Cross-species analysis of enhancer logic using deep learning.

Area of Science:

• Genomics
• Computational Biology
• Cancer Research

Background:

• Understanding the genomic regulatory code of enhancers is crucial for cellular identity and interpreting noncoding variation.
• Enhancer function knowledge can improve cell type-specific gene therapy drivers.
• Melanoma, with its distinct cell states, presents a relevant model for studying enhancer codes.

Purpose of the Study:

• To develop explainable enhancer models using deep learning and cross-species chromatin accessibility profiling.
• To decipher the enhancer code in melanoma, focusing on distinct cell states.
• To identify transcription factor binding sites and understand enhancer evolution.

Main Methods:

• Trained and validated a deep learning model (DeepMEL) using chromatin accessibility data from 26 melanoma samples across six species.
• Applied DeepMEL to analyze enhancer architectures and transcription factor binding sites in different melanoma states.
• Utilized DeepMEL to identify orthologous enhancers across species and pinpoint nucleotide substitutions in enhancer turnover.

Main Results:

• DeepMEL demonstrated high accuracy on the CAGI5 challenge, outperforming existing models for the melanoma enhancer of *IRF4*.
• The model identified specific transcription factor roles in nucleosome displacement and enhancer activation within distinct melanoma cell states.
• DeepMEL successfully identified conserved enhancers across distantly related species and highlighted nucleotide substitutions driving enhancer evolution.

Conclusions:

• DeepMEL provides a powerful tool for predicting and optimizing candidate enhancers, and prioritizing mutations, applicable to various cancer and normal cell types.
• The computational strategy advances the understanding of enhancer function and regulatory code.
• This approach has broad implications for gene therapy and personalized medicine.