Benchmarking DNA binding affinity models using allele-specific transcription factor binding data.

Area of Science:

• Genomics and Molecular Biology
• Bioinformatics and Computational Biology

Background:

• Transcription factors (TFs) exhibit sequence-specific DNA binding, which can lead to allele-specific binding (ASB) at heterozygous loci.
• Standard ChIP-seq assays often lack the statistical power to detect ASB at the individual variant level across the genome.
• Accurate TF binding models are crucial for understanding gene regulation and its variations.

Approach:

• Developed a framework for benchmarking sequence-to-affinity models by evaluating their ability to predict allelic imbalances in ChIP-seq counts.
• Utilized a likelihood function based on an over-dispersed binomial distribution to aggregate evidence for allelic preference genome-wide.
• Introduced PyProBound, an extensible reimplementation of the ProBound machine learning framework, incorporating sequence-specific bias in DNA fragmentation.

Key Points:

• The proposed likelihood function effectively aggregates evidence for allelic preference without requiring significance at individual variants.
• PyProBound, configured to account for DNA fragmentation bias, improves TF binding models trained on ChIP-seq data.
• The framework facilitates systematic comparison of multiple TF binding models for the same transcription factor.

Conclusions:

• The developed framework provides a robust method for benchmarking TF binding models and enhancing the detection of allele-specific binding.
• The PyProBound implementation offers improved TF binding predictions by accounting for sequence-specific biases.
• The likelihood function can be effectively utilized for de novo motif discovery directly from raw allele-aware ChIP-seq counts.