Prediction method for identifying enhancer in DNA (deoxyribonucleic acid) sequence and predicting activity intensity of enhancer
Through a predictor based on context language embedding, using single-thermal encoding and nucleotide chemical property feature encoding, combined with Dense Net and self-attention mechanism, the problem of low accuracy of enhancer prediction in the prior art is solved, achieving high accuracy Enhancer predictions for degree and strong generalization ability.
Patent Information
- Application Number
- CN202410992982.0
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-07-24
- Publication Date
- 2024-11-12
AI Technical Summary
Existing enhancers and their intensity prediction methods rely on time-consuming and expensive experimental techniques, and the accuracy of deep learning methods needs to be improved, making it difficult to extract comprehensive nucleotide patterns from DNA sequences.
Using a predictor based on context language embedding, feature encoding is performed through one-hot encoding and nucleotide chemistry, and using improved Dense Net and self-attention mechanism to extract advanced features, combining multi-layer perceptron and integrated learning for prediction, and modeling Tune and cross-validate for improved accuracy.
The prediction accuracy of enhancers and their intensity is significantly improved, and high accuracy and strong generalization capabilities are obtained. The ACC and MCC values of all data sets are higher than that of the state-of-the-art methods.
Smart Images

Figure CN118942544A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of prediction of functional sites in sequence analysis in bioinformatics, and in particular to a method for identifying enhancers in DNA sequences and predicting their activity strength. Background Art
[0002] Enhancers play a crucial role in gene transcriptional regulation, human disease progression, and cell differentiation. Enhancers are short DNA fragments ranging in length from 100 to 1000 base pairs, located upstream of specific genes. Gene enhancers are non-coding segments of DNA that play a central role in the transcriptional process that regulates development, cell identity, and evolution. Enhancers contain a variety of genetic marker sites, the most common of which are transcription factor binding sites. Enhancers regulate gene expression by interacting with their target gene promoters. This interaction can be cis or trans. Cis-action refers to enhancers and their action site genes on the same chromosome, and trans-action refers to enhancers and their action site genes on different chromosomes. Many human diseases have been shown to be affected by enhancer genetic variations, such as various cancers and inflammatory bowel disease. Therefore, it is of great significance to explore enhancer identification based on DNA sequences.
[0003] For the research on enhancers and their strength identification, for a long time in the past, the prediction of enhancers relied on biological experimental techniques, but they are still time-consuming and expensive. Therefore, it is necessary to develop more effective computational methods or tools to supplement laboratory experiments. With the rapid development of machine learning and deep learning, many prediction models have been established to identify enhancers. However, these methods usually require careful design of manual features or use of an ensemble of multiple models based on different features. Their performance depends largely on the quality of handcrafted features or ensembles. In addition, based on limited experience and domain knowledge, it is difficult to extract comprehensive nucleotide patterns from DNA sequences. Therefore, some researchers began to use deep learning methods to identify enhancers. Although these methods can identify enhancers, the accuracy of the model needs to be further improved, and it should be possible to develop better models using new deep learning frameworks. Summary of the invention
[0004] The purpose of the present invention is to solve the problem of the accuracy of the existing prediction of enhancers and their strengths, and to provide a predictor based on contextual language embedding for identifying enhancers and their strengths. The prediction method can predict enhancers and classify enhancer strengths, and at the same time construct an optimal two-layer classification model, thereby improving the accuracy of predicting enhancers and their strengths.
[0005] The following is a technical solution to achieve the purpose of the present invention, comprising the following steps:
[0006] 1. Collect sample data sets: obtain positive sample DNA sequences and negative sample DNA sequences of training sets and independent test sets;
[0007] 2. Feature encoding: One-hot encoding and nucleotide chemical properties (NCP) are used for feature encoding. Finally, each sequence is represented by a two-dimensional numerical matrix.
[0008] 3. Build the model: After obtaining the feature encoding, use the improved Dense Net to extract high-level features, use the self-attention mechanism to extract relatively important features, and input them into the multi-layer perceptron (MLP) to obtain the final prediction results;
[0009] 4. Model tuning: Use batch normalization, dropout, and early stopping methods to optimize the prediction model;
[0010] 5. Ensemble model: The model uses five-fold cross validation, where each fold is tested using an independent test set, and each test enhancer sequence produces five prediction probabilities, and finally uses ensemble learning for classification voting;
[0011] 6. Model evaluation: Model evaluation was performed based on 5-fold cross validation and independent testing, using sensitivity (Sn), specificity (Sp), accuracy (Acc), Matthews correlation coefficient (MCC) and area under the ROC curve (AUC)
[0012] Conduct performance measurement.
