Efficient High-Throughput DNA Breathing Features Generation Using Jax-EPBD

Area of Science:

• Computational Biology
• Molecular Dynamics
• Genomics

Background:

• DNA breathing dynamics, transient base-pair openings, are crucial for DNA-templated processes.
• Existing models like EPBD face computational limits for analyzing long DNA sequences.

Purpose of the Study:

• Introduce JAX-EPBD, a GPU-accelerated framework for high-throughput DNA breathing simulations.
• Enhance the scalability and speed of analyzing DNA dynamics for genome-scale applications.

Main Methods:

• Developed JAX-EPBD using JAX for GPU acceleration, enabling Langevin molecular dynamics.
• Achieved up to 30x speedup and improved scalability over traditional C-based EPBD.
• Applied JAX-EPBD to predict transcription factor binding affinity and analyze promoter dynamics.

Main Results:

• JAX-EPBD efficiently captures time-dependent DNA breathing behaviors like bubble lifetimes and base flipping kinetics.
• Incorporating DNA breathing features improved transcription factor binding affinity predictions.
• Observed sequence-specific bubble dynamics in the AAV P5 promoter correlating with transcriptional activity.

Conclusions:

• JAX-EPBD is a powerful, scalable tool for studying DNA breathing dynamics.
• The framework facilitates understanding DNA breathing's role in gene regulation and TF binding.
• Enables genome-scale analyses of DNA dynamics for biological insights.