Bidirectional dinucleotide position-specific preference and point-wise mutual information encoding methods for DNA / RNA sequences

The bidirectional dinucleotide position-specific preference and point mutual information encoding method enhances DNA/RNA sequence feature extraction, addressing redundancy issues and improving methylation site recognition accuracy.

CN112365925BActive Publication Date: 2024-12-06SHAANXI NORMAL UNIV
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202011236138.3
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2020-11-09
Publication Date
2024-12-06
Estimated Expiration
2040-11-09

AI Technical Summary

Technical Problem

Existing DNA/RNA sequence encoding methods fail to effectively extract high-dimensional features with low redundancy, leading to low accuracy in methylation site recognition models.

Method used

A bidirectional dinucleotide position-specific preference and point mutual information encoding method is developed, constructing nucleotide and dinucleotide matrices to encode DNA/RNA sequences, reducing redundancy and enhancing feature extraction.

Benefits of technology

The method improves the accuracy and specificity of DNA/RNA methylation site recognition, outperforming existing methods in prediction and classification tasks.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN112365925B_ABST
    Figure CN112365925B_ABST
Patent Text Reader

Abstract

A method for encoding bidirectional dinucleotide position-specific preference and point mutual information of DNA / RNA sequences, comprising the steps of constructing a DNA / RNA sequence nucleotide position-specific preference matrix, constructing a DNA / RNA sequence bidirectional dinucleotide position-specific preference matrix, determining the point mutual information value of DNA / RNA sequence nucleotides, feature combination, and DNA / RNA sequence encoding. In order to extract more dinucleotide position information from DNA / RNA sequence data, a parameter α is introduced to represent the distance between dinucleotides, and the numerical feature vectors with different values ​​of α are combined into a global high-dimensional numerical feature vector. 6 A has very good performance in methylation site identification. The DNA / RNA numerical feature data obtained by the present invention has the advantages of more classification information, low redundancy between features, high recognition accuracy of the trained model, etc., and can be used for encoding DNA / RNA sequences.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The invention belongs to the technical field of sequence data analysis, and in particular relates to a DNA / RNA sequence encoding method. Background Art

[0002] Machine learning technology plays a key role in the identification of DNA / RNA methylation sites in the post-genomic era. Generally speaking, using machine learning technology to solve this problem mainly includes three steps: sequence data encoding, model construction and performance evaluation. Among them, sequence data encoding is the most important step to solve this problem. That is, how to extract numerical features containing more classification identification information from DNA / RNA sequence data is the key to accurately identify DNA / RNA methylation sites.

[0003] At present, the existing encoding methods have the defects of low encoding feature dimensionality and inability to extract key identification information from DNA / RNA sequence data, resulting in low accuracy of the established prediction and recognition model. Using multiple encoding methods to encode DNA / RNA sequence data and combining the encoded numerical features into high-dimensional features can overcome the shortcomings of a single encoding method, but it is bound to cause high redundancy of numerical features and waste of computing resources, and the improvement of the accuracy of DNA / RNA methylation site recognition models is very limited. Therefore, developing a DNA / RNA sequence data encoding method that effectively encodes DNA / RNA sequences into high-dimensional numerical features containing more classification and recognition information and has low redundancy between features is the key to solving this type of problem and is also a hot topic in this field. Summary of the invention

[0004] The technical problem to be solved by the present invention is to overcome the shortcomings of the above-mentioned existing technologies and provide a bidirectional dinucleotide position-specific preference and point mutual information DNA / RNA sequence encoding method with rich classification and recognition information, low redundancy between features, and high recognition and prediction accuracy of the established model.

[0005] The technical solution adopted to solve the above technical problems consists of the following steps:

[0006] (1) Construction of DNA / RNA sequence nucleotide position-specific preference matrix

[0007] Given a DNA / RNA sequence dataset D, the dataset consists of a positive class dataset and a negative class dataset.

[0008] The nucleotide position-specific preference matrix of the positive class dataset is determined as follows:

[0009]

[0010] A, C, G, and X are four nucleotides of DNA / RNA. X is represented by nucleotide T in the DNA sequence dataset and by nucleotide U in the RNA sequence dataset. i is the position of the nucleotide, 1≤i≤l, and the value of i is a finite positive integer. l is the nucleotide length of the DNA / RNA sequence sample, and the value of l is an odd number. are the occurrence frequencies of nucleotides A, C, G, and X at the i-th position of all sequence samples in the positive data set.

[0011] The nucleotide position-specific preference matrix of the negative class dataset is determined as follows:

[0012]

[0013] in, are the occurrence frequencies of nucleotides A, C, G, and X at the i-th position of all sequence samples in the negative dataset, respectively.

[0014] (2) Construction of a bidirectional dinucleotide position-specific preference matrix for DNA / RNA sequences

[0015] The forward dinucleotide position-specific preference matrix of the positive data set is determined as follows:

[0016]

[0017] Among them, AA, AC, ..., XX are 16 dinucleotides composed of the four nucleotides A, C, G, and X of DNA / RNA, α represents the distance between two nucleotides, 0≤α≤(l-3) / 2, the value of α is a finite positive integer, j is the position of the nucleotide, α+2≤j≤l-α-1, the value of j is a finite positive integer, are the occurrence frequencies of dinucleotides AA, AC, …, XX at the j-th and j+α+1-th positions of all sequence samples in the positive data set, respectively.

