Datasets for predicting TF binding using Virtual ChIP-seq

This repository contains datasets necessary for using the Virtual ChIP-seq software.

Virtual ChIP-seq requires the following datasets to predict transcription factor binding:

• chipExpDir_AtoH_V1.0.0.tar.gz: Reference matrices of correlation between TF binding and gene expression for TFs starting with letters A-H.

• chipExpDir_ItoZ_V1.0.0.tar.gz: Reference matrices of correlation between TF binding and gene expression for TFs starting with letters I-Z.

• refTables_V1.1.0.tar.gz: PhastCons genomic conservation, FIMO PWM scores for JASPAR motifs, and ChIP-seq data of ENCODE and Cistrome database.

• hg38_chrsize.tsv: Length of chromosomes in hg38

• trainedModels_V1.0.0.tar.gz: Virtual ChIP-seq scikit-learn trained models saved in joblib format

• <CellType>.tar.gz: Pre-calculated matrices suitable for training with other algorithms or re-training with Virtual ChIP-seq.

Some predictive features of TF binding are the same in each cell type and are stored together for simplicity in refTables_V1.0.0.tar.gz. You can use datasets from other cell types (named here as  <CellType>.tar.gz) for the purpose of re-training the model. The <CellType>.tar.gz files contain pre-calculated predictive features of transcription factor binding in 4 chromosomes (5, 10, 15, 20).

These features include:

• PhastCons genomic conservation

• FIMO score for sequence motifs of TF in the JASPAR database

• Chromatin accessibility

• TF binding in ENCODE + Cistrome DB datasets

• Virtual ChIP-seq expression score