iEnhancer-DCSA: identifying enhancers via dual-scale convolution and spatial attention.

Area of Science:

• Bioinformatics
• Genomics
• Computational Biology

Background:

• Identifying enhancers and their regulatory strength presents significant bioinformatics challenges due to their dynamic nature.
• Deep learning models have advanced enhancer detection, but often overlook multi-scale motif information or treat all features equally.
• Effectively integrating diverse motif information while disregarding irrelevant data is crucial for accurate enhancer prediction.

Purpose of the Study:

• To develop an accurate and stable predictor, named iEnhancer-DCSA, for enhancer identification and classification.
• To address limitations in existing methods by effectively using multi-scale motif information and spatial attention.

Main Methods:

• The iEnhancer-DCSA predictor employs dual-scale fusion to extract features from motifs of varying lengths.
• Spatial attention mechanisms are incorporated to selectively focus on critical features, enhancing prediction accuracy.
• The model automatically extracts and processes motif features, reducing the need for manual feature engineering.

Main Results:

• iEnhancer-DCSA demonstrated superior performance compared to state-of-the-art methods on test datasets.
• Enhancer identification accuracy improved by 3.45% and MCC by 9.41%.
• Enhancer classification accuracy increased by 7.65% and MCC by 18.1%, with ablation studies confirming the efficacy of dual-scale fusion and spatial attention.

Conclusions:

• iEnhancer-DCSA serves as a valuable computational tool for identifying and classifying enhancers.
• The predictor is particularly effective for novel enhancers not present in the training dataset.
• The model's ability to handle multi-scale features and spatial attention offers a significant advancement in enhancer prediction.