Insulation Score

The insulation score is computed by sliding a square window along the diagonal of a Hi-C contact matrix. For each genomic bin, the algorithm sums all interaction frequencies within that window, quantifying the local contact density. A sharp drop in the insulation score indicates a boundary where interactions are depleted, signifying a transition between two self-interacting domains. The window size, typically ranging from 100kb to 1Mb, determines the scale of detected boundaries.

TAD boundaries are identified as local minima in the insulation score profile. A deep trough represents a genomic locus with significantly fewer contacts than its flanking regions, acting as an insulating barrier to loop extrusion. The strength of a boundary is quantified by the delta vector, which measures the difference in insulation between the boundary and its neighboring regions. Strong boundaries often correlate with convergent CTCF binding sites.

Raw Hi-C contact frequencies exhibit a strong distance-dependent decay, where proximal loci interact more frequently than distal ones. Insulation scores must be normalized to account for this genomic distance bias. Without normalization, the center of large domains would artifactually appear as boundaries. Common approaches include dividing by the genome-wide average contact probability at each distance or using iterative correction methods like ICE.

Genome folding is hierarchical, with nested sub-TADs existing within larger domains. Computing insulation scores at multiple window sizes reveals this structural hierarchy. A small window detects fine-scale boundaries, while a large window identifies mega-domain borders. This multi-scale approach is critical for understanding how regulatory elements are partitioned at different levels of chromatin organization.

Deep learning models like Akita and DeepC predict insulation scores directly from DNA sequence and epigenomic features. By framing 3D genome folding as a sequence-to-insulation regression task, these models bypass the need for experimental Hi-C data. The predicted insulation profile can then be used to infer TAD boundaries, A/B compartments, and chromatin loops in silico for any cell type or genetic variant.

Not all boundaries are equal. The insulation score depth quantifies boundary strength, which correlates with the occupancy of architectural proteins like CTCF and cohesin. Dynamic boundaries that change across cell types or developmental stages exhibit variable insulation scores. Tracking these quantitative shifts reveals how 3D genome reorganization drives differential gene expression during differentiation and disease progression.

A complementary metric that quantifies the upstream or downstream bias of chromatin interactions at a given genomic bin.

• Computed from Hi-C contact maps to infer the directionality of loop extrusion
• Positive values indicate downstream-biased interactions; negative values indicate upstream bias
• Used alongside insulation score to precisely map TAD boundaries and their orientation
• Often visualized as a track aligned with insulation profiles for boundary confirmation

A self-interacting genomic region where DNA sequences physically interact more frequently with each other than with sequences outside the domain.

• Insulation score minima directly identify TAD boundaries
• TADs function as fundamental structural units of chromosome folding
• Boundaries are enriched for CTCF binding sites and housekeeping genes
• Disruption of TAD boundaries is associated with developmental disorders and cancer

A reproducibility metric specifically designed for Hi-C data that measures similarity between two contact maps while accounting for distance-dependent signal.

• Stratifies contacts by genomic distance before computing correlation
• Prevents the dominant distance-decay signal from inflating similarity scores
• Used to benchmark insulation score reproducibility across replicates
• Essential for validating predicted contact maps against experimental data

A statistical correction applied to Hi-C contact maps to account for the expected background contact frequency decay as a function of linear genomic distance.

• Contact probability follows a power-law decay: P(s) ~ s^(-1) at megabase scales
• Normalization enables accurate insulation score computation by removing distance-dependent bias
• Methods include iterative correction (ICE) and matrix balancing
• Without normalization, insulation scores would reflect genomic distance rather than true boundary strength

A mechanistic model wherein cohesin complexes actively reel DNA to form progressively larger loops until blocked by CTCF boundary elements.

• Explains the formation of TADs and the insulation patterns observed in Hi-C data
• Insulation score quantifies the barrier strength that halts extrusion
• Convergent CTCF motifs create strong boundaries with low insulation scores
• Provides the biophysical foundation for interpreting insulation score profiles

The classification of genomic regions into open, active 'A' compartments or closed, inactive 'B' compartments based on long-range interaction patterns.

• Derived from the first principal component of Hi-C correlation matrices
• Operates at a larger scale than TADs and insulation score domains
• A compartments correlate with active transcription and high gene density
• Insulation score boundaries can exist within both A and B compartments, reflecting hierarchical genome organization