DNA sequence+shape kernel enables alignment-free modeling of transcription factor binding

Area of Science:

• Computational Biology
• Genomics
• Bioinformatics

Background:

• Transcription factors (TFs) bind specific DNA motifs, influenced by local DNA shape properties like minor groove width.
• Existing TF-DNA binding prediction methods often require aligned TF binding site training data.
• Understanding sequence and shape contributions is crucial for accurate binding affinity prediction.

Purpose of the Study:

• To develop and evaluate a novel sequence + shape kernel for predicting protein-DNA binding affinity.
• To assess the performance improvement offered by incorporating DNA shape information into TF binding prediction models.
• To provide a computational tool that leverages both sequence and shape for enhanced binding analysis.

Main Methods:

• Developed a sequence + shape kernel extending k-mer based sequence kernels, specifically the di-mismatch kernel.
• Utilized three in vitro benchmark datasets: universal protein binding microarrays (uPBMs), genomic context PBMs (gcPBMs), and SELEX-seq data.
• Compared performance of sequence-only models (k-spectrum, di-mismatch) against combined sequence and shape models.

Main Results:

• The k-spectrum + shape model outperformed the classical k-spectrum kernel, especially for small k values.
• The di-mismatch kernel showed better performance than the k-mer kernel for larger k.
• The di-mismatch + shape kernel demonstrated superior predictive ability compared to the di-mismatch kernel for intermediate k values.

Conclusions:

• Integrating DNA shape information significantly enhances the prediction of protein-DNA binding affinity.
• The developed sequence + shape kernel offers improved accuracy over traditional sequence-based methods.
• This approach provides a more comprehensive understanding of the factors governing protein-DNA interactions.