ncRNA recognition method, system and device based on support vector machine with low-precision RBF kernel

By using low-precision RBF kernel function and Taylor unfolding approximation mode in ncRNA recognition, the problem of low training and prediction efficiency of existing models is solved, and the computational efficiency is improved and the recognition accuracy is maintained.

CN114821577BActive Publication Date: 2024-11-15WUHAN UNIV
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202210264629.1
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2022-03-17
Publication Date
2024-11-15
Estimated Expiration
2042-03-17

AI Technical Summary

Technical Problem

The existing support vector machine model based on RBF core is inefficient in training and prediction in ncRNA recognition, which leads to a long calculation time and affects the promotion of technology.

Method used

The low-precision RBF kernel function is used to scale the exponential function through Taylor expansion approximation mode (first order, second order, third order, fourth order, fifth order) to reduce calculation errors and accelerate calculations, and improve the training and prediction efficiency of the model.

Benefits of technology

It significantly reduces the computational time-consuming of model training and prediction, improves the computational efficiency of ncRNA recognition, and maintains high recognition accuracy within the acceptable accuracy range.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN114821577B_ABST
    Figure CN114821577B_ABST
Patent Text Reader

Abstract

The present invention discloses a method, system and device for identifying ncRNA based on a support vector machine with a low-precision RBF kernel, including three steps: data processing, model training and model prediction. In the first step, RNA data is processed into a standard format required by the model. In the second step, a precision mode is selected and RNA data is input into the model for training. In the third step, the trained model predicts the RNA data to be identified. Experimental results show that the above method improves the computational efficiency of training and prediction of the support vector machine model based on the RBF kernel while more accurately identifying ncRNA.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention belongs to the technical field of non-coding RNA recognition, and relates to a ncRNA recognition method, system and device, and in particular to a ncRNA recognition method, system and device based on a support vector machine model with a low-precision RBF kernel. Background Art

[0002] ncRNA (non-coding RNA) is an RNA molecule that has functions but is not translated into proteins. ncRNAs play an important role in cell life activities and disease pathogenesis. With the application of high-throughput technology, the number of ncRNAs is growing rapidly. In order to better understand biology and discover candidate drug targets, accurate and rapid identification of ncRNAs has important practical significance and application value.

[0003] Support vector machine is a classification method with a solid theoretical foundation. It uses inner product kernel function to replace high-dimensional nonlinear mapping. The most commonly used kernel function is RBF kernel. RBF kernel has good performance in the identification of ncRNA, but the exponential function operation cost in RBF kernel is often too high, and the training and prediction process of support vector machine model based on RBF kernel takes up a large proportion of time, resulting in low efficiency of model training and prediction, which is not conducive to the promotion of technology. Summary of the invention

[0004] In view of the above problems, the present invention provides a ncRNA recognition method, system and device based on a support vector machine model with a low-precision RBF kernel, which are used to quickly and accurately identify ncRNA and promote the popularization and use of the technology.

[0005] The technical solution adopted by the method of the present invention is: a ncRNA recognition method based on a support vector machine with a low-precision RBF kernel, comprising the following steps:

[0006] Step 1: Preprocess the RNA data to be identified;

[0007] Step 2: Extract data features from the preprocessed RNA data;

[0008] Step 3: Input the extracted data feature information into the support vector machine with low-precision RBF kernel for ncRNA identification;

[0009] The low-precision RBF kernel is k(x1,x2)=exp(-γ(||x1-x2||) 2 ), where γ is the kernel function coefficient, and x1 and x2 are vectors composed of the features of any two RNAs.

[0010] The technical solution adopted by the system of the present invention is: a ncRNA recognition system based on a support vector machine with a low-precision RBF kernel, comprising the following modules:

[0011] Module 1 is used to pre-process the RNA data to be identified;

[0012] Module 2 is used to extract data features from the preprocessed RNA data;

[0013] Module 3 is used to input the extracted data feature information into a support vector machine with a low-precision RBF kernel for ncRNA identification;

[0014] The low-precision RBF kernel is k(x1,x2)=exp(-γ(||x1-x2||) 2 ), where γ is the kernel function coefficient, and x1 and x2 are vectors composed of the features of any two RNAs.

