CASPIAN

CASPIAN identifies chromatin topologically associated domain (TAD) boundaries from Hi-C contact matrices using hierarchical spatial density clustering to enable analysis of boundary-associated regulatory factors and their relationship to gene expression.

Key Features:

• TAD boundary detection: Identifies chromatin TAD boundaries from Hi-C contact matrices across varying resolutions.
• Hierarchical density-based clustering (HDBSCAN): Implements the HDBSCAN algorithm as the core spatial density clustering method for boundary identification.
• Multiple distance metrics: Computes pairwise bin distances using Euclidean, Manhattan, and Chebyshev metrics to accommodate diverse Hi-C contact matrix characteristics from simulation or normalized methods.
• Minimal parameterization: Operates with a minimal set of parameters for the clustering procedure.
• Regulatory factor enrichment: Detects enrichment of regulatory elements and proteins at TAD boundaries, including CTCF, H3K4me1, H3K4me3, RAD21, POLR2A, and SMC3, with enrichment patterns comparable to Insulation Score and TopDom.
• Boundary-centered distribution analysis: Analyzes the spatial distribution of various factors anchored at TAD boundaries to characterize their positional patterns relative to boundaries.

Scientific Applications:

• TAD mapping: Produces boundary calls for analyses of chromatin topology from Hi-C datasets.
• Regulatory element analysis: Enables enrichment and positional studies of CTCF, H3K4me1, H3K4me3, RAD21, POLR2A, and SMC3 at TAD boundaries to investigate regulatory roles.
• Method comparison: Facilitates comparative assessment of boundary-associated factor enrichment relative to Insulation Score and TopDom.
• Chromatin structure–function studies: Supports investigation of relationships between TAD boundaries and gene expression patterns.

Methodology:

CASPIAN applies a spatial density clustering approach using HDBSCAN and computes pairwise bin distances with Euclidean, Manhattan, and Chebyshev metrics on Hi-C contact matrices (including simulated or normalized matrices).