[0018] The backward dinucleotide position-specific preference matrix of the positive class dataset is determined as follows:

[0019]

[0020] in, are the occurrence frequencies of dinucleotides AA, AC, …, XX at the j-th and j-α-1-th positions of all sequence samples in the positive data set, respectively.

[0021] The forward dinucleotide position-specific preference matrix of the negative class data set is determined as follows:

[0022]

[0023] in, are the occurrence frequencies of dinucleotides AA, AC, …, XX at the j-th and j+α+1-th positions of all sequence samples in the negative data set, respectively.

[0024] The backward dinucleotide position-specific preference matrix for the negative class dataset is determined as follows:

[0025]

[0026] in, are the occurrence frequencies of dinucleotides AA, AC, …, XX at the j-th and j-α-1-th positions of all sequence samples in the negative data set, respectively.

[0027] (3) Determine the point mutual information value of DNA / RNA sequence nucleotides

[0028] (3.1) Determine the forward point mutual information value of the nucleotides of the DNA / RNA sequence to be encoded in the positive class data set as follows

[0029]

[0030] Where x is the nucleotide at the jth position, x∈{A,C,G,X}, z is the nucleotide at the j+α+1th position, z∈{A,C,G,X}, is the frequency of occurrence of dinucleotide xz at the jth and j+α+1th positions of all sequence samples in the positive dataset, is the frequency of occurrence of nucleotide x at the jth position of all sequence samples in the positive dataset, is the frequency of occurrence of nucleotide z at the j+α+1th position of all sequence samples in the positive data set.

[0031] Determine the backward point mutual information value of the nucleotides of the DNA / RNA sequence to be encoded in the positive data set as follows:

[0032]

[0033] Where y is the nucleotide at position j-α-1, y∈{A,C,G,X}, is the frequency of occurrence of dinucleotide xy at the jth and j-α-1th positions of all sequence samples in the positive data set, is the frequency of occurrence of nucleotide y at the j-α-1th position of all sequence samples in the positive data set.

[0034] Point mutual information encoding value of the nucleotide at the jth position of the DNA / RNA sequence sample to be encoded in the positive class data set Defined as the forward point mutual information value and the backward point mutual information value The mean of the DNA / RNA sequence samples of length l is encoded as the point mutual information feature vector V with a length of l-2α-2 + :

[0035]

[0036]

[0037] (3.2) Determine the forward point mutual information value of the nucleotides of the DNA / RNA sequence to be encoded in the negative class data set as follows

[0038]

[0039] in, is the frequency of occurrence of dinucleotide xz at the jth and j+α+1th positions of all sequence samples in the negative dataset, is the frequency of occurrence of nucleotide x at the jth position of all sequence samples in the negative dataset, is the frequency of occurrence of nucleotide z at the j+α+1th position of all sequence samples in the negative dataset.

[0040] The backward point mutual information value of the nucleotides of the DNA / RNA sequence to be encoded in the negative data set is determined as follows:

[0041]

[0042] in, is the frequency of occurrence of dinucleotide xy at the jth and j-α-1th positions of all sequence samples in the negative dataset, is the frequency of occurrence of nucleotide y at the j-α-1th position of all sequence samples in the negative dataset.

[0043] Point mutual information encoding value of the nucleotide at the jth position of the DNA / RNA sequence sample to be encoded in the negative class data set Defined as the forward point mutual information value and the backward point mutual information value The mean of the DNA / RNA sequence samples of length l is encoded as the point mutual information feature vector V with a length of l-2α-2 - :

[0044]

[0045]

[0046] (3.3) Given a DNA / RNA sequence sample of length l to be encoded, we can use the vector V + and V - Subtract the corresponding elements to determine its eigenvector V:

[0047] V=[V α+2 ,V α+3 ,…,V j ]

[0048]

[0049] (4) Feature combination

[0050] When the parameter α is 0, the eigenvector V(0) is [V2, V3, V4, …, V l-2 ,V l-1 ], the number of elements is l-2, when α is 1, the eigenvector V(1) is [V3,V4,V5,…,V l-3 ,V l-2 ], the number of elements is l-4, ..., α is (l-5) / 2, the eigenvector V((l-5) / 2) is [V (l-1) / 2 ,V (l+1) / 2 ,V (l+3) / 2 ], the number of elements is 3, and when α is (l-3) / 2, the eigenvector V((l-3) / 2) is [V (l+1) / 2 ], the number of elements is 1; the eigenvectors determined by different values ​​of parameter α are combined into an element number of (l-1) 2 / 4 high-dimensional feature vector [V(0),V(1),…,V((l-5) / 2),V((l-3) / 2)].

[0051] (5) DNA / RNA sequence encoding

[0052] Using the above steps (1) to (4), the DNA / RNA sequence data set D is encoded into a numerical data set D'. s is the number of samples in the numerical data set D'. The value of s is a finite positive integer, (l-1) 2 / 4 is the number of features of the numerical data set D'.

