TAPIOCA

TAPIOCA predicts topologically associated domains (TADs) from epigenetic features to infer chromatin topology for studies of gene regulation and genomic organization.

Key Features:

• TAD prediction from epigenetic features: Predicts topologically associated domains (TADs) using epigenetic track data as input features.
• Self-attention transformer architecture: Implements a self-attention-based deep learning transformer algorithm inspired by the sequence transduction transformer network architecture.
• Hi-C-independent inference: Infers chromatin topology without requiring labeled Hi-C sequencing data as input.
• Cross-cell-line generalization: Demonstrates generalization of TAD prediction performance across cell lines beyond the training set.
• Quantitative performance: Reports superior performance on established metrics for TAD prediction compared to prior methods.

Scientific Applications:

• Chromatin topology mapping: Use predicted TADs to characterize chromatin organization and structural domains in the genome.
• Gene regulation studies: Relate TAD boundaries and topology to regulation of gene expression and maintenance of genomic integrity.
• Comparative cell-line analysis: Compare chromatin architecture across different cell lines using epigenetic-feature-based TAD predictions.
• Analysis without Hi-C data: Enable TAD inference and downstream analyses in contexts where Hi-C sequencing data are unavailable.

Methodology:

Uses epigenetic track data as model input and applies a self-attention-based deep learning transformer algorithm inspired by the sequence transduction transformer network architecture to predict TADs.