iEnhancer-ELM: improve enhancer identification by extracting position-related multiscale contextual information based...

Area of Science:

• Genomics
• Bioinformatics
• Computational Biology

Background:

• Enhancers are crucial cis-regulatory elements controlling gene transcription.
• Existing feature extraction methods struggle to capture position-dependent, multi-scale contextual information from DNA sequences.

Purpose of the Study:

• To propose a novel enhancer identification method, iEnhancer-ELM, leveraging BERT-like enhancer language models.
• To address the limitations of current methods in learning position-related multiscale contextual information from raw DNA sequences.

Main Methods:

• iEnhancer-ELM tokenizes DNA sequences using multi-scale k-mers.
• It employs a multi-head attention mechanism to extract positional and contextual information from different k-mer scales.
• The method evaluates and ensembles various k-mer scales for optimal performance.

Main Results:

• The proposed iEnhancer-ELM model demonstrates superior performance compared to state-of-the-art methods on benchmark datasets.
• A case study revealed 30 potential enhancer motifs using a 3-mer model, with 12 validated by STREME and JASPAR.
• The model shows interpretability and potential for uncovering enhancer biological mechanisms.

Conclusions:

• iEnhancer-ELM offers an effective approach for enhancer identification by capturing essential sequence features.
• The method provides insights into enhancer motifs and their regulatory roles.
• The developed models and code are publicly available for further research.