A scalable computational framework for predicting gene expression from candidate cis-regulatory elements.

Area of Science:

• Genomics
• Computational Biology
• Molecular Biology

Background:

• Understanding cis-regulatory element (CRE) function in gene expression is crucial but challenging due to dynamic CREs.
• Predicting gene expression from CREs remains a significant unsolved problem in molecular biology.

Purpose of the Study:

• To develop a scalable computational framework (ScPGE) for predicting gene expression from candidate CREs (cCREs).
• To improve the accuracy of identifying active enhancer-gene interactions and understanding regulatory mechanisms.

Main Methods:

• ScPGE integrates DNA sequences, transcription factor (TF) binding scores, and epigenomic data from cCREs into 3D tensors.
• It employs a hybrid model combining convolutional neural networks and transformers to analyze cCRE-gene relationships.
• Attention mechanisms are utilized to identify key enhancer-gene interactions.

Main Results:

• ScPGE outperforms existing state-of-the-art models in gene expression prediction and enhancer-gene interaction identification.
• Analysis revealed that the regulatory effect of cCREs decreases with distance from the target gene.
• Incorporating chromatin loops enhanced ScPGE's ability to capture distal cCRE-gene interactions.
• ScPGE identified crucial TF motifs and elucidated different regulatory roles of cCREs.

Conclusions:

• ScPGE provides a powerful and scalable framework for deciphering CRE-gene regulatory relationships.
• The model's findings on distance-dependent effects and the utility of chromatin loops offer new insights into gene regulation.
• ScPGE aids in discovering regulatory elements and understanding their functional mechanisms.