[0053] In order to encode DNA / RNA sequence data into numerical data containing more classification identification information and low redundancy between features, the present invention proposes a bidirectional dinucleotide position-specific preference method. Based on the nucleotide position-specific preference matrix and the bidirectional dinucleotide position-specific preference matrix of positive and negative DNA / RNA sequence data, the DNA / RNA sequence samples are encoded into numerical feature samples using the point mutual information method. In order to extract more dinucleotide position information from the DNA / RNA sequence samples, the parameter α is introduced in the process of constructing the bidirectional dinucleotide position-specific preference matrix to represent the dinucleotide spacing, and the numerical feature vectors determined by different values ​​of α are combined into a global high-dimensional numerical feature vector containing more classification identification information and low redundancy between features. The DNA / RNA sequence encoding method of the present invention is used to conduct a comparative simulation experiment with 7 existing encoding methods. The experimental results show that the support vector machine model established by the encoding method of the present invention has a good performance in the N-dimensional encoding of Caenorhabditis elegans DNA. 4 -methylcytosine (N 4 The accuracy, sensitivity, specificity, MCC, AUROC and AUPRC of site recognition prediction of 4-methylcytosine (4mC) were 0.905, 0.902, 0.909, 0.811, 0.967 and 0.966 respectively, which were higher than those of the other 7 comparative encoding methods. The support vector machine established based on the encoding method of the present invention predicted the N 6 -methyladenosine (N 6 Methyladenosine, m 6 A) The accuracy, sensitivity, specificity, MCC, AUROC, and AUPRC of site recognition prediction were 0.905, 0.916, 0.894, 0.810, 0.968, and 0.967, respectively, which were higher than those of the other seven comparative coding methods. BRIEF DESCRIPTION OF THE DRAWINGS

[0054] Figure 1 It is a flow chart of the method of embodiment 1 of the present invention.

[0055] Figure 2 The N-coding of DNA of Caenorhabditis elegans by the support vector machine model established based on the present invention and 7 encoding methods 4 -AUROC value curve for the prediction of methylcytosine site identification.

[0056] Figure 3 The support vector machine model established based on the present invention and 7 encoding methods is used to calculate the N of Caenorhabditis elegans DNA. 4 -AUPRC value curves predicted by methylcytosine site identification.

[0057] Figure 4The support vector machine model established based on the present invention and 7 encoding methods is used to calculate the N of yeast RNA. 6 -AUROC value curve for the prediction of methyladenosine site identification.

[0058] Figure 5 The support vector machine model established based on the present invention and 7 encoding methods is used to calculate the N of yeast RNA. 6 -AUPRC value curve predicted by methyladenosine site identification. DETAILED DESCRIPTION

[0059] The present invention will be further described in detail below in conjunction with the accompanying drawings and examples, but the present invention is not limited to the following embodiments.

[0060] Example 1

[0061] Based on the literature iDNA4mC:identifying DNA N 4 -methylcytosine sites based on nucleotide chemical properties of Caenorhabditis elegans DNA 4 -methylcytosine (N 4 -methylcytosine, 4mC) data set as an example. This data set has 3108 DNA sequence samples, among which the number of samples in the positive data set, i.e. the real N 4 -The number of methylcytosine samples is 1554, and the number of negative class data set samples is non-N 4 The number of -methylcytosine samples is 1554, and the length l of each sequence sample is 41. The bidirectional dinucleotide position-specific preference and point mutual information DNA encoding method of this embodiment consists of the following steps (see Figure 1 ):

[0062] (1) Construction of DNA sequence nucleotide position-specific preference matrix

[0063] Given a DNA sequence dataset D, the dataset consists of a positive class dataset and a negative class dataset.

[0064] The nucleotide position-specific preference matrix of the positive class dataset is determined as follows:

[0065]

[0066] Wherein, A, C, G, and T are four nucleotides of DNA, i is the position of the nucleotide, 1≤i≤l, the value of i is a finite positive integer, l is the nucleotide length of the DNA sequence sample, the value of l is an odd number, and the value of l in this embodiment is 41. are the occurrence frequencies of nucleotides A, C, G, and T at the i-th position of all sequence samples in the positive data set.

[0067] The nucleotide position-specific preference matrix of the negative class dataset is determined as follows:

[0068]

[0069] in, are the occurrence frequencies of nucleotides A, C, G, and T at the i-th position of all sequence samples in the negative dataset, respectively.

[0070] (2) Construction of a DNA sequence bidirectional dinucleotide position-specific preference matrix

[0071] The forward dinucleotide position-specific preference matrix of the positive data set is determined as follows:

[0072]

[0073] Wherein, AA, AC, ..., TT are 16 dinucleotides composed of the four nucleotides A, C, G, and T of DNA, α represents the distance between two nucleotides, 0≤α≤(l-3) / 2, the value of α is a finite positive integer, 0≤α≤19 in this embodiment, j is the position of the nucleotide, α+2≤j≤l-α-1, α+2≤j≤40-α in this embodiment, the value of j is a finite positive integer, are the occurrence frequencies of dinucleotides AA, AC, …, TT at the j-th and j+α+1-th positions of all sequence samples in the positive data set, respectively.

