4mC locus prediction method based on convolutional fuzzy neural network
By introducing a convolutional fuzzy neural network into the 4mC site prediction method, extracting local features and selecting membership values, it solves the problem that traditional methods are difficult to deal with large data and high characteristics, and achieves better model generalization ability and test performance.
Patent Information
- Application Number
- CN202411831819.2
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-12-12
- Publication Date
- 2025-03-28
AI Technical Summary
The traditional 4mC site prediction method is difficult to process large data, and has high requirements for input characteristics, so it cannot effectively filter outlier noise values, resulting in large limitations in the model and poor test performance.
The 4mC site prediction method based on the convolutional fuzzy neural network is adopted to extract local features through the convolutional neural network and input them into the fuzzy neural network for selection of membership values, and a convolutional fuzzy neural network is constructed for prediction.
This method can effectively process large data, reduce the requirements for input features, filter outlier noise values, and improve the generalization ability and test performance of the model.
Smart Images
Figure CN119724361A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of site prediction, and in particular to a 4mC site prediction method based on convolutional fuzzy neural network. Background Art
[0002] The prediction of N4-methylcytosine (4mC) sites helps to deeply understand the disease mechanism and promote drug development. However, traditional biological experimental methods not only consume a lot of time and financial resources, but also cannot meet the needs of high-throughput data.
[0003] Among the 4mC site prediction methods, the prediction method based on convolutional neural networks can extract local features very well, but this method is overly dependent on the quality of the input features, has high requirements for the input features, cannot filter outlier noise values, has large model limitations, and performs poorly in the test set.
[0004] Fuzzy networks can only process small data sets. As the input data increases, the number of nodes in the fuzzy network will explode exponentially. When the number of inputs is too large, the combination of fuzzy reasoning becomes too complex, and these data volumes are difficult to implement with any hardware. Therefore, traditional fuzzy networks have difficulty calculating large data. Summary of the invention
[0005] In view of this, the purpose of the present invention is to propose a 4mC site prediction method based on convolutional fuzzy neural network to solve the technical problem that existing methods are difficult to calculate large data.
[0006] The technical means adopted by the present invention are as follows:
[0007] A 4mC site prediction method based on convolutional fuzzy neural network comprises the following steps:
[0008] S1. Obtain a DNA data set, preprocess the DNA data set, represent the preprocessed data set and perform multi-scale feature fusion;
[0009] S2, using convolutional neural network to extract local features of data feature representation to obtain important features;
[0010] S3, sending the important features into the fuzzy neural network to select the membership value, thereby constructing a convolutional fuzzy neural network;
[0011] S4. Use convolutional fuzzy neural network to predict the data to be predicted and obtain the prediction result.
[0012] Furthermore, S1 specifically includes the following steps:
[0013] There are six data sets, each of which has an input DNA sequence length of 41 nucleobases, a middle nucleobase of cytosine, and 20 different nucleobases upstream and downstream; an equal number of positive and negative samples are randomly selected to construct a balanced data set;
[0014] The input DNA sequence is represented by features and multi-scale feature fusion is performed.
[0015] Furthermore, S2 specifically includes the following steps:
[0016] CNN is used to extract feature information, and the convolution layer with a convolution kernel of (3, 3) is used to extract local features;
[0017] Use the maximum pooling layer to reduce the dimension and build a deep network structure by stacking multiple convolutional layers and pooling layers to capture the high-level semantic information and global information of the input data;
[0018] The feature data obtained by the convolution layer and the pooling layer are used as the input of the fuzzy neural network.
