ChIPXpress: using publicly available gene expression data to improve ChIP-seq and ChIP-chip target gene ranking

ChIPx (i.e., ChIP-seq and ChIP-chip) is increasingly used to map genome-wide transcription factor (TF) binding sites. A single ChIPx experiment can identify thousands of TF bound genes, but typically only a fraction of these genes are functional targets that respond transcriptionally to perturbations of TF expression. To identify promising functional target genes for follow-up studies, researchers usually collect gene expression data from TF perturbation experiments to determine which of the TF targets respond transcriptionally to binding. Unfortunately, approximately 40% of ChIPx studies do not have accompanying gene expression data from TF perturbation experiments. For these studies, genes are often prioritized solely based on the binding strengths of ChIPx signals in order to choose follow-up candidates. ChIPXpress is a novel method that improves upon this ChIPx-only ranking approach by integrating ChIPx data with large amounts of Publicly available gene Expression Data (PED).

We demonstrate that PED does contain useful information to identify functional TF target genes despite its inherent heterogeneity. A truncated absolute correlation measure is developed to better capture the regulatory relationships between TFs and their target genes in PED. By integrating the information from ChIPx and PED, ChIPXpress can significantly increase the chance of finding functional target genes responsive to TF perturbation among the top ranked genes. ChIPXpress is implemented as an easy-to-use R/Bioconductor package. We evaluate ChIPXpress using 10 different ChIPx datasets in mouse and human and find that ChIPXpress rankings are more accurate than rankings based solely on ChIPx data and may result in substantial improvement in prediction accuracy, irrespective of which peak calling algorithm is used to analyze the ChIPx data.

ChIPXpress provides a new tool to better prioritize TF bound genes from ChIPx experiments for follow-up studies when investigators do not have their own gene expression data. It demonstrates that the regulatory information from PED can be used to boost ChIPx data analyses. It also represents an important step towards more fully utilizing the valuable, but highly heterogeneous data contained in public gene expression databases.