significantly different, and then judge whether the influence of different factors on the results is significant.

[0076] Step 206, reducing the dimension of the multidimensional sequence features by using the ANOVA feature selection algorithm, calculating the statistical value of the variance analysis of the multidimensional sequence features and the target variable, performing feature selection based on the statistical value, and obtaining input features.

[0077] In one embodiment, a 2OM methylation modification site identification method based on feature selection and deep neural network is provided, which may also include a process of using an ANOVA feature selection algorithm, and the specific process includes: dividing the multidimensional sequence features into several feature groups through the ANOVA feature selection algorithm; calculating the inter-group mean square and the intra-group mean square of each feature group; and calculating the statistical value based on the target variable according to the inter-group mean square and the intra-group mean square.

[0078] Specifically, in this embodiment, for feature X i and the target variable Y i The F value of the variance analysis statistic can be calculated, and its calculation formula can be expressed as: Among them, MS b is the mean square between groups, MS w It is the mean square within groups. In ANOVA, we need to pay attention to whether the mean square between groups is significantly greater than the mean square within groups, that is, MS b >MS w By calculating the F-value of different features, the strength of the relationship between them and the target variable can be evaluated.

[0079] Step 208, for RNA sequences with modification sites on C, input features to the first classifier, perform activation and pooling processing through the first classifier, and output classification results after processing through a multi-head self-attention mechanism; for RNA sequences with modification sites on A, G, and U, input features to the second classifier, perform data processing through the Dropout layer and the linear layer in the second classifier, and output classification results after processing through a Sigmoid layer.

[0080] Among them, the first classifier can be a type of deep learning neural network, which is mainly used to process and analyze data such as images and sequences. The main feature is to use convolutional layers to efficiently capture local features, thereby achieving efficient learning and representation of data. In addition, in order to make the network more expressive and robust and reduce the number of parameters, in this embodiment, activation layers and pooling layers can be interspersed between multiple convolutional layers, and the convolutional layers, activation layers and pooling layers are stacked together to form a convolutional neural network module in this embodiment.

[0081] There are a total of 3 convolutional layers in the MHCNN module in this embodiment, and the convolution output layers are 16, 32, and 16 respectively, the padding is 2, and the convolution kernel size is 3. Among them, a multi-head self-attention mechanism is also used in the convolutional network. Unlike a single attention mechanism that only performs weighted processing on the input sequence from one perspective, the multi-head self-attention designs multiple attention "heads" running in parallel. Each head focuses on different subspace features of the input sequence through different linear transformations. By independently calculating the attention weights and then splicing the results of each head, it is possible to capture multiple association patterns and contextual information of the sequence data at the same time. Specifically, convolution is good at extracting continuous nucleotide local pattern features, and the self-attention module further enhances this ability, which is of great significance for understanding complex RNA secondary structures and predicting their potential biological functions. Among them, the number of multi-heads in this embodiment is set to 8.

[0082] In one embodiment, a 2OM methylation modification site identification method based on feature selection and deep neural network is provided, which may also include a process of obtaining a classification result through a first classifier, and the specific process includes: for an RNA sequence with a modification site on C, inputting the input feature into the first classifier to enter the convolution layer, and outputting the convolved data through the output channel; subjecting the convolved data to ReLU activation and maximum pooling processing, and outputting a tensor; inputting the output tensor into a multi-head self-attention mechanism, and obtaining a classification result after processing through a fully connected layer.

[0083] Specifically, in this embodiment, the site data of Cm will enter the first classifier after feature selection, and the specific steps include:

[0084] Step 1: The data after ANOVA feature selection is used as input;

[0085] Step 2: Input the PseKNC vector into the first Conv layer. The sequence is originally 1 channel and the output channel dimension is 16.

[0086] Step 3: After ReLU activation and maximum pooling processing;

[0087] Step 4: Input the tensor output from step 3 into the second Conv layer, with 16 input channels and 32 output channels.