[0019] Further, in S3, the convolutional fuzzy neural network includes two sub-networks: a convolutional neural network and a fuzzy neural network;
[0020] The convolutional fuzzy neural network includes the first to eighth layers:
[0021] The first layer: input layer, the DNA data set is divided into training set and validation set according to the ratio of 8:2, of which 80% is the training set and 20% is the test set; the training set X = [x1,...,x i ,...,x N ]∈R L*N ; where x i represents the i-th training data. There are N training data in total, and each training data consists of L nucleotides;
[0022] The second layer: feature representation layer, which encodes DNA sequences using six different DNA sequence encoding schemes: cumulative nucleotide frequency, binary encoding, electron-ion interaction, nucleotide composition, chemical properties of nucleotides, and 3-nucleotide composition;
[0023] The third layer: aggregation layer, which fuses the six individual feature vectors to obtain a feature set called Fusion6, or F for short, which is used for subsequent feature extraction;
[0024] The fourth layer: convolutional layer, the feature set F obtained by fusing the features of the third layer is sent to CNN for feature extraction, and the result of the fourth layer is in, The number of mappings is s, and there are M fuzzy sets in the membership function;
[0025] Fifth layer: Send each feature map separately into the kth fuzzy set to obtain:
[0026]
[0027] in, is a fuzzy membership function, and a Gaussian membership function is used to construct the classifier, x ij is the j-th dimension feature of the i-th feature input; the membership function has 3 fuzzy sets;
[0028] The sixth layer: Aggregate the results obtained by mapping different features in the same fuzzy set:
[0029]
[0030] The seventh layer: Normalize the aggregated results to get:
[0031]
[0032] The output of the kth fuzzy set is:
[0033]
[0034] The output for all fuzzy sets is:
[0035]
[0036] The eighth layer: The overall parameters are expressed as:
[0037] W=(W 1 ,W 2 ,...,W M )
[0038] The predicted values are:
[0039] Furthermore, S4 also includes optimizing the convolutional fuzzy neural network based on the training set, specifically including model parameter optimization and model hyperparameter optimization;
[0040] The loss between the true value and the predicted value is calculated through the cross entropy loss function, and the model parameters are optimized through the gradient descent method;
[0041] The optimal hyperparameters are found through grid analysis.
[0042] Furthermore, S4 specifically includes the following steps:
[0043] The data to be predicted is input into the convolutional fuzzy neural network model. After feature representation, feature extraction, and probability prediction, the final output is the probability value of cytosine being a 4mC site. The site with a probability value greater than the rated value is a 4mC site.
[0044] The present invention also provides a storage medium, which includes a stored program, wherein when the program is run, any of the above-mentioned 4mC site prediction methods based on convolutional fuzzy neural networks is executed.
[0045] The present invention also provides an electronic device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes any one of the above-mentioned 4mC site prediction methods based on convolutional fuzzy neural networks through the computer program.
[0046] Compared with the prior art, the present invention has the following advantages:
[0047] The technical solution provided by the present invention first performs PSTNP encoding on the input data, then uses a convolutional neural network to extract local features from the encoded data, and then sends the extracted features to FNN for membership value selection, assigning a lower membership value to the outlier value, and vice versa. We finally predict the data of the test set to obtain the prediction result. The present invention not only uses CNN to extract local features, but also uses FNN to assign different membership values to the input data. The present invention has low requirements on the quality of the input features, can filter outlier noise values, has small model limitations, and performs well in the test set. BRIEF DESCRIPTION OF THE DRAWINGS
[0048] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative labor.
[0049] Figure 1 This is a convolutional fuzzy neural network (CFNN) framework diagram of the present invention.
[0050] Figure 2 This is the network model diagram of the FNN of the present invention. DETAILED DESCRIPTION
[0051] In order to enable those skilled in the art to better understand the scheme of the present invention, the technical scheme in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work should fall within the scope of protection of the present invention.
[0052] It should be noted that the terms "first", "second", etc. in the specification and claims of the present invention and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence. It should be understood that the data used in this way can be interchanged where appropriate, so that the embodiments of the present invention described herein can be implemented in an order other than those illustrated or described herein. In addition, the terms "including" and "having" and any variations thereof are intended to cover non-exclusive inclusions, for example, a process, method, system, product or device that includes a series of steps or units is not necessarily limited to those steps or units that are clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.
