An Explainable Deep Learning Framework Integrating DNA Sequence and Transcription Initiation Signals for Gene...

Area of Science:

• Genomics
• Computational Biology
• Molecular Biology

Background:

• Gene expression regulation is crucial for cellular function and phenotype.
• Predicting gene expression accurately is a challenge in human genetics.
• Existing deep learning models often neglect transcription initiation signals and lack interpretability.

Purpose of the Study:

• To develop an interpretable deep learning framework for predicting gene expression levels.
• To integrate DNA sequence, mRNA half-life, and transcription initiation signals.
• To enhance understanding of gene regulatory mechanisms.

Main Methods:

• Developed an interpretable deep learning framework using convolutional neural networks and Gated Recurrent Units.
• Integrated DNA sequence, mRNA half-life, and transcription initiation signals as input features.
• Validated the model on GM12878, HepG2, and K562 human cell line datasets.

Main Results:

• The framework efficiently predicts gene expression levels.
• The model demonstrates robustness across different human cell lines.
• Interpretability analysis provides valuable insights into gene expression mechanisms.

Conclusions:

• The developed deep learning framework is a robust tool for predicting gene expression.
• The model enhances the understanding of gene regulatory mechanisms.
• This approach offers practical utility for gene expression studies.