EMDLP
EMDLP identifies RNA methylation sites using natural language processing and a multiscale deep learning architecture that combines dilated convolutional neural networks and Bidirectional LSTM to predict m1A and m6A modification positions.
Key Features:
- Sequence representation: Uses NLP-based encodings including RNA word embedding, One-hot encoding, and RGloVe to represent RNA sequences.
- Model architecture: Employs a Dilated Convolutional Bidirectional LSTM (DCB) network integrating dilated convolutional neural networks (DCNNs) with BiLSTM layers.
- Multiscale feature extraction: Captures local and global sequence context via dilated convolutions for local responses and BiLSTM for sequential context.
- Ensemble integration: Combines predictions from different encodings/models using a soft voting mechanism.
- Performance metrics: Reports AUROC scores of 95.56% for m1A and 85.24% for m6A, surpassing existing state-of-the-art models.
Scientific Applications:
- m1A site prediction: Predicts m1A methylation sites in RNA sequences.
- m6A site prediction: Predicts m6A methylation sites in RNA sequences.
- Epi-transcriptome studies: Facilitates investigation of RNA modifications involved in post-transcriptional regulation.
- Method comparison and benchmarking: Enables evaluation of predictive performance against other models using AUROC.
Methodology:
EMDLP represents RNA sequences with RNA word embedding, One-hot encoding, and RGloVe; inputs these encodings into a Dilated Convolutional Bidirectional LSTM (DCB) network combining DCNNs and BiLSTM; and integrates model outputs via a soft voting mechanism.
Topics
Details
- License:
- Not licensed
- Cost:
- Free of charge
- Tool Type:
- web application
- Operating Systems:
- Mac, Linux, Windows
- Programming Languages:
- Python
- Added:
- 9/13/2022
- Last Updated:
- 11/24/2024
Operations
Publications
Wang H, Liu H, Huang T, Li G, Zhang L, Sun Y. EMDLP: Ensemble multiscale deep learning model for RNA methylation site prediction. BMC Bioinformatics. 2022;23(1). doi:10.1186/s12859-022-04756-1. PMID:35676633. PMCID:PMC9178860.