[0015] The technical solution adopted by the device of the present invention is: an ncRNA recognition device based on a support vector machine with a low-precision RBF kernel, comprising:

[0016] one or more processors;

[0017] The storage device is used to store one or more programs. When the one or more programs are executed by the one or more processors, the one or more processors implement the ncRNA recognition method of the support vector machine with low-precision RBF kernel.

[0018] The present invention can more accurately identify ncRNA by using eight data features of RNA and a support vector machine model, and improves the computational efficiency of model training and prediction by using a low-precision RBF kernel function. At the same time, it reduces the computational time consumed by the support vector machine model based on the RBF kernel in the training and prediction of ncRNA. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] Figure 1 The figure is a flow chart of a method according to an embodiment of the present invention. DETAILED DESCRIPTION

[0020] In order to facilitate ordinary technicians in the field to understand and implement the present invention, the present invention is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the implementation examples described herein are only used to illustrate and explain the present invention, and are not used to limit the present invention.

[0021] Please see Figure 1 The present invention provides a method for identifying ncRNA based on a support vector machine with a low-precision RBF kernel, comprising the following steps:

[0022] Step 1: Preprocess the RNA data to be identified and process each RNA data into fasta format data.

[0023] Step 2: Extract data features from the preprocessed RNA data;

[0024] The data features of this embodiment include eight features. The first feature is the ΔG° of the double-stranded RNA calculated by the Dynalign algorithm. total =ΔG°1+ΔG°2+number of gaps×ΔG° gp ; ΔG°1 and ΔG°2 are the predicted folding free energy changes of the secondary structures of the two sequences, respectively, and ΔG° gp is the penalty imposed on gaps; the second feature is the length of the shorter of the two sequences; the third feature is the frequency of base 'A' in one sequence; the fourth feature is the frequency of base 'U' in one sequence; the fifth feature is the frequency of base 'C' in one sequence; the sixth feature is the frequency of base 'A' in the other sequence; the seventh feature is the frequency of base 'U' in the other sequence; the eighth feature is the frequency of base 'C' in the other sequence.

[0025] Step 3: Input the extracted data feature information into the support vector machine with low-precision RBF kernel for ncRNA identification;

[0026] Among them, the low-precision RBF kernel is k(x1,x2)=exp(-γ(||x1-x2||) 2 ), where γ is the kernel function coefficient, and x1 and x2 are vectors composed of the features of any two RNAs.

[0027] The support vector machine with a low-precision RBF kernel used in this embodiment is a trained support vector machine with a low-precision RBF kernel;

[0028] In this embodiment, the support vector machine training process of the low-precision RBF kernel includes the following sub-steps:

[0029] (1) Construct a training set, process the RNA data in the training set into the fasta standard format required by the support vector machine, and extract eight features and labels of the RNA data in the training set;

[0030] The first of the eight features in this example is the ΔG° of the double-stranded RNA calculated by the Dynalign algorithm. total =ΔG°1+ΔG°2+number of gaps×ΔG° gp ; ΔG°1 and ΔG°2 are the predicted folding free energy changes of the secondary structures of the two sequences, respectively, and ΔG° gpis the penalty imposed on gaps; the second feature is the length of the shorter of the two sequences; the third feature is the frequency of base ‘A’ in one sequence; the fourth feature is the frequency of base ‘U’ in one sequence; the fifth feature is the frequency of base ‘C’ in one sequence; the sixth feature is the frequency of base ‘A’ in the other sequence; the seventh feature is the frequency of base ‘U’ in the other sequence; the eighth feature is the frequency of base ‘C’ in the other sequence;

[0031] In this embodiment, the label of ncRNA data is 1, and the label of non-ncRNA data is -1;

[0032] (2) Set the stopping condition tol and input the training set data into the support vector machine with low-precision RBF kernel for training;

