TFAnalysis

TFAnalysis predicts transcription factor binding sites and cell-type-specific interactions using ensemble random forest classifiers to support gene regulation analyses.

Key Features:

• Ensemble random forest models: Employs an ensemble learning strategy using random forest classifiers to predict TF binding sites across tissues and cell types.
• DNase1-seq data integration: Integrates DNase1-seq data and uses DNase1-seq peaks rather than raw signals to reduce false positives in binding predictions.
• Position Specific Energy Matrices (PSEMs): Uses PSEMs as transcription factor motif descriptors for model training and evaluation.
• ChIP-seq data integration: Uses transcription factor ChIP-seq data as gold-standard labels for training and evaluating predictive performance.
• Protein-protein interaction context: Feature-importance analyses reveal selection of motifs from TFs that are close interaction partners in protein-protein interaction networks.

Scientific Applications:

• Gene regulation studies: Predicts TF binding sites to aid elucidation of gene regulation mechanisms across various tissues and cell types.
• ENCODE-DREAM challenge evaluation: Developed and evaluated within the ENCODE-DREAM in vivo TF binding site prediction challenge.

Methodology:

Trains ensemble random forest classifiers using DNase1-seq peaks and PSEMs as input features, evaluates models against ChIP-seq data, and performs feature-importance analysis to identify cell-type-specific co-factors and motifs linked via protein-protein interaction networks.