Quantitative modeling of gene expression using DNA shape features of binding sites

Area of Science:

• Genomic biology
• Transcriptional regulation
• Computational biology

Background:

• Predicting gene expression from regulatory sequences is crucial in genomics.
• Current sequence-to-expression models primarily use position weight matrices (PWMs) for DNA binding specificity.
• The utility of DNA shape features in these models remains largely unexplored.

Purpose of the Study:

• To develop and evaluate a statistical thermodynamics model for gene expression prediction using DNA shape features.
• To compare the performance of DNA shape-based models against traditional PWM-based models.
• To investigate the combined predictive power of DNA shape and PWM features.

Main Methods:

• Developed a statistical thermodynamics model incorporating DNA shape features of transcription factor binding sites.
• Evaluated model performance using expression data from 37 enhancers in Drosophila embryos.
• Compared predictions from DNA shape models, PWM models, and combined models.

Main Results:

• DNA shape-based models demonstrated superior performance compared to PWM-based models in predicting gene expression.
• DNA shape features provide complementary information to PWMs, enhancing predictive accuracy.
• Combining DNA shape and PWM features yielded improved gene expression predictions.

Conclusions:

• DNA shape-based models offer a valuable alternative and complement to PWMs for sequence-to-expression modeling.
• Local DNA shape features are effective for predicting gene expression driven by regulatory sequences.
• This study provides a framework for advancing gene expression prediction beyond current PWM limitations.