[0053] like Figure 1 and 2 As shown, the present invention provides a 4mC site prediction method based on convolutional fuzzy neural network, comprising the following steps:
[0054] S1, obtain a data set, preprocess the data set, and perform PSTNP encoding on the preprocessed data set;
[0055] We have a total of six data sets, each of which has an input DNA sequence length of 41 nucleobases (bp), with cytosine (C) in the middle and 20 different nucleobases upstream and downstream. We randomly selected an equal number of positive and negative samples to construct a balanced data set.
[0056] The input DNA sequence is subjected to feature extraction and multi-scale feature fusion.
[0057] S2, using convolutional neural network to extract local features of the data set to obtain important features;
[0058] For 4mC sites, different organisms have their own unique nucleobase preferences. If we can extract this preference information from the training set, we can use this preference information to build a model and apply it to the test set.
[0059] CNN is a deep learning model that can effectively process high-dimensional data. CNN can automatically learn features without manually designing or selecting feature extractors. The deep CNN network structure can improve the model's expressiveness and generalization capabilities.
[0060] Based on this, we use CNN to extract feature information and use the convolution layer with a convolution kernel of (3, 3) to extract local features. Use the maximum pooling layer to reduce the dimension, thereby reducing the number of parameters and computational complexity. By stacking multiple convolutional layers and pooling layers to build a deep network structure, we can capture the high-level semantic information and global information of the input data, thereby improving the generalization ability of the model. We use the feature data obtained by the convolutional layer and the pooling layer as the input of the next fuzzy neural network (FNN).
[0061] S3, sending the important features into the fuzzy neural network to select the membership value, thereby constructing a convolutional fuzzy neural network;
[0062] The original 4mC site prediction model directly extracts feature information from the input sequence, without considering the negative effects of noise points and outliers on the model. To solve this problem, we introduced the concept of fuzziness, which assigns lower membership values to those noise points and outliers to reduce their impact on the model, thereby improving the generalization ability of the model.
[0063] FNN has five layers in total, namely input layer, fuzzification layer, fuzzy rule layer, normalization layer, and output layer.
[0064] The first layer is the input layer, which takes the eigenvalues in the fully connected layer as input. The second layer is the fuzzification layer, where each input component corresponds to a group of nodes, with the number of nodes being m. i is relative to x i The number of fuzzy classifications performed, where each node represents a fuzzy verbal variable value. The role of each node in the second layer is to calculate the membership function of each component of the input vector belonging to each verbal variable fuzzy set. Each node in the third layer represents a fuzzy rule, which is used to match the antecedent of the fuzzy rule and calculate the practicality of each rule. There are n groups of membership functions in the second layer, and one membership function is taken from each group without duplication to form the nodes of the third layer. For a given input, only those linguistic variable values near the input point have a larger membership value, and the membership of the linguistic variable values far away from the input point is either very small (Gaussian membership function) or 0 (triangular membership function). When the membership function is very small (for example, less than 0.05), it is approximately taken as 0, so in the output a of the third layer, only a few nodes have non-zero outputs, while the outputs of most nodes are 0, which is similar to the local approximation network. The number of nodes in the fourth layer is the same as that of the third layer, that is, the applicability of each rule is normalized. Only a few nodes in this layer have large outputs, and the outputs of most nodes are close to zero. The fifth layer is the output layer, which realizes the clear calculation, that is,
[0065] Many quantities of the FNN model structure need to be set in advance, such as the number of fuzzy classifications for each input component; the selection of the membership function form (normal distribution, triangular distribution, trapezoidal distribution, etc.). The parameters that need to be learned are mainly the weights W of the last layer. ij As well as the center value and width of the membership function of the second layer. FNN is essentially a multi-layer feedforward network, so the parameter adjustment learning algorithm can be designed by imitating the BP network using the error back propagation method. The gradient of each layer is calculated based on the partial derivative of the previous layer. When the actual output of the network is consistent with the ideal output, it indicates that the training is over. Otherwise, the parameters of each layer are corrected through error back propagation until the error falls within the required range. When the actual output of the network is consistent with the ideal output, it indicates that the training is over. Otherwise, the parameters of each layer are corrected through error back propagation until the error falls within the required range.