[0088] Step 5: After ReLU activation and maximum pooling processing;

[0089] Step 6: Input the tensor output from step 5 into the second Conv layer, with 32 input channels and 16 output channels.

[0090] Step 7: After ReLU activation and maximum pooling processing;

[0091] Step 8: The output tensor is input into the multi-head self-attention, and the number of heads is set to 8;

[0092] Step 9: Concatenate the tensor of the integer code after a series of processing and the tensor output in step 8 and input them into the fully connected layer to finally obtain the classification result.

[0093] In one embodiment, a 2OM methylation modification site identification method based on feature selection and deep neural network is provided, and the process of obtaining a classification result through a second classifier is also provided. The specific process includes: for RNA sequences with modification sites on A, G, and U, the input features are input into the second classifier, and a discarding ratio is set in the Dropout layer in the second classifier; the input features pass through the Dropout layer, the input features are processed based on the discarding ratio, and the processed features are input into the linear layer and the Sigmoid layer for processing to obtain the classification results.

[0094] Among them, the second classifier has a powerful representation learning ability, which can mine deep and abstract 2OM site features in RNA from the features after feature selection, which helps to distinguish subtle sequence differences and hidden biochemical characteristics.

[0095] Specifically, in this embodiment, the 2OM site data at Am, Gm, and Um will be input into the second classifier, and the specific steps include:

[0096] Step 1: After the Dropout layer, set the dropout ratio to 0.4;

[0097] Step 2, pass through the linear layer (1024*128);

[0098] Step 3: After the Dropout layer, set the dropout ratio to 0.3;

[0099] Step 4, pass through the linear layer (128*1);

[0100] Step 5: Get the final classification result through the Sigmoid layer.

[0101] The classification results are obtained by the first classifier and the second classifier based on the 2OM site data of Cm, Am, Gm, and Um, respectively, so that the 2OM methylation modification site identification results can be determined.

[0102] Step 210, determining the 2OM methylation modification site identification result based on the classification result.

[0103] In one embodiment, a 2OM methylation modification site identification method based on feature selection and deep neural network is provided, which may also include a process for determining the 2OM methylation modification site identification result, and the specific process includes: obtaining a score threshold, comparing the classification result with the score threshold, and obtaining a comparison result; if the comparison result is that the classification result is higher than the score threshold, then determining that the 2OM methylation modification site identification result is positive; if the comparison result is that the classification result is not higher than the classification threshold, then determining that the 2OM methylation modification site identification result is negative.

[0104] In this embodiment, the score threshold may be 0.5, and a comparison result higher than 0.5 may be determined as a positive result for identifying the 2OM methylation modification site; and a comparison result lower than 0.5 may be determined as a negative result for identifying the 2OM methylation modification site.

[0105] In one embodiment, Figure 3 As shown, a method for identifying 2OM methylation modification sites based on feature selection and deep neural network is provided, and the specific process includes:

[0106] Input RNA sequence;

[0107] According to RNA sequences, 8 features were obtained, including ANF, NCP, PseKNC, PseEIIP, PS2, DPCP_2, K-mer, and ENAC;

[0108] Use the ANOVA feature selection algorithm;

[0109] Input the RNA sequence with the modification site on C into the first classifier and output the recognition result;

[0110] Input the RNA sequence with modification sites on A, G, and U into the second classifier and output the recognition result;

[0111] Thus, the final 2OM methylation modification site identification results are obtained.

[0112] In one embodiment, a 2OM methylation modification site identification method based on feature selection and deep neural network can be implemented by a designed 2OM prediction model (DeepR2OM) based on feature selection and deep neural network, wherein the structure of the model is as follows: Figure 4 As shown, the model combines 8 RNA features, ANOVA feature selection, and two neural network feature classifiers, which can effectively extract and utilize local and global features of RNA sequences.

