Cross-species regulatory sequence activity prediction.

Area of Science:

• Genomics
• Computational Biology
• Machine Learning

Background:

• Machine learning models predict nucleic acid regulatory activity, aiding gene regulation studies.
• Human genome research is extensive, but model organisms offer unique data for training and analysis.
• Model organism data, including tissue and cell states, is underutilized in genomic studies.

Purpose of the Study:

• To develop a strategy for training deep convolutional neural networks on multiple genomes simultaneously.
• To improve gene expression prediction accuracy by training on combined human and mouse genomic data.
• To apply cross-species regulatory models for analyzing human genetic variants linked to disease.

Main Methods:

• Developed a deep convolutional neural network strategy for simultaneous multi-genome training.
• Applied the models to large datasets of human and mouse gene expression data.
• Utilized trained mouse regulatory models to analyze human genetic variants associated with molecular phenotypes and disease.

Main Results:

• Simultaneous training on human and mouse genomes significantly improved gene expression prediction accuracy.
• Enhanced prediction accuracy was observed for held-out sequences and genetic variants.
• Demonstrated a novel method for applying mouse regulatory models to human genetic variant analysis.

Conclusions:

• Multi-genome deep learning enhances the predictive power of gene regulatory models.
• Cross-species model application provides a powerful new approach for human genetic variant interpretation.
• Leveraging model organism data accelerates the investigation of gene regulation in human diseases.