[0066] The first layer: input layer, the DNA data set is divided into training set and validation set in a ratio of 8:2, of which 80% is the training set and 20% is the test set; the training set X = [x1,...,x i ,...,x N ]∈R L*N ; where x i represents the i-th training data. There are N training data in total, and each training data consists of L nucleotides;
[0067] The second layer: the feature representation layer, uses six different DNA sequence encoding schemes to encode DNA sequences: Accumulated nucleotide frequency (ANF), binary encoding (binary), electron-ion interaction (EIIP), nucleotide composition (NAC), nucleotide chemical property (NCP), and 3-nucleotide composition (3NCs).
[0068] The third layer is the aggregation layer. We fuse the six individual feature vectors to obtain a feature set called Fusion6, or F for short, which is used for subsequent feature extraction.
[0069] The fourth layer is the convolution layer. We send the feature set F obtained by fusing the features of the third layer into CNN for feature extraction, and the result of the fourth layer is in, The number of mappings is s, and there are M fuzzy sets in the membership function.
[0070] Fifth layer: Send each feature map separately into the kth fuzzy set to obtain:
[0071]
[0072] in, is a fuzzy membership function, and a Gaussian membership function is used to construct the classifier, x ij The j-th dimension feature of the i-th feature input. The membership function has three fuzzy sets;
[0073] The sixth layer: Aggregate the results obtained by mapping different features in the same fuzzy set:
[0074]
[0075] The seventh layer: Normalize the aggregated results to get:
[0076]
[0077] The output of the kth fuzzy set is:
[0078]
[0079] The output for all fuzzy sets is:
[0080]
[0081] The eighth layer: The overall parameters are expressed as:
[0082] W=(W 1 ,W 2 ,...,W M )
[0083] The predicted values are:
[0084] S4. Use convolutional fuzzy neural network to predict the data to be predicted and obtain the prediction result.
[0085] Predict the probability that cytosine is a 4mC site, and distinguish whether it is a 4mC site or not. Use a nonlinear activation function to output the final probability value of a 4mC site. Use the cross entropy loss function to calculate the loss between the true value and the predicted value, and further optimize the parameters within the model.
[0086] Model testing:
[0087] Based on the known 4mC sites, 80% of the balanced sample data is selected as the training set, and the remaining 20% of the balanced sample data is selected as the test set. The data passes through the CNN feature extraction module and the FNN membership value assignment module in the model to finally obtain the probability prediction value of the 4mC site. The loss value between the real sample label value and the model prediction value is recorded, as well as the indicator value for evaluating the model classification ability.
[0088] Test various parameters and important modules that affect model performance. Observe the impact of the number of channels of the feature extractor, the feature dimensions of the network nodes, and the number of fuzzy nodes on the model test results. Develop a method to remove the CNN feature extraction module or the FNN membership value assignment module, and test the model results containing different modules.
[0089] Test the model's ability to predict 4mC sites in the test set. For our six data sets, output the model's performance on the six data sets.
[0090] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or replace some or all of the technical features therein with equivalents. However, these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.
Claims
1. A 4mC site prediction method based on convolutional fuzzy neural network, characterized in that: The steps include: S1. Obtain a DNA data set, preprocess the DNA data set, represent the preprocessed data set and perform multi-scale feature fusion; S2, using convolutional neural network to extract local features of data feature representation to obtain important features; S3, sending the important features into the fuzzy neural network to select the membership value, thereby constructing a convolutional fuzzy neural network; S4. Use convolutional fuzzy neural network to predict the data to be predicted and obtain the prediction result.