[0113] like Figure 4As shown, the DeepR2OM model can include three parts: data preparation, feature processing, classifier and results. Among them, feature processing can be performed using the ANOVA feature selection algorithm, and two classifiers are set to process RNA sequences at different positions of the modification site respectively, and the final 2OM methylation modification site recognition results are obtained according to the output recognition results. That is, firstly, the RNA sequence text is converted into the encoding methods of ANF, NCP, PseKNC, PseEIIP, PS2, DPCP_2, K-mer and ENAC; then, the feature dimension is reduced using the ANOVA feature selection algorithm; for RNA sequences with modification sites on A, G, and U, after feature selection, they will enter the deep neural network to obtain the recognition results; for RNA sequences with modification sites on C, the convolutional neural network with multi-head self-attention will be integrated to obtain the final recognition results.

[0114] It should be understood that, although the steps in the above-mentioned flowcharts are shown in sequence according to the indication of the arrows, these steps are not necessarily executed in sequence according to the order indicated by the arrows. Unless there is a clear explanation in this article, the execution of these steps is not strictly limited in order, and these steps can be executed in other orders. Moreover, at least a part of the steps in the above-mentioned flowcharts may include a plurality of sub-steps or a plurality of stages, and these sub-steps or stages are not necessarily executed at the same time, but can be executed at different times, and the execution order of these sub-steps or stages is not necessarily to be carried out in sequence, but can be executed in turn or alternately with other steps or at least a part of the sub-steps or stages of other steps.

[0115] In one embodiment, Figure 5 As shown, a 2OM methylation modification site recognition system based on feature selection and deep neural network is provided, including: a feature conversion module 510, a feature splicing module 520, a feature selection module 530, a classification module 540 and a recognition module 550, wherein:

[0116] The feature conversion module 510 is used to collect RNA sequences and convert the encoding method of the RNA sequences to obtain sequence features;

[0117] A feature concatenation module 520 is used to concatenate sequence features to obtain multi-dimensional sequence features, and input the multi-dimensional sequence features into the ANOVA feature selection algorithm;

[0118] A feature selection module 530 is used to reduce the dimension of the multi-dimensional sequence features by using an ANOVA feature selection algorithm, calculate the statistical value of the variance analysis of the multi-dimensional sequence features and the target variable, and perform feature selection based on the statistical value to obtain input features;

[0119] Classification module 540, for RNA sequences with modification sites on C, inputting input features into a first classifier, performing activation and pooling processing through the first classifier, and outputting classification results after processing through a multi-head self-attention mechanism; for RNA sequences with modification sites on A, G, and U, inputting input features into a second classifier, performing data processing through a Dropout layer and a linear layer in the second classifier, and outputting classification results after processing through a Sigmoid layer;

[0120] The identification module 550 is used to determine the 20M methylation modification site identification result based on the classification result.

[0121] In one embodiment, Figure 6 As shown, a feature selection and deep neural network 2OM methylation modification site identification system provided also includes: a data set determination module 560, which is used to collect sequences containing 2OM sites as data sets; based on the data set, a nucleotide window size of 20bp is selected from the upstream and downstream of each 2OM site as a positive sample; based on the data set, a nucleotide window size of 41bp is randomly selected from the far end of each 2OM site as a negative sample; positive samples and negative samples are converted into proportional data sets by downsampling, and the proportional data sets are divided into a training data set and a test data set.

[0122] In one embodiment, the data set determination module 560 is also used to divide the proportional data set into four subsets of A, U, C, and G according to the different modified bases at the 2OM site; the RNA sequence with the modified site on C is the C subset; the RNA sequences with the modified sites on A, G, and U are the A, G, and U subsets respectively.

[0123] In one embodiment, the feature conversion module 510 is also used to convert the RNA sequence text into ANF, NCP, PseKNC, PseEIIP, PS2, DPCP_2, K-mer, ENAC encoding to obtain a feature vector.

