iDNA-OpenPrompt

iDNA-OpenPrompt leverages the OpenPrompt learning framework to identify DNA methylation sites by combining prompt templates, a prompt verbalizer, and pre-trained language models with a DNA vocabulary library, BERT tokenizer, and specific label words.

Key Features:

• OpenPrompt Learning Framework: The model constructs a prompt-learning framework for DNA methylation sequences using a prompt template, prompt verbalizer, and Pre-trained Language Model (PLM).
• DNA-Specific Vocabulary and Tokenization: A DNA vocabulary library, BERT tokenizer, and specific label words are integrated to enable accurate identification of DNA methylation sites.
• Multi-Species and Multi-Modification Coverage: The model is evaluated on 17 benchmark datasets covering various species and three DNA methylation modifications (4mC, 5hmC, 6mA).
• Performance Evaluation: Extensive analysis evaluates predictive, reliability, and consistency capabilities across benchmark datasets.

Scientific Applications:

• DNA Methylation Site Identification: Identifies DNA methylation sites across different species and methylation modification types without labor-intensive bisulfite sequencing.
• 4mC Prediction: Predicts N4-methylcytosine modification sites from DNA sequences.
• 5hmC Prediction: Predicts 5-hydroxymethylcytosine modification sites from DNA sequences.
• 6mA Prediction: Predicts N6-methyladenine modification sites from DNA sequences.
• Cross-Species Epigenetic Analysis: Supports analysis across various species through evaluation on diverse benchmark datasets.

Methodology:

The model combines a prompt template, prompt verbalizer, and Pre-trained Language Model (PLM) to construct the prompt-learning framework for DNA methylation sequences, with a DNA vocabulary library, BERT tokenizer, and specific label words introduced to enable accurate identification.