Virtual ChIP-seq: Datasets for Predicting Transcription Factor Binding

Datasets for the Virtual ChIP-seq software contain reference matrices, trained models, and pre-calculated features for predicting transcription factor binding. The data includes correlation matrices for TFs A-Z, genomic conservation scores, PWM scores from JASPAR, and ChIP-seq data from ENCODE and Cistrome. This repository was created by Mehran Karimzadeh of the University of Toronto.

Use Cases

• Predicting transcription factor binding based on correlation matrices with gene expression.
• Training or retraining machine learning models using pre-calculated features like PhastCons conservation and FIMO scores.
• Analyzing the relationship between chromatin accessibility and TF binding using ENCODE and Cistrome data.
• Benchmarking new algorithms against the provided scikit-learn trained models.

Strengths

• Includes reference data for transcription factors spanning letters A to Z, covering a wide range.
• Provides pre-calculated predictive features for four specific chromosomes (5, 10, 15, 20).
• Integrates data from multiple established sources, including PhastCons, JASPAR, ENCODE, and Cistrome.

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• Row count is unknown, which may limit suitability assessment.
• Last update date is unknown; freshness unverified.

Provenance

Source

University of Toronto (Mehran Karimzadeh), paperswithcode platform.

Collection Method

Likely compiled from public genomic databases (ENCODE, Cistrome) and computational predictions.