[0124] In one embodiment, the feature selection module 530 is also used to divide the multidimensional sequence features into several feature groups through the ANOVA feature selection algorithm; calculate the inter-group mean square and the intra-group mean square of each feature group; and calculate the statistical value based on the target variable according to the inter-group mean square and the intra-group mean square.

[0125] In one embodiment, the classification module 540 is also used to input the input features into the first classifier and enter the convolution layer for the RNA sequence with the modification site on C, and output the convolved data through the output channel; the convolved data is subjected to ReLU activation and maximum pooling processing to output a tensor; the output tensor is input into the multi-head self-attention mechanism, and the classification result is obtained after processing through the fully connected layer.

[0126] In one embodiment, the classification module 540 is also used to input the input features into the second classifier for RNA sequences with modification sites on A, G, and U, and set the discard ratio in the Dropout layer in the second classifier; the input features pass through the Dropout layer, the input features are processed based on the discard ratio, and the processed features are input into the linear layer and the Sigmoid layer for processing to obtain the classification results.

[0127] In one embodiment, the identification module 550 is also used to obtain a score threshold, compare the classification result with the score threshold, and obtain a comparison result; if the comparison result is that the classification result is higher than the score threshold, then the 2OM methylation modification site identification result is determined to be positive; if the comparison result is that the classification result is not higher than the classification threshold, then the 2OM methylation modification site identification result is determined to be negative.

[0128] In one embodiment, a method and system for identifying 2OM methylation modification sites based on feature selection and deep neural network is provided, and experimental data of the application of the system in the DeepR2OM model are provided, specifically:

[0129] The DeepR2OM model shows excellent performance in multiple evaluation indicators. Under the training validation set test, the recognition rate of Am is ACC 91.9%, RECALL is 89.4%, PRE is 93.9%, and MCC is 83.9%; the recognition rate of Um is ACC 84.4%, RECALL is 83.9%, PRE is 84.7%, and MCC is 68.8%; the recognition rate of Cm is ACC 88.5%, RECALL is 86.5%, PRE is 90.4%, and MCC is 78.8%; the recognition rate of Gm is ACC 89.4%, RECALL is 86.8%, PRE is 91.4%, and MCC is 78.8%. The average performance of the model under the training validation set is ACC 88.5%, RECALL 86.7%, PRE 90.1%, and MCC 77.1%.

[0130] Under the data of the independent test set, the recognition rate of Am is ACC 90.2, RECALL is 87.5%, PRE is 91.6%, and MCC is 82.4%; the recognition rate of Um is ACC 84.3%, RECALL is 82.1%, PRE is 82.4%, and MCC is 66.7; the recognition rate of Cm is ACC 86.7%, RECALL is 85.5%, PRE is 88.3%, and MCC is 77.0%; the recognition rate of Gm is ACC 87.4%, RECALL is 86.8%, PRE is 89.4%, and MCC is 75.7%. The average performance of the model under the training validation set is ACC 87.1%, RECALL 85.5%, PRE 87.9%, and MCC 75.7%. All four indicators are higher than the existing models. The box plot of the average ACC performance of the three methods on the four data sets is shown in the figure below. Figure 7 The specific data is shown in the following table:

[0131]

[0132] In one embodiment, a computer device is provided. The computer device may be a terminal, and its internal structure diagram may be as follows: Figure 8 As shown. The computer device includes a processor, a memory, a network interface, a display screen and an input device connected by a system bus. Among them, the processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and a computer program. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The network interface of the computer device is used to communicate with an external terminal through a network connection. When the computer program is executed by the processor, a 2OM methylation modification site identification method based on feature selection and deep neural network is implemented. The display screen of the computer device can be a liquid crystal display screen or an electronic ink display screen, and the input device of the computer device can be a touch layer covered on the display screen, or a button, trackball or touchpad set on the computer device housing, or an external keyboard, touchpad or mouse, etc.

