CCIP

CCIP predicts chromatin loops mediated by CTCF (CCCTC-binding factor) using a two-stage random-forest machine-learning approach that leverages transitivity-related information together with genomic and functional genome data to study loop formation, topologically associating domains (TADs), and enhancer–promoter interactions.

Key Features:

• CTCF-mediated loop prediction: Predicts chromatin loops mediated by CTCF (CCCTC-binding factor).
• Two-stage random-forest model: Implements a novel two-stage random-forest-based machine-learning approach.
• Transitivity-related information: Incorporates transitivity-related information of interacting CTCF anchors, where transitivity denotes anchors interacting via a common third anchor through loop extrusion.
• Feature integration: Integrates functional genome data and genomic data as predictive features.
• Stage 1 — transitivity-based candidate identification: Identifies potential interactions based on transitivity in the first stage.
• Stage 2 — prediction refinement: Refines initial predictions using additional genomic features in the second stage.
• Tandem loop and enhancer–promoter inference: Captures tandem CTCF loops and facilitates inference about enhancer–promoter interactions.
• Performance validation: Experimental validation reported higher accuracy compared to existing methods.
• Insights into TADs and regulation: Provides insights into formation of topologically associating domains (TADs) and implications for transcriptional regulation.

Scientific Applications:

• Chromatin architecture mapping: Predicts CTCF-mediated chromatin loops to study chromatin organization and loop networks.
• Mechanistic studies: Investigates mechanisms of loop formation and loop extrusion involving CTCF anchors.
• TAD and regulatory analysis: Analyzes the role of CTCF-mediated loops in TAD formation and transcriptional regulation.
• Enhancer–promoter interaction inference: Infers enhancer–promoter interactions mediated or facilitated by CTCF loops.
• Method benchmarking: Serves to compare predictive performance against existing loop-prediction methods.

Methodology:

Two-stage random-forest-based machine learning integrating transitivity-related information with functional genome data and genomic data, where stage 1 identifies potential interactions based on transitivity and stage 2 refines predictions using additional genomic features.