[0033] In this embodiment, the support vector machine of the low-precision RBF kernel includes support vector machines of the RBF kernel of five approximation modes, namely, a first-order approximation mode, a second-order approximation mode, a third-order approximation mode, a fourth-order approximation mode and a fifth-order approximation mode;

[0034] The first-order approximation mode uses the first-order Taylor expansion: exp(k) = 1 + k; k represents the RBF kernel function exp(-γ(||x1-x2||) 2 ) Zoom 2 r Then the variable k∈(0,1) satisfies Equation 2 r ×exp(k)=exp(-γ(||x1-x2||) 2 ), γ is the RBF kernel function coefficient, x1 and x2 are vectors composed of the features of any two RNAs, and r is the exponential position; the complete Taylor expansion has an error term O(k n ), if k>1, the error term will obviously become larger, and the difference from the true value will be larger. Scaling the exp function so that k∈(0,1) will reduce the error and make it closer to the true value. The scaling ratio is 2 r It is convenient to utilize displacement operations in the program to speed up calculations.

[0035] The second-order approximation model uses the second-order Taylor expansion: exp(k) = 1+k+k 2 / 2! ;

[0036] The third-order approximation model uses the third-order Taylor expansion: exp(k) = 1+k+k 2 / 2+k 3 / 3! ;

[0037] The fourth-order approximation model uses the fourth-order Taylor expansion: exp(k) = 1+k+k 2 / 2+k 3 / 3! +k 4 / 4! ;

[0038] The fifth-order approximation model uses the fifth-order Taylor expansion: exp(k) = 1 + k + k 2 / 2+k 3 / 3! +k 4 / 4! +k 5 / 5! ;

[0039] (3) When the calculation error is less than the stopping condition tol, the calculation is terminated and a support vector machine with a trained low-precision RBF kernel is obtained.

[0040] In order to better compare the training and prediction effects of the support vector machine model with a low-precision RBF kernel, the exp function in the math.h library function of C++ is used to calculate the RBF kernel function as a high-precision control group. The training set is input into the support vector machine model with a high-precision RBF kernel for training, and the test set is input into the trained model.

[0041] For the five approximate modes and the high-precision control group, recorded as examples one to six, the training time, test time, training set accuracy and test set accuracy are recorded respectively, and then the training time ratio, training set prediction time ratio and test set prediction time ratio are calculated. The experimental results show that the training time, training set training time and test set prediction time of examples one to five are faster than those of example six. Although the training set accuracy and test set accuracy of examples one to three are lower than those of example six, the maximum training set accuracy difference and the maximum test set accuracy difference are 2.59% and 3.45% respectively. Within the acceptable range, the summary results are shown in Table 1:

[0042] Table 1

[0043]

[0044]

[0045] The training set accuracy is the number of correctly predicted data in the training set divided by the number of training sets, the test set accuracy is the number of correctly predicted data in the test set divided by the number of predicted data in the prediction set, the training time ratio of an instance is the training time of instance six divided by the training time of the instance, the training set prediction time ratio of an instance is the training set prediction time of instance six divided by the training set prediction time of the instance, and the test set prediction time ratio of an instance is the test set prediction time of instance six divided by the test set prediction time of the instance.

[0046] For the test cases of Example 1 to Example 6, the sensitivity, specificity and F-value are calculated respectively, which can also prove the effectiveness of the present invention in terms of accuracy. The results are shown in Table 2:

[0047] Table 2

[0048] Instance number Sensitivity Specificity F-value Example 1 83.88% 93.17% 84.93% Example 2 86.82% 94.75% 88.00% Example 3 88.04% 94.99% 88.90% Example 4 89.54% 95.53% 90.22% Example 5 89.54% 95.53% 90.22% Example 6 89.54% 95.53% 90.22%

[0049] It should be understood that the above description of the preferred embodiment is relatively detailed and cannot be regarded as limiting the scope of patent protection of the present invention. Under the enlightenment of the present invention, ordinary technicians in this field can also make substitutions or modifications without departing from the scope of protection of the claims of the present invention, which all fall within the scope of protection of the present invention. The scope of protection requested for the present invention shall be based on the attached claims.