2. The 4mC site prediction method based on convolutional fuzzy neural network according to claim 1, characterized in that: S1 specifically includes the following steps: There are six data sets, each of which has an input DNA sequence length of 41 nucleobases, a middle nucleobase of cytosine, and 20 different nucleobases upstream and downstream; an equal number of positive and negative samples are randomly selected to construct a balanced data set; The input DNA sequence is represented by features and multi-scale features are fused.
3. The 4mC site prediction method based on convolutional fuzzy neural network according to claim 1, characterized in that: S2 specifically includes the following steps: CNN is used to extract feature information, and the convolution layer with a convolution kernel of (3, 3) is used to extract local features; Use the maximum pooling layer to reduce the dimension and build a deep network structure by stacking multiple convolutional layers and pooling layers to capture the high-level semantic information and global information of the input data; The feature data obtained by the convolution layer and the pooling layer are used as the input of the fuzzy neural network.
4. The 4mC site prediction method based on convolutional fuzzy neural network according to claim 1, characterized in that: In S3, the convolutional fuzzy neural network includes two sub-networks: a convolutional neural network and a fuzzy neural network; The convolutional fuzzy neural network includes the first to eighth layers: The first layer: input layer, the DNA data set is divided into training set and validation set in a ratio of 8:2, of which 80% is the training set and 20% is the test set; the training set X = [x1,...,x i ,...,x N ]∈R L*N ; where x i represents the i-th training data. There are N training data in total, and each training data consists of L nucleotides; The second layer: feature representation layer, which encodes DNA sequences using six different DNA sequence encoding schemes: cumulative nucleotide frequency, binary encoding, electron-ion interaction, nucleotide composition, chemical properties of nucleotides, and 3-nucleotide composition; The third layer: aggregation layer, which fuses the six individual feature vectors to obtain a feature set called Fusion6, or F for short, which is used for subsequent feature extraction; The fourth layer: convolutional layer, the feature set F obtained by fusing the features of the third layer is sent to CNN for feature extraction, and the result of the fourth layer is in, The number of mappings is s, and there are M fuzzy sets in the membership function; Fifth layer: Send each feature map separately into the kth fuzzy set to obtain: in, is a fuzzy membership function, and a Gaussian membership function is used to construct the classifier, x ij is the j-th dimension feature of the i-th feature input; the membership function has 3 fuzzy sets; The sixth layer: Aggregate the results obtained by mapping different features in the same fuzzy set: The seventh layer: Normalize the aggregated results to get: The output of the kth fuzzy set is: The output for all fuzzy sets is: The eighth layer: The overall parameters are expressed as: In=(In 1 ,IN 2 ,...,IN M ) The predicted values are:
5. The 4mC site prediction method based on convolutional fuzzy neural network according to claim 1, characterized in that: S4 also includes optimizing the convolutional fuzzy neural network based on the training set, specifically including model parameter optimization and model hyperparameter optimization; The loss between the true value and the predicted value is calculated through the cross entropy loss function, and the model parameters are optimized through the gradient descent method; The optimal hyperparameters are found through grid analysis.
6. The 4mC site prediction method based on convolutional fuzzy neural network according to claim 1, characterized in that: S4 specifically includes the following steps: The data to be predicted is input into the convolutional fuzzy neural network model. After feature representation, feature extraction, and probability prediction, the final output is the probability value of cytosine being a 4mC site. The site with a probability value greater than the rated value is a 4mC site.
7. A storage medium, characterized in that: The storage medium includes a stored program, wherein when the program is run, the 4mC site prediction method based on convolutional fuzzy neural network described in any one of claims 1 to 6 is executed.
8. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein: The processor executes the 4mC site prediction method based on convolutional fuzzy neural network according to any one of claims 1 to 6 through the operation of the computer program.