Benchmarking and building DNA binding affinity models using allele-specific and allele-agnostic transcription factor...

Area of Science:

• Genomics
• Bioinformatics
• Molecular Biology

Background:

• Transcription factors (TFs) exhibit sequence-specific DNA binding.
• Allele-specific binding (ASB) differences in TFs are observable at heterozygous loci.
• Current genome-scale assays like ChIP-seq have limitations in detecting ASB due to read coverage and variant representation.

Purpose of the Study:

• To develop and benchmark methods for predicting TF binding allelic imbalances from sequence.
• To quantitatively assess the reliability of predicting allelic differences in TF binding.
• To facilitate de novo inference of high-quality TF binding models from in vivo data.

Main Methods:

• Proposed methods for benchmarking sequence-to-affinity models using a likelihood function based on an over-dispersed binomial distribution.
• Introduced PyProBound, an extensible reimplementation of a biophysically interpretable machine learning framework for de novo model inference.
• Incorporated assay-specific bias in DNA fragmentation rate when training models on ChIP-seq data.

Main Results:

• Developed a method to aggregate evidence for allelic preference across the genome without requiring individual variant significance.
• PyProBound facilitates de novo motif discovery using allele-specific ChIP-seq counts.
• Accounting for DNA fragmentation bias improved TF binding models trained on ChIP-seq data.

Conclusions:

• Provided new strategies for predicting the functional impact of non-coding variants.
• Enabled more accurate assessment of TF binding variations at the allelic level.
• Enhanced the prediction of TF binding from sequence data.