Claims

1. A method for identifying ncRNA based on a support vector machine with a low-precision RBF kernel, characterized in that: The following steps are involved: Step 1: Preprocess the RNA data to be identified; Step 2: Extract data features from the preprocessed RNA data; Step 3: Input the extracted data feature information into the support vector machine with low-precision RBF kernel for ncRNA identification; The low-precision RBF kernel is k(x1,x2)=exp(-γ(||x1-x2||) 2 ), where γ is the kernel function coefficient, x1 and x2 are vectors composed of the features of any two RNAs; The support vector machine of the low-precision RBF kernel is a trained support vector machine of the low-precision RBF kernel; The support vector machine training process of the low-precision RBF kernel includes the following sub-steps: (1) Construct a training set and extract eight features and labels of RNA data in the training set; The first of the eight features is the ΔG° of the double-stranded RNA calculated by the Dynalign algorithm. total =ΔG°1+ΔG°2+number of gaps×ΔG° gp ; ΔG°1 and ΔG°2 are the predicted folding free energy changes of the secondary structures of the two sequences, respectively, and ΔG° gp is the penalty imposed on gaps; the second feature is the length of the shorter of the two sequences; the third feature is the frequency of base ‘A’ in one sequence; the fourth feature is the frequency of base ‘U’ in one sequence; the fifth feature is the frequency of base ‘C’ in one sequence; the sixth feature is the frequency of base ‘A’ in the other sequence; the seventh feature is the frequency of base ‘U’ in the other sequence; the eighth feature is the frequency of base ‘C’ in the other sequence; The label, the label of ncRNA data is 1, and the label of non-ncRNA data is -1; (2) Set the stopping condition tol and input the training set data into the support vector machine with low-precision RBF kernel for training; The support vector machine of the low-precision RBF kernel includes support vector machines of the RBF kernel with five approximate modes, namely, a first-order approximate mode, a second-order approximate mode, a third-order approximate mode, a fourth-order approximate mode and a fifth-order approximate mode; The first-order approximation mode uses the first-order Taylor expansion: exp(k) = 1 + k; where k represents the RBF kernel function exp(-γ(||x1-x2||) 2 ) Zoom 2 r Then the variable k∈(0,1) satisfies Equation 2 r ×exp(k)=exp(-γ(||x1-x2||) 2 ), γ is the RBF kernel function coefficient, x1 and x2 are vectors composed of the features of any two RNAs, and r is the exponential position; The second-order approximation model uses the second-order Taylor expansion: exp(k) = 1+k+k 2 / 2! ; The third-order approximation model uses the third-order Taylor expansion: exp(k) = 1+k+k 2 / 2+k 3 / 3! ; The fourth-order approximation model uses the fourth-order Taylor expansion: exp(k) = 1+k+k 2 / 2+k 3 / 3! +k 4 / 4! ; The fifth-order approximation model uses the fifth-order Taylor expansion: exp(k) = 1 + k + k 2 / 2+k 3 / 3! +k 4 / 4! +k 5 / 5! ; (3) When the calculation error is less than the stopping condition tol, the calculation is terminated and a support vector machine with a trained low-precision RBF kernel is obtained.

2. The ncRNA recognition method based on the support vector machine of the low-precision RBF kernel according to claim 1, characterized in that: The preprocessing described in step 1 is to process each RNA data into fasta format data.

