EAP-LSTM: A Bi-LSTM-Based Deep Learning Framework for Quantitatively Predicting Enhancer Activity in Drosophila and...

Area of Science:

• Genomics and Bioinformatics
• Molecular Biology
• Computational Biology

Background:

• Enhancer activity is crucial for gene regulation, impacting development and disease.
• Accurate prediction of enhancer activity is vital for understanding gene regulatory mechanisms.
• Current prediction models require improvement for broader applicability.

Purpose of the Study:

• To develop a novel deep learning framework, EAP-LSTM, for quantitative enhancer activity prediction.
• To evaluate EAP-LSTM's performance across different species and cell lines.
• To investigate the role of transcription factor binding sites in enhancer function.

Main Methods:

• Developed EAP-LSTM (Enhancer Activity Prediction based on Bi-LSTM) framework.
• Integrated DNA sequence features (Word2Vec, k-mer variants) and epigenomic data.
• Validated model on six diverse cell lines (human and Drosophila).

Main Results:

• EAP-LSTM consistently outperformed state-of-the-art models (DeepSTARR, HEAP) across all datasets.
• Achieved a PCC of 0.7944 on K562 data, surpassing competitors.
• Demonstrated robust performance in small-sample learning scenarios.
• Identified critical motifs within transcription factor binding sites linked to enhancer activity.

Conclusions:

• EAP-LSTM offers a significant advancement in enhancer activity prediction accuracy.
• The framework provides valuable insights into the molecular mechanisms of enhancer function.
• Findings contribute to a better understanding of gene regulation in development and disease.