Anchor

Anchor predicts transcription factor (TF) binding sites across diverse cell types by training machine learning models on integrated epigenomic datasets to extend binding-site inference to untested cellular conditions.

Key Features:

• Multisource Bias Correction: Applies methods to correct biases arising from multiple data sources and batch effects in genomic datasets.
• Information Imbalance Handling: Manages uneven data coverage across cell types to support predictions when data are sparse or unevenly distributed.
• Nonlinear Interaction Modeling: Models complex nonlinear interactions between TF binding motifs and chromatin accessibility extending up to 1500 base pairs from the region of interest.
• Generalization Across Cell Types: Trains predictive models that generalize TF binding-site predictions to untested cell types beyond DNA sequence and DNase-seq footprints.

Scientific Applications:

• Transcription Factor Binding Site Prediction: Predicts TF binding sites in untested cell types to inform gene regulatory mechanism studies across cellular contexts.
• Epigenomic Research: Correlates TF binding with chromatin accessibility and other regulatory elements to study epigenetic landscapes.
• Systems Biology: Supplies binding-site predictions for integration into models of cellular function and regulatory interaction networks.

Methodology:

Anchor employs a machine learning pipeline that integrates ChIP-seq, DNase-seq, and other epigenomic datasets, implements multisource bias mitigation and strategies for information imbalance, trains models capturing nonlinear and indirect relationships between TF motifs and chromatin accessibility up to 1500 bp while accounting for cellular context, and was validated and benchmarked in the 2017 ENCODE-DREAM in vivo TF binding site prediction challenge.