[0013] Beneficial effects:
[0014] This model uses one-hot encoding and nucleotide chemical properties (NCP) for encoding. An improved Dense Net is used to extract high-level features from the original encoding. The importance of high-level features is evaluated using the self-attention module and then input into the fully connected neural network to obtain the prediction probability. Finally, the final classification results are integrated through voting. The experimental results show that we have achieved significant performance improvement compared with existing methods. The ACC values and MCC of all datasets are higher than the state-of-the-art methods, which shows that our proposed enhancer prediction model has strong generalization ability while having high accuracy. BRIEF DESCRIPTION OF THE DRAWINGS
[0015] Figure 1 A flowchart of a method for identifying enhancers and predicting their strength;
[0016] Figure 2 This is the ROC curve of the first layer of the model identifying enhancers under 10 rounds of 5-fold cross validation;
[0017] Figure 3 This is the ROC curve of the second-layer model identifying enhancer strength under 10 rounds of 5-fold cross validation.
[0018] Figure 4 For standard data set table
[0019] Figure 5 Summary of NCP feature encoding
[0020] Figure 6 Comparison of enhancer identification with state-of-the-art methods on an independent test set
[0021] Figure 7 Comparison of enhancer strength prediction with state-of-the-art methods on an independent test set DETAILED DESCRIPTION
[0022] The specific implementation modes of the present invention are explained below in conjunction with the accompanying drawings. The accompanying drawings are for illustrative purposes only and cannot be understood as limiting the present invention. The accompanying drawings are for reference and illustration only and do not constitute a limitation on the scope of patent protection of the present invention, because many changes can be made to the present invention without departing from the spirit and scope of the present invention.
[0023] like Figure 1 As shown, Figure 1 The workflow of this work is shown. One-hot encoding and nucleotide chemical properties (NCP) are used for encoding, and an improved densely connected network (Dense Net) is used to extract high-level features from the original encoding. The importance of high-level features is evaluated using the self-attention mechanism, and then input into the fully connected neural network to obtain the predicted probability. Finally, the final classification results are integrated and learned by voting to obtain the final predicted classification results.
[0024] To build and evaluate the constructed model, we used two data sets for two-layer models. The first layer of the model identified whether it was an enhancer, and the second layer of the model classified the enhancer strength. The two-layer model used the same network structure and parameters, only the input data was different. The specific information of the data set is shown in Table 1 below. In addition, the NCP coding scheme is summarized in Table 2 below.
[0025] Table 1. Standard datasets
[0026]
[0027] Table 2. Summary of NCP feature encoding
[0028]
[0029] We used 10 rounds of 5-fold cross validation on the training set to test the effectiveness of our proposed model, and used an independent dataset to verify the generalization ability of the predictor. To evaluate the predictor, we used the following five indicators: sensitivity (Sn), specificity (Sp), accuracy (Acc), Matthews correlation coefficient (MCC), and area under the ROC curve (AUC).
[0030] Finally, our method is compared with the existing most advanced methods. The comparison results of independent tests are shown in Tables 3 and 4. From the comparison results, it can be seen that the prediction accuracy of the two-layer model in this example is significantly better than that of other methods.
[0031] Table 3. Comparison of enhancer identification with state-of-the-art methods on an independent test set
[0032]
[0033] Table 4. Comparison of enhancer strength prediction with state-of-the-art methods on an independent test set
[0034]
Claims
1. A method for identifying enhancers and predicting their activity strength, the process comprising the following steps: 1) Collect sample data sets: obtain positive sample DNA sequences and negative sample DNA sequences of training sets and independent test sets; 2) Feature encoding: One-hot encoding and nucleotide chemical properties (NCP) are used for feature encoding. Finally, each sequence is represented by a two-dimensional numerical matrix. 3) Model construction: After obtaining the feature encoding, the improved dense network (Dense Net) is used to extract high-level features, the self-attention mechanism is used to evaluate the relatively important features, and the features are input into the multi-layer perceptron (MLP) to obtain the final prediction results; 4) Model tuning: Use batch normalization, dropout, and early stopping methods to optimize the prediction model. 5) Ensemble model: The model uses five-fold cross validation, where each fold is tested using an independent test set, and each test enhancer sequence produces five prediction probabilities, and finally uses ensemble learning for classification voting; 6) Model evaluation: Model evaluation was performed based on 5-fold cross validation and independent testing, using sensitivity (Sn), specificity (Sp), accuracy (Acc), and Matthews correlation coefficient (MCC) to measure performance.
2. This method uses single-hot and nucleotide chemical properties (NCP) for encoding, and uses an improved densely connected network (Dense Net) to extract high-level features from the original encoding. The importance of high-level features is evaluated using the self-attention mechanism, and then input into the fully connected neural network to obtain the predicted probability. Finally, the final classification results are integrated and learned by voting.
3. A method for identifying enhancers and predicting their activity strength according to claim 1, characterized in that The feature encoding scheme and model construction described in step 2) and step 3) enable the model to retain and capture more important information while achieving high performance prediction results.