[0074] The backward dinucleotide position-specific preference matrix of the positive class dataset is determined as follows:

[0075]

[0076] in, are the occurrence frequencies of dinucleotides AA, AC, …, TT at the j-th and j-α-1-th positions of all sequence samples in the positive data set, respectively.

[0077] The forward dinucleotide position-specific preference matrix of the negative class data set is determined as follows:

[0078]

[0079] in, are the occurrence frequencies of dinucleotides AA, AC, …, TT at the j-th and j+α+1-th positions of all sequence samples in the negative data set, respectively.

[0080] The backward dinucleotide position-specific preference matrix for the negative class dataset is determined as follows:

[0081]

[0082] in, are the occurrence frequencies of dinucleotides AA, AC, …, TT at the j-th and j-α-1-th positions of all sequence samples in the negative data set, respectively.

[0083] (3) Determine the point mutual information value of DNA sequence nucleotides

[0084] (3.1) Determine the forward point mutual information value of the nucleotides of the DNA sequence to be encoded in the positive class data set as follows

[0085]

[0086] Where x is the nucleotide at the jth position, x∈{A,C,G,T}, z is the nucleotide at the j+α+1th position, z∈{A,C,G,T}, is the frequency of occurrence of dinucleotide xz at the jth and j+α+1th positions of all sequence samples in the positive dataset, is the frequency of occurrence of nucleotide x at the jth position of all sequence samples in the positive dataset, is the frequency of occurrence of nucleotide z at the j+α+1th position of all sequence samples in the positive data set.

[0087] The backward point mutual information value of the nucleotides of the DNA sequence to be encoded in the positive data set is determined as follows:

[0088]

[0089] Where y is the nucleotide at position j-α-1, y∈{A,C,G,T}, is the frequency of occurrence of dinucleotide xy at the jth and j-α-1th positions of all sequence samples in the positive data set, is the frequency of occurrence of nucleotide y at the j-α-1th position of all sequence samples in the positive data set.

[0090] The point mutual information encoding value of the nucleotide at the jth position of the DNA sequence sample to be encoded in the positive class data set Defined as the forward point mutual information value and the backward point mutual information value The mean of the DNA sequence samples of length l is encoded as the point mutual information feature vector V with a length of l-2α-2 + :

[0091]

[0092]

[0093] The value of l in this embodiment is 41.

[0094] (3.2) Determine the forward point mutual information value of the nucleotides of the DNA sequence to be encoded in the negative class data set as follows

[0095]

[0096] in, is the frequency of occurrence of dinucleotide xz at the jth and j+α+1th positions of all sequence samples in the negative dataset, is the frequency of occurrence of nucleotide x at the jth position of all sequence samples in the negative dataset, is the frequency of occurrence of nucleotide z at the j+α+1th position of all sequence samples in the negative dataset.

[0097] The backward point mutual information value of the nucleotides of the DNA sequence to be encoded in the negative data set is determined by the following formula:

[0098]

[0099] in, is the frequency of occurrence of dinucleotide xy at the jth and j-α-1th positions of all sequence samples in the negative dataset, is the frequency of occurrence of nucleotide y at the j-α-1th position of all sequence samples in the negative dataset.

[0100] The point mutual information encoding value of the nucleotide at the jth position of the DNA sequence sample to be encoded in the negative class data set Defined as the forward point mutual information value and the backward point mutual information value The mean of the DNA sequence samples of length l is encoded as the point mutual information feature vector V with a length of l-2α-2 - :

[0101]

[0102]

[0103] The value of l in the embodiment is 41.

[0104] (3.3) Given a DNA sequence sample of length l to be encoded, we can use the vector V + and V - Subtract the corresponding elements to determine its eigenvector V:

[0105] V=[V α+2 ,V α+3 ,…,V j ]

[0106]

[0107] (4) Feature combination

[0108] When the parameter α is 0, the eigenvector V(0) is [V2, V3, V4, …, V l-2 ,V l-1 ], the number of elements is l-2, when α is 1, the eigenvector V(1) is [V3,V4,V5,…,V l-3 ,V l-2 ], the number of elements is l-4, ..., α is (l-5) / 2, the eigenvector V((l-5) / 2) is [V (l-1) / 2 ,V (l+1) / 2 ,V (l+3) / 2 ], the number of elements is 3, and when α is (l-3) / 2, the eigenvector V((l-3) / 2) is [V (l+1) / 2 ], the number of elements is 1; the eigenvectors determined by different values ​​of parameter α are combined into an element number of (l-1) 2 / 4 high-dimensional feature vector [V(0), V(1),…, V((l-5) / 2), V((l-3) / 2)], where l in this embodiment is 41.

[0109] (5) DNA sequence encoding

[0110] Using the above steps (1) to (4), the DNA sequence data set D is encoded into a numerical data set D'. s is the number of samples in the numerical data set D'. The value of s is a finite positive integer. The value of s in this embodiment is 3108, (l-1) 2 / 4 is the feature number of the numerical data set D', which completes the DNA sequence encoding.

