Predictive modelling of gene expression from transcriptional regulatory elements

Area of Science:

• Computational Biology
• Genomics
• Systems Biology

Background:

• Predictive modeling of gene expression is crucial for understanding transcriptional regulation.
• Existing models identify gene/miRNA dysregulation linked to transcription factor (TF) binding and histone modifications (HMs).
• A comparative analysis of modeling algorithms and feature extraction methods is currently lacking.

Purpose of the Study:

• To define and compare methods for quantifying gene-TF/HM interactions.
• To evaluate the suitability of different methods for integrating heterogeneous ChIP-seq data.
• To construct and compare log-linear and epsilon-support vector regression models for predicting gene expression.

Main Methods:

• Quantification of pairwise gene-TF/HM interactions using three distinct methods.
• Development of log-linear and epsilon-support vector regression models.
• Utilized mouse embryonic stem cell (mESC) and human lymphoblastoid (GM12878) datasets, incorporating both ChIP-seq and in silico TF-binding data (PWM).

Main Results:

• Transcription factor binding and histone modifications are highly predictive of mRNA transcript abundance.
• Predictive accuracy was consistent across different algorithms and cell types.
• Both ChIP-seq and position weight matrix (PWM)-derived TF-binding data yielded reliable predictions.

Conclusions:

• The study provides a comparative framework for predictive modeling of gene expression.
• TF-binding and HMs are key determinants of gene expression levels.
• The developed framework is open-source and available as a virtual environment for broader research application.