Hi-ChiP-ML

Hi-ChIP-ML predicts chromatin folding patterns and Topologically Associating Domains (TADs) in Drosophila from epigenetic datasets including chromatin immunoprecipitation profiles, DNA-binding protein distributions, and spatial DNA structure data.

Key Features:

• Machine Learning Models: Implements linear regression with four types of regularization, gradient boosting methods, and recurrent neural networks using a bidirectional long short-term memory (BiLSTM) architecture.
• Epigenetic Input: Uses chromatin immunoprecipitation (ChIP) data and distributions of DNA-binding proteins as predictive features.
• Specific Predictive Markers: Identifies protein Chromator (Chriz) and histone modification H3K4me3 as particularly informative for TAD prediction.
• Organism and Target: Focused on Drosophila and the prediction and characterization of Topologically Associating Domains (TADs).
• Data Integration: Integrates large-scale epigenetic datasets that include spatial DNA structure information.
• Cross-cell-line Analysis: Analyzes chromatin marks across different cell lines to inform predictions.
• Performance: The BiLSTM architecture produced superior prediction scores compared to other tested models.

Scientific Applications:

• Gene Regulation Studies: Enables investigation of how TAD organization relates to regulation of gene expression in Drosophila.
• TAD Formation Mechanisms: Provides insights into the epigenetic features and mechanisms underlying TAD formation and genomic organization.
• Feature Prioritization: Supports identification of key epigenetic predictors (e.g., Chromator, H3K4me3) for downstream functional studies of gene regulation.

Methodology:

Analyzes chromatin marks across different cell lines and applies machine learning models—linear regression with regularization, gradient boosting, and BiLSTM recurrent neural networks—to characterize DNA folding patterns associated with TADs.