[0111] In order to verify the beneficial effects of the present invention, the inventors used the DNA sequence encoding method of Example 1 and the PSNP (position-specific nucleotide propensities), PSDP (position-specific dinucleotide propensities), KNF (Knucleotide frequencies), KSNPF (Kspaced nucleotide pair frequencies), NPPS (nucleotide pair position specificity), PBE (positional binary encoding), NCPNC (nucleotide chemical property and nucleotide composition) encoding methods to apply to the N sequence of Caenorhabditis elegans DNA. 4 -methylcytosine site recognition prediction, compare the performance of the support vector machine model established by each encoding method. The average classification accuracy (Accuracy), sensitivity (Sensitivity), specificity (Specificity), MCC (Mathew's correlation coefficient), AUROC (Area under the receiver operating characteristic curve), AUPRC (Area under the precision recall curve) of the 10-fold cross validation method were used for evaluation. The experimental method is as follows:

[0112] 1. The N of DNA of Caenorhabditis elegans was quantified according to the method of Example 1. 4 -Methylcytosine dataset sequence samples are encoded.

[0113] 2. Dataset Standardization

[0114] The numerical data set D' is standardized using the following maximum minimization method:

[0115]

[0116] Among them, g m,n is the nth eigenvalue of the mth sample in the numerical data set D', g m,n The normalized value is g' m,n ,max(g n ) and min(g n ) represents the maximum and minimum eigenvalues ​​on the nth column of the numerical data set D', 1≤m≤s, 1≤n≤(l-1)2 / 4, the values ​​of m and n are finite positive integers. In this embodiment, the value of l is 41, and the value of s is 3108.

[0117] 3. Divide the data set

[0118] The standardized numerical data set D' is divided into 10 parts using the K-fold cross-validation method (K = 10), and one of them is used as the test set D' in turn. Te , the remaining 9 are used as training set D' Tr , a total of 10 runs, each time with the training set D' Tr With the test set D' Te The ratio is 9:1.

[0119] 4. Model training and testing

[0120] Using the training set D' Tr Train the support vector machine model and use the test set D' Te Test the performance of the support vector machine model.

[0121] The same operation was performed on the seven comparative experiments according to steps 2 to 4 in the experimental method to determine the N 4 -methylcytosine sites were used for identification and prediction. The experimental results of average classification accuracy, sensitivity, specificity, and MCC are shown in Table 1, and the experimental results of AUROC are shown in Figure 2 , the experimental results of AUPRC can be found in Figure 3 .

[0122] Table 1 Comparative experimental results of the method in Example 1 and seven methods

[0123]

[0124] As can be seen from Table 1, the support vector machine model established based on the DNA sequence encoding method of the present invention can effectively identify the N 4 The accuracy, sensitivity, specificity, and MCC of -methylcytosine site recognition prediction were 0.905, 0.902, 0.909, and 0.811, respectively, which were all higher than those of the other seven comparative coding methods.

[0125] Depend on Figure 2 It can be seen that the support vector machine model established based on the DNA sequence encoding method of the present invention can effectively identify the N 4 The AUROC value of -methylcytosine site recognition prediction was 0.967, which was higher than that of the other seven contrast encoding methods.

[0126] Depend on Figure 3It can be seen that the support vector machine model established based on the DNA sequence encoding method of the present invention can effectively identify the N 4 The AUPRC value of -methylcytosine site recognition prediction was 0.966, which was higher than that of the other seven comparison coding methods.

[0127] Example 2

[0128] Benchmark data for identifying N 6 N-methyladenosine sites in the Saccharomyces cerevisiae genome of yeast RNA 6 -methyladenosine (N 6 Methyladenosine, m 6 A) dataset as an example, the dataset has a total of 2614 RNA sequence samples, among which the number of samples in the positive dataset, that is, the real N 6 -The number of methyladenosine samples is 1307, and the number of negative class data set samples is non-N 6 The number of -methyladenosine samples is 1307, and the length l of each sequence sample is 51. The bidirectional dinucleotide position-specific preference and point mutual information RNA sequence encoding method of this embodiment consists of the following steps (see Figure 1 ).

[0129] (1) Construction of RNA sequence nucleotide position-specific preference matrix

[0130] Given an RNA sequence dataset D, the dataset consists of a positive class dataset and a negative class dataset.

[0131] The nucleotide position-specific preference matrix of the positive class dataset is determined as follows:

[0132]

[0133] Wherein, A, C, G, and U are four nucleotides of RNA, i is the position of the nucleotide, 1≤i≤l, the value of i is a finite positive integer, l is the nucleotide length of the RNA sequence sample, the value of l is an odd number, and the value of l in this embodiment is 51. are the occurrence frequencies of nucleotides A, C, G, and U at the i-th position of all sequence samples in the positive data set, respectively.

[0134] The nucleotide position-specific preference matrix of the negative class dataset is determined as follows:

[0135]

[0136] in, are the occurrence frequencies of nucleotides A, C, G, and U at the i-th position of all sequence samples in the negative data set, respectively.

