Randomized trial demonstrates a novel Bayesian neural network modeling TF-DNA interactions, suggesting improved understanding of gene regulation.
Key Points
This research aims to develop a Bayesian neural network approach for understanding transcription factor interactions with DNA by integrating sequence and methylation data.
Developed BayesPI-FLY, a Bayesian neural network for de novo motif discovery.
Utilized a two-layer inference architecture for model parameter estimation.
Validated on synthetic and high-throughput sequencing datasets, including whole-genome bisulfite sequencing.
Successfully characterized methylation effects on TF binding at single-nucleotide and motif levels.
Generated position weight matrices and sequence logos for motif interpretation.
Reproduced known methylation-associated TF-binding patterns and inferred strand-specific associations.