[0133] Those skilled in the art will understand that Figure 8 The structure shown in the figure is only a block diagram of a part of the structure related to the solution of the present application, and does not constitute a limitation on the computer device to which the solution of the present application is applied. The specific computer device may include more or fewer components than those shown in the figure, or combine certain components, or have a different arrangement of components.

[0134] In one embodiment, a computer device is provided, including a memory and a processor, wherein a computer program is stored in the memory, and when the processor executes the computer program, the steps of a 2OM methylation modification site identification method based on feature selection and deep neural network are implemented.

[0135] In one embodiment, a computer-readable storage medium is provided, on which a computer program is stored. When the computer program is executed by a processor, the steps of a 2OM methylation modification site identification method based on feature selection and deep neural network are implemented.

[0136] Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be completed by instructing the relevant hardware through a computer program, and the computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to memory, storage, database or other media used in the embodiments provided in this application can include non-volatile and / or volatile memory. Non-volatile memory can include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM) or flash memory. Volatile memory can include random access memory (RAM) or external cache memory. As an illustration and not limitation, RAM is available in many forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link (Synchlink) DRAM (SLDRAM), memory bus (Rambus) direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM).

[0137] The technical features of the above embodiments may be arbitrarily combined. To make the description concise, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0138] The above-mentioned embodiments only express several implementation methods of the present application, and the descriptions thereof are relatively specific and detailed, but they cannot be understood as limiting the scope of the invention patent. It should be pointed out that, for a person of ordinary skill in the art, several variations and improvements can be made without departing from the concept of the present application, and these all belong to the protection scope of the present application. Therefore, the protection scope of the patent of the present application shall be subject to the attached claims.

Claims

1. A method for identifying 2OM methylation modification sites based on feature selection and deep neural network, characterized in that: The method comprises: Collecting RNA sequences and converting the encoding mode of the RNA sequences to obtain sequence features; The sequence features are spliced ​​to obtain multi-dimensional sequence features, and the multi-dimensional sequence features are input into the ANOVA feature selection algorithm; The dimension of the multidimensional sequence feature is reduced by the ANOVA feature selection algorithm, and the statistical value of the variance analysis of the multidimensional sequence feature and the target variable is calculated, and feature selection is performed based on the statistical value to obtain input features; For RNA sequences with modification sites on C, the input features are input into the first classifier, activated and pooled by the first classifier, and the classification results are output after being processed by the multi-head self-attention mechanism; for RNA sequences with modification sites on A, G, and U, the input features are input into the second classifier, data processing is performed by the Dropout layer and the linear layer in the second classifier, and the classification results are output after being processed by the Sigmoid layer; The 2OM methylation modification site identification result is determined based on the classification result.

2. The 2OM methylation modification site identification method based on feature selection and deep neural network according to claim 1, characterized in that: The method further comprises: Sequences containing 2OM sites were collected as data sets; Based on the data set, a nucleotide window size of 20 bp was selected from the upstream and downstream of each 2OM site as a positive sample; Based on the dataset, a nucleotide window size of 41 bp was randomly selected from the distal end of each 2OM site as a negative sample; The positive samples and negative samples are converted into proportional data sets by downsampling, and the proportional data sets are divided into a training data set and a test data set.

3. The 2OM methylation modification site identification method based on feature selection and deep neural network according to claim 2, characterized in that: The method further comprises: According to the different modified bases at the 2OM site, the ratio dataset was divided into four subsets: A, U, C, and G; The RNA sequences whose modification sites are on C are the C subset; the RNA sequences whose modification sites are on A, G, and U are the A, G, and U subsets, respectively.

4. The method for identifying 2OM methylation modification sites based on feature selection and deep neural network according to claim 1, characterized in that: The coding mode of the RNA sequence is converted to obtain sequence features, including: The RNA sequence text is converted into ANF, NCP, PseKNC, PseEIIP, PS2, DPCP_2, K-mer, and ENAC encoding to obtain the feature vector.