[0137] (2) Construction of RNA sequence bidirectional dinucleotide position-specific preference matrix

[0138] The forward dinucleotide position-specific preference matrix of the positive data set is determined as follows:

[0139]

[0140] Among them, AA, AC, ..., UU are 16 dinucleotides composed of the four nucleotides A, C, G, and U of RNA, α represents the distance between two nucleotides, 0≤α≤(l-3) / 2, the value of α is a finite positive integer, 0≤α≤24 in this embodiment, j is the position of the nucleotide, α+2≤j≤l-α-1, α+2≤j≤50-α in this embodiment, the value of j is a finite positive integer, are the occurrence frequencies of dinucleotides AA, AC, …, UU at the j-th and j+α+1-th positions of all sequence samples in the positive data set, respectively.

[0141] The backward dinucleotide position-specific preference matrix of the positive class dataset is determined as follows:

[0142]

[0143] in, are the occurrence frequencies of dinucleotides AA, AC, …, UU at the j-th and j-α-1-th positions of all sequence samples in the positive class dataset, respectively.

[0144] The forward dinucleotide position-specific preference matrix of the negative class data set is determined as follows:

[0145]

[0146] in, are the occurrence frequencies of dinucleotides AA, AC, …, UU at the j-th and j+α+1-th positions of all sequence samples in the negative data set, respectively.

[0147] The backward dinucleotide position-specific preference matrix for the negative class dataset is determined as follows:

[0148]

[0149] in, are the occurrence frequencies of dinucleotides AA, AC, …, UU at the j-th and j-α-1-th positions of all sequence samples in the negative data set, respectively.

[0150] (3) Determine the point mutual information value of RNA sequence nucleotides

[0151] (3.1) Determine the forward point mutual information value of the nucleotides of the RNA sequence to be encoded in the positive class data set as follows

[0152]

[0153] Where x is the nucleotide at the jth position, x∈{A,C,G,U}, z is the nucleotide at the j+α+1th position, z∈{A,C,G,U}, is the frequency of occurrence of dinucleotide xz at the jth and j+α+1th positions of all sequence samples in the positive dataset, is the frequency of occurrence of nucleotide x at the jth position of all sequence samples in the positive dataset, is the frequency of nucleotide z at the j+α+1th position of all sequence samples in the positive data set.

[0154] The backward point mutual information value of the nucleotide of the RNA sequence to be encoded in the positive data set is determined as follows:

[0155]

[0156] Where y is the nucleotide at position j-α-1, y∈{A,C,G,U}, is the frequency of occurrence of dinucleotide xy at the jth and j-α-1th positions of all sequence samples in the positive data set, is the frequency of occurrence of nucleotide y at the j-α-1th position of all sequence samples in the positive data set.

[0157] The point mutual information encoding value of the nucleotide at the jth position of the RNA sequence sample to be encoded in the positive class data set Defined as the forward point mutual information value and the backward point mutual information value The mean of the RNA sequence samples of length l is encoded as the point mutual information feature vector V with a length of l-2α-2 + :

[0158]

[0159]

[0160] The value of l in this embodiment is 51.

[0161] (3.2) Determine the forward point mutual information value of the nucleotides of the RNA sequence to be encoded in the negative class data set as follows

[0162]

[0163] in, is the frequency of occurrence of dinucleotide xz at the jth and j+α+1th positions of all sequence samples in the negative dataset, is the frequency of occurrence of nucleotide x at the jth position of all sequence samples in the negative dataset, is the frequency of occurrence of nucleotide z at the j+α+1th position of all sequence samples in the negative dataset.

[0164] The backward point mutual information value of the nucleotides of the RNA sequence to be encoded in the negative data set is determined as follows:

[0165]

[0166] in, is the frequency of occurrence of dinucleotide xy at the jth and j-α-1th positions of all sequence samples in the negative dataset, is the frequency of occurrence of nucleotide y at the j-α-1th position of all sequence samples in the negative dataset.

[0167] Point mutual information encoding value of the nucleotide at the jth position of the RNA sequence sample to be encoded in the negative class data set Defined as the forward point mutual information value and the backward point mutual information value The mean of the RNA sequence samples of length l is encoded as the point mutual information feature vector V with a length of l-2α-2 - :

[0168]

[0169]

[0170] The value of l in this embodiment is 51.

[0171] (3.3) Given a sample of RNA sequence to be encoded with length l, we can use the vector V + and V - Subtract the corresponding elements to determine its eigenvector V:

[0172] V=[V α+2 ,V α+3 ,…,V j ]

[0173]

[0174] (4) Feature combination

[0175] When the parameter α is 0, the eigenvector V(0) is [V2, V3, V4, …, V l-2 ,V l-1 ], the number of elements is l-2, when α is 1, the eigenvector V(1) is [V3,V4,V5,…,V l-3 ,V l-2 ], the number of elements is l-4, ..., α is (l-5) / 2, the eigenvector V((l-5) / 2) is [V (l-1) / 2 ,V (l+1) / 2 ,V (l+3) / 2 ], the number of elements is 3, and when α is (l-3) / 2, the eigenvector V((l-3) / 2) is [V (l+1) / 2 ], the number of elements is 1; the eigenvectors determined by different values ​​of parameter α are combined into an element number of (l-1) 2 / 4 high-dimensional feature vector [V(0), V(1),…, V((l-5) / 2), V((l-3) / 2)], where l in this embodiment is 51.

