SACSANN

SACSANN predicts chromosomal A/B compartment annotations from genomic sequence using stacked artificial neural networks to infer cell-type-specific megabase-scale compartmentalization.

Key Features:

• Sequence-based features: Uses genomic sequence-derived features, including predicted transcription factor binding sites and transposable elements.
• Stacked artificial neural networks: Employs stacked artificial neural networks in a hierarchical model to classify A and B chromosomal compartments.
• Megabase-scale, cell-type-specific predictions: Models compartmentalization at the megabase scale and provides cell-type-specific compartment annotations.
• Hi-C-independent prediction: Predicts compartment annotations without requiring Hi-C data.
• Cross-species transferability: Demonstrates transferability of predictions across analogous human and mouse cell types.
• Sequence determinant identification: Identifies key genomic sequence determinants associated with compartmentalization.
• Reference-genome focus: Operates solely on features derived from reference genomes.

Scientific Applications:

• Compartment inference without Hi-C: Enables prediction of A/B compartments in species or cell types lacking Hi-C data.
• 3D genome evolution studies: Facilitates comparative analyses of compartmentalization across species to study 3D genome evolution.
• Gene regulation investigation: Supports analysis of relationships between compartmentalization and transcriptional activity (A compartments versus B compartments).
• Comparative cell-type analyses: Allows model transfer and comparison between analogous human and mouse cell types.

Methodology:

Extracts sequence-derived features (predicted transcription factor binding sites and transposable elements) from reference genomes and trains stacked artificial neural networks in a hierarchical model to predict A/B compartments.