3. The ncRNA recognition method based on the support vector machine of the low-precision RBF kernel according to claim 1, characterized in that: The data features described in step 2 include eight features. The first feature is the ΔG° of the double-stranded RNA calculated by the Dynalign algorithm. total =ΔG°1+ΔG°2+number of gaps×ΔG° gp ; where ΔG°1 and ΔG°2 are the predicted folding free energy changes of the secondary structures of the two sequences of the double-stranded RNA, respectively, and ΔG° gp It is a penalty imposed on the gap; The second feature is the length of the shorter of the two sequences of the double-stranded RNA; the third feature is the frequency of base 'A' in one sequence; the fourth feature is the frequency of base 'U' in one sequence; the fifth feature is the frequency of base 'C' in one sequence; the sixth feature is the frequency of base 'A' in the other sequence; the seventh feature is the frequency of base 'U' in the other sequence; and the eighth feature is the frequency of base 'C' in the other sequence.

4. A ncRNA recognition system based on a support vector machine with a low-precision RBF kernel, characterized in that: Includes the following modules: Module 1 is used to pre-process the RNA data to be identified; Module 2 is used to extract data features from the preprocessed RNA data; Module 3 is used to input the extracted data feature information into a support vector machine with a low-precision RBF kernel for ncRNA identification; The low-precision RBF kernel is k(x1,x2)=exp(-γ(||x1-x2||) 2 ), where γ is the kernel function coefficient, x1 and x2 are vectors composed of the features of any two RNAs; The support vector machine of the low-precision RBF kernel is a trained support vector machine of the low-precision RBF kernel; The support vector machine training process of the low-precision RBF kernel includes the following sub-steps: (1) Construct a training set and extract eight features and labels of RNA data in the training set; The first of the eight features is the ΔG° of the double-stranded RNA calculated by the Dynalign algorithm. total =ΔG°1+ΔG°2+number of gaps×ΔG° gp ; ΔG°1 and ΔG°2 are the predicted folding free energy changes of the secondary structures of the two sequences, respectively, and ΔG° gp is the penalty imposed on gaps; the second feature is the length of the shorter of the two sequences; the third feature is the frequency of base ‘A’ in one sequence; the fourth feature is the frequency of base ‘U’ in one sequence; the fifth feature is the frequency of base ‘C’ in one sequence; the sixth feature is the frequency of base ‘A’ in the other sequence; the seventh feature is the frequency of base ‘U’ in the other sequence; the eighth feature is the frequency of base ‘C’ in the other sequence; The label, the label of ncRNA data is 1, and the label of non-ncRNA data is -1; (2) Set the stopping condition tol and input the training set data into the support vector machine with low-precision RBF kernel for training; The support vector machine of the low-precision RBF kernel includes support vector machines of the RBF kernel with five approximate modes, namely, a first-order approximate mode, a second-order approximate mode, a third-order approximate mode, a fourth-order approximate mode and a fifth-order approximate mode; The first-order approximation mode uses the first-order Taylor expansion: exp(k) = 1 + k; where k represents the RBF kernel function exp(-γ(||x1-x2||) 2 ) Zoom 2 r Then the variable k∈(0,1) satisfies Equation 2 r ×exp(k)=exp(-γ(||x1-x2||) 2 ), γ is the RBF kernel function coefficient, x1 and x2 are vectors composed of the features of any two RNAs, and r is the exponential position; The second-order approximation model uses the second-order Taylor expansion: exp(k) = 1+k+k 2 / 2! ; The third-order approximation model uses the third-order Taylor expansion: exp(k) = 1+k+k 2 / 2+k 3 / 3! ; The fourth-order approximation model uses the fourth-order Taylor expansion: exp(k) = 1+k+k 2 / 2+k 3 / 3! +k 4 / 4! ; The fifth-order approximation model uses the fifth-order Taylor expansion: exp(k) = 1 + k + k 2 / 2+k 3 / 3! +k 4 / 4! +k 5 / 5! ; (3) When the calculation error is less than the stopping condition tol, the calculation is terminated and a support vector machine with a trained low-precision RBF kernel is obtained.

5. An ncRNA recognition device based on a support vector machine with a low-precision RBF kernel, characterized in that: include: one or more processors; A storage device for storing one or more programs, when the one or more programs are executed by the one or more processors, enables the one or more processors to implement the ncRNA recognition method of the support vector machine with a low-precision RBF kernel as described in any one of claims 1 to 3.