[0176] (5) RNA sequence encoding

[0177] Using the above steps (1) to (4), the RNA sequence dataset D is encoded into a numerical dataset D'. s is the number of samples in the numerical data set D'. The value of s is a finite positive integer. The value of s in this embodiment is 2614, (l-1) 2 / 4 is the feature number of the numerical data set D', which completes the RNA sequence encoding.

[0178] In order to verify the beneficial effects of the present invention, the inventors used the RNA sequence encoding method of Example 2 and the PSNP (position-specific nucleotide propensities), PSDP (position-specific dinucleotide propensities), KNF (Knucleotide frequencies), KSNPF (Kspaced nucleotide pair frequencies), NPPS (nucleotide pair position specificity), PBE (positional binary encoding), NCPNC (nucleotide chemical property and nucleotide composition) encoding methods to apply to the N6 -methyladenosine site recognition prediction, compare the performance of the support vector machine model established by each encoding method. The average classification accuracy (Accuracy), sensitivity (Sensitivity), specificity (Specificity), MCC (Mathew's correlation coefficient), AUROC (Area under the receiver operating characteristic curve), AUPRC (Area under the precision recall curve) of the 10-fold cross validation method were used for evaluation. The experimental method is as follows:

[0179] 1. According to the method of Example 2, the N 6 -Methyladenosine dataset sequence samples are encoded.

[0180] 2. Dataset Standardization

[0181] The numerical data set D' is standardized using the following maximum minimization method:

[0182]

[0183] Among them, g m,n is the nth eigenvalue of the mth sample in the numerical data set D', g m,n The normalized value is g' m,n ,max(g n ) and min(g n ) represents the maximum and minimum eigenvalues ​​on the nth column of the numerical data set D', 1≤m≤s, 1≤n≤(l-1) 2 / 4, the values ​​of m and n are finite positive integers. In this embodiment, the value of l is 51, and the value of s is 2614.

[0184] 3. Divide the data set

[0185] The standardized numerical data set D' is divided into 10 parts using the K-fold cross-validation method (K = 10), and one of them is used as the test set D' in turn. Te , the remaining 9 are used as training set D' Tr , a total of 10 runs, each time with the training set D' Tr With the test set D' Te The ratio is 9:1.

[0186] 4. Model training and testing

[0187] Using the training set D' Tr Train the support vector machine model and use the test set D'Te Test the performance of the support vector machine model.

[0188] The same operation was performed on the seven comparative experiments according to steps 2 to 4 in the experimental method to determine the N 6 -methyladenosine sites were used for identification and prediction. The experimental results of average classification accuracy, sensitivity, specificity and MCC are shown in Table 2.

[0189] The experimental results of AUROC are shown in Figure 4 , the experimental results of AUPRC can be found in Figure 5 .

[0190] Table 2 Comparative experimental results of the method in Example 2 and 7 methods

[0191]

[0192] As can be seen from Table 2, the support vector machine model established based on the RNA sequence encoding method of the present invention can effectively predict the N value of yeast RNA. 6 The accuracy, sensitivity, specificity and MCC of -methyladenosine site recognition prediction were 0.905, 0.916, 0.894 and 0.810, respectively, which were higher than those of the other seven comparative coding methods.

[0193] Depend on Figure 4 It can be seen that the support vector machine model established based on the RNA sequence encoding method of the present invention can effectively predict the N value of yeast RNA. 6 The AUROC value of -methyladenosine site identification prediction was 0.968, which was higher than that of the other 7 contrast encoding methods.

[0194] Depend on Figure 5 It can be seen that the support vector machine model established based on the RNA sequence encoding method of the present invention can effectively predict the N value of yeast RNA. 6 The AUPRC value of the -methyladenosine site identification prediction was 0.967, which was higher than that of the other seven comparison coding methods.

