基于剪枝预训练模型与人工特征编码融合的4mC位点识别方法
By fusing the pruned pre-trained model DNABert with artificial feature encoding, the problem of insufficient DNA sequence feature representation was solved, the recognition accuracy of the 4mC site was improved, and more accurate prediction results were achieved.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- GUANGDONG UNIV OF TECH
- Filing Date
- 2023-09-07
- Publication Date
- 2026-07-17
AI Technical Summary
Existing DNA sequence feature characterization capabilities are insufficient, resulting in low prediction accuracy in 4mC site identification, especially with overfitting and computational resource consumption issues on small datasets.
We employ a method that combines the pruned pre-trained model DNABert with artificial feature encoding. We expand the feature representation space through Kmer encoding and CKSNAP encoding, and combine a bidirectional LSTM network and an attention fusion module to extract deep and shallow feature information. Finally, we use a feedforward neural network for classification and prediction.
It significantly improves the identification accuracy of 4mC sites on six independent benchmark test datasets, outperforming existing models and achieving more accurate prediction results.
Smart Images
Figure CN117216656B_ABST