PICS

PICS infers transcription factor binding events from ChIP-Seq directional short-read data using probabilistic Bayesian models to localize binding sites and quantify enrichment.

Key Features:

• Probabilistic inference: Employs an empirical Bayes mixture model to model local concentrations of directional reads for robust identification of binding sites.
• Bayesian hierarchical t-mixture model: Uses a Bayesian hierarchical t-mixture model to distinguish closely adjacent binding events.
• DNA fragment length priors: Incorporates prior information on DNA fragment length to improve binding event localization.
• Read mappability adjustment: Leverages precalculated whole-genome read mappability profiles and a truncated t-distribution to account for reads missing due to local genome repetitiveness.
• Uncertainty estimation: Estimates uncertainties in model parameters and provides confidence regions on binding event locations.
• Enrichment scoring and FDR estimation: Calculates per-event enrichment scores relative to control samples and uses controls to estimate false discovery rates.

Scientific Applications:

• Genome-wide TF binding site identification: Identifies transcription factor–DNA associations genome-wide from ChIP-Seq data.
• High-resolution localization: Precisely localizes closely adjacent binding events using hierarchical t-mixture modeling and fragment length priors.
• Mappability-aware analysis: Enables analysis that accounts for local repetitiveness and mappability biases using whole-genome mappability profiles.
• Enrichment quantification and validation: Provides per-site enrichment scoring and control-based FDR estimates for validation of identified sites.
• Robustness assessment: Supports analyses requiring robustness to model misspecification as demonstrated in simulation studies.

Methodology:

Models directional read distributions using an empirical Bayes mixture framework and a Bayesian hierarchical t-mixture model; incorporates DNA fragment length priors; adjusts for missing reads using precalculated whole-genome read mappability profiles and a truncated t-distribution; estimates parameter uncertainties and computes per-event enrichment scores with control-based FDR estimation.