Claims

1. A method for encoding bidirectional dinucleotide position-specific preference and point mutual information DNA / RNA sequences, characterized in that It consists of the following steps: (1) Construction of DNA / RNA sequence nucleotide position-specific preference matrix Given a DNA / RNA sequence dataset D, which consists of a positive dataset and a negative dataset; The nucleotide position-specific preference matrix of the positive class dataset is determined as follows: A, C, G, and X are four nucleotides of DNA / RNA. X is represented by nucleotide T in the DNA sequence dataset and by nucleotide U in the RNA sequence dataset. i is the position of the nucleotide, 1≤i≤l, and the value of i is a finite positive integer. l is the nucleotide length of the DNA / RNA sequence sample, and the value of l is an odd number. are the occurrence frequencies of nucleotides A, C, G, and X at the i-th position of all sequence samples in the positive class data set; The nucleotide position-specific preference matrix of the negative class dataset is determined as follows: in, are the occurrence frequencies of nucleotides A, C, G, and X at the i-th position of all sequence samples in the negative dataset; (2) Construction of a bidirectional dinucleotide position-specific preference matrix for DNA / RNA sequences The forward dinucleotide position-specific preference matrix of the positive data set is determined as follows: Among them, AA, AC, ..., XX are 16 dinucleotides composed of the four nucleotides A, C, G, and X of DNA / RNA, α represents the distance between two nucleotides, 0≤α≤(l-3) / 2, j is the position of the nucleotide, α+2≤j≤l-α-1, and the value of j is a finite positive integer. are the occurrence frequencies of dinucleotides AA, AC, …, XX at the jth and j+α+1th positions of all sequence samples in the positive dataset, respectively; The backward dinucleotide position-specific preference matrix of the positive class dataset is determined as follows: in, are the occurrence frequencies of dinucleotides AA, AC, …, XX at the j-th and j-α-1-th positions of all sequence samples in the positive class dataset, respectively; The forward dinucleotide position-specific preference matrix of the negative class data set is determined as follows: in, are the occurrence frequencies of dinucleotides AA, AC, …, XX at the jth and j+α+1th positions of all sequence samples in the negative dataset, respectively; The backward dinucleotide position-specific preference matrix for the negative class dataset is determined as follows: in, are the occurrence frequencies of dinucleotides AA, AC, …, XX at the j-th and j-α-1-th positions of all sequence samples in the negative dataset, respectively; (3) Determine the point mutual information value of DNA / RNA sequence nucleotides (3.1) Determine the forward point mutual information value of the nucleotides of the DNA / RNA sequence to be encoded in the positive class data set as follows Where x is the nucleotide at the jth position, x∈{A,C,G,X}, z is the nucleotide at the j+α+1th position, z∈{A,C,G,X}, is the frequency of occurrence of dinucleotide xz at the jth and j+α+1th positions of all sequence samples in the positive dataset, is the frequency of occurrence of nucleotide x at the jth position of all sequence samples in the positive dataset, is the frequency of occurrence of nucleotide z at the j+α+1th position of all sequence samples in the positive data set; Determine the backward point mutual information value of the nucleotides of the DNA / RNA sequence to be encoded in the positive data set as follows: Where y is the nucleotide at position j-α-1, y∈{A,C,G,X}, is the frequency of occurrence of dinucleotide xy at the jth and j-α-1th positions of all sequence samples in the positive data set, is the frequency of occurrence of nucleotide y at the j-α-1th position of all sequence samples in the positive data set; Point mutual information encoding value of the nucleotide at the jth position of the DNA / RNA sequence sample to be encoded in the positive class data set Defined as the forward point mutual information value and the backward point mutual information value The mean of the DNA / RNA sequence samples of length l is encoded as the point mutual information feature vector V with a length of l-2α-2 + : (3.2) Determine the forward point mutual information value of the nucleotides of the DNA / RNA sequence to be encoded in the negative class data set as follows in, is the frequency of occurrence of dinucleotide xz at the jth and j+α+1th positions of all sequence samples in the negative dataset, is the frequency of occurrence of nucleotide x at the jth position of all sequence samples in the negative dataset, is the occurrence frequency of nucleotide z at the j+α+1th position of all sequence samples in the negative dataset; The backward point mutual information value of the nucleotides of the DNA / RNA sequence to be encoded in the negative data set is determined as follows: in, is the frequency of occurrence of dinucleotide xy at the jth and j-α-1th positions of all sequence samples in the negative dataset, is the occurrence frequency of nucleotide y at the j-α-1th position of all sequence samples in the negative dataset; Point mutual information encoding value of the nucleotide at the jth position of the DNA / RNA sequence sample to be encoded in the negative class data set Defined as the forward point mutual information value and the backward point mutual information value The mean of the DNA / RNA sequence samples of length l is encoded as the point mutual information feature vector V with a length of l-2α-2 - : (3.3) Given a DNA / RNA sequence sample of length l to be encoded, we can use the vector V + and V - Subtract the corresponding elements to determine its eigenvector V: V=[V α+2 ,V α+3 ,…,V j ] (4) Feature combination When the parameter α is 0, the eigenvector V(0) is [V2, V3, V4, …, V l-2 ,V l-1 ], the number of elements is l-2, when α is 1, the eigenvector V(1) is [V3,V4,V5,…,V l-3 ,V l-2 ], the number of elements is l-4, ..., α is (l-5) / 2, the eigenvector V((l-5) / 2) is [V (l-1) / 2 ,V (l+1) / 2 ,V (l+3) / 2 ], the number of elements is 3, and when α is (l-3) / 2, the eigenvector V((l-3) / 2) is [V (l+1) / 2 ], the number of elements is 1; the eigenvectors determined by different values ​​of parameter α are combined into an element number of (l-1) 2 / 4 high-dimensional feature vector [V(0), V(1), …, V((l-5) / 2), V((l-3) / 2)]; (5) DNA / RNA sequence encoding Using the above steps (1) to (4), the DNA / RNA sequence data set D is encoded into a numerical data set D'. s is the number of samples in the numerical data set D'. The value of s is a finite positive integer, (l-1) 2 / 4 is the number of features of the numerical data set D'.