Understanding Transcriptional Regulatory Redundancy by Learnable Global Subset Perturbations

Area of Science:

• Genomics
• Computational Biology
• Molecular Biology

Background:

• Transcriptional regulation by cis-regulatory elements (CREs) is vital for biological functions.
• Disruption of CREs can lead to diseases, and their redundancy necessitates methods to identify collaborative regulatory sets.
• Understanding how CREs work together is key to deciphering gene expression control.

Purpose of the Study:

• To introduce GRIDS, an in silico computational method for dissecting combinatorial CRE effects.
• To identify sets of CREs that collaboratively regulate gene expression effectively, accounting for redundancy.
• To provide global explanations of gene regulation by CREs.

Main Methods:

• GRIDS employs a two-phase approach as a global feature explanation challenge.
• It constructs a differentiable surrogate function to model complex gene regulatory processes and enable cross-modal translations.
• Learnable perturbations within a state transition framework are used for efficient navigation of the combinatorial feature landscape.

Main Results:

• GRIDS demonstrates superior explanatory capabilities compared to existing methods in comprehensive benchmarks.
• Global explanations from GRIDS reveal intricate regulatory redundancy across diverse cell types and states.
• The method effectively navigates the complex combinatorial feature landscape of CREs.

Conclusions:

• GRIDS offers a powerful computational approach to understand combinatorial cis-regulatory element effects.
• The findings highlight the significance of regulatory redundancy in cellular regulation.
• GRIDS has the potential to significantly advance biological research in gene expression and disease mechanisms.