5. The method for identifying 2OM methylation modification sites based on feature selection and deep neural network according to claim 1, characterized in that: Calculating the statistical value of the variance analysis of the multidimensional sequence features and the target variable by the ANOVA feature selection algorithm includes: Dividing the multidimensional sequence features into a plurality of feature groups by the ANOVA feature selection algorithm; Calculate the inter-group mean square and the intra-group mean square of each feature group; According to the between-group mean square and the within-group mean square, a statistical value is calculated based on the target variable.

6. The method for identifying 2OM methylation modification sites based on feature selection and deep neural network according to claim 1, characterized in that: For the RNA sequence with the modification site on C, the input feature is input into the first classifier, activated and pooled by the first classifier, and processed by the multi-head self-attention mechanism to output the classification result, including: For the RNA sequence with the modification site on C, the input feature is input into the first classifier to enter the convolution layer, and the convolved data is output through the output channel; The convolved data is processed by ReLU activation and maximum pooling to output a tensor; The output tensor is input into the multi-head self-attention mechanism and processed through the fully connected layer to obtain the classification result.

7. The method for identifying 2OM methylation modification sites based on feature selection and deep neural network according to claim 1, characterized in that: For RNA sequences with modification sites on A, G, and U, the input features are input into the second classifier, data is processed by the Dropout layer and the linear layer in the second classifier, and the classification results are output after being processed by the Sigmoid layer, including: For RNA sequences with modification sites on A, G, and U, the input features are input into a second classifier, and a discard ratio is set in a Dropout layer in the second classifier; The input features pass through the Dropout layer, the input features are processed based on the discard ratio, and the processed features are input to the linear layer and the Sigmoid layer for processing to obtain the classification results.

8. The method for identifying 2OM methylation modification sites based on feature selection and deep neural network according to claim 1, characterized in that: Determining the 2OM methylation modification site identification result based on the classification result includes: Obtaining a score threshold, and comparing the classification result with the score threshold to obtain a comparison result; If the comparison result is that the classification result is higher than the score threshold, the 2OM methylation modification site recognition result is determined to be positive; If the comparison result is that the classification result is not higher than the classification threshold, the 2OM methylation modification site identification result is determined to be negative.

9. A 2OM methylation modification site recognition system based on feature selection and deep neural network, characterized in that: The system comprises: A feature conversion module is used to collect RNA sequences and convert the encoding method of the RNA sequences to obtain sequence features; A feature splicing module is used to splice the sequence features to obtain multi-dimensional sequence features, and input the multi-dimensional sequence features into the ANOVA feature selection algorithm; A feature selection module, used to reduce the dimension of the multidimensional sequence features by using the ANOVA feature selection algorithm, calculate the statistical value of the variance analysis of the multidimensional sequence features and the target variable, perform feature selection based on the statistical value, and obtain input features; A classification module, for RNA sequences with modification sites on C, inputting the input features into a first classifier, performing activation and pooling processing through the first classifier, and outputting classification results after processing through a multi-head self-attention mechanism; for RNA sequences with modification sites on A, G, and U, inputting the input features into a second classifier, performing data processing through a Dropout layer and a linear layer in the second classifier, and outputting classification results after processing through a Sigmoid layer; An identification module is used to determine the 2OM methylation modification site identification result based on the classification result.

10. The 2OM methylation modification site identification system based on feature selection and deep neural network according to claim 9, characterized in that: The system further comprises a data set determination module, which is used to collect sequences containing 2OM sites as data sets; Based on the data set, a nucleotide window size of 20bp is selected from the upstream and downstream of each 2OM site as a positive sample; based on the data set, a nucleotide window size of 41bp is randomly selected from the distal end of each 2OM site as a negative sample; the positive samples and negative samples are converted into a proportional data set by downsampling, and the proportional data set is divided into a training data set and a test data set.

Images