3D Genome Reconstruction

The core mathematical engine that converts pairwise interaction frequencies into 3D coordinates. Hi-C contact maps are first normalized and transformed into a distance matrix using an inverse power-law relationship (contact frequency ∝ 1/distance^α). The algorithm then iteratively adjusts the spatial positions of genomic loci to minimize the discrepancy between the derived distance matrix and the distances in the reconstructed 3D structure. Common solvers include multidimensional scaling (MDS) and simulated annealing with conjugate gradient descent, often constrained by excluded volume and polymer chain connectivity.

Reconstruction algorithms integrate principles from polymer physics to ensure biologically plausible structures. Key constraints include:

• Excluded volume: No two chromatin segments can occupy the same space
• Chain connectivity: Adjacent loci must maintain covalent bond distances (~1–2 nm)
• Bending rigidity: The persistence length of chromatin (~50–300 nm) limits sharp turns
• Confinement: Chromosomes are constrained within the nuclear volume These priors prevent overfitting to noisy Hi-C data and produce structures consistent with DNA FISH validation measurements.

Hi-C data represents a population average of millions of cells, each with a unique genome conformation. Reconstruction methods must decide between:

• Consensus structures: A single 3D model representing the average folding pattern across the cell population
• Ensemble structures: A distribution of conformations that collectively satisfy the contact probability matrix Ensemble approaches better capture the dynamic and heterogeneous nature of chromatin, revealing transient interactions and cell-to-cell variability that consensus models obscure. Methods like Bayesian inference and molecular dynamics are used to sample the conformational landscape.

Raw Hi-C matrices contain systematic biases that must be corrected before reconstruction to avoid structural artifacts. The Iterative Correction and Eigenvector decomposition (ICE) algorithm applies matrix balancing to equalize row and column sums, correcting for:

• GC content bias: Regions with high GC content are over-represented in sequencing
• Mappability: Repetitive regions have ambiguous read alignment
• Restriction fragment length: Longer fragments generate more ligation products Without proper normalization, reconstructed structures would reflect technical artifacts rather than true chromatin conformation.

Reconstructed structures must be validated against independent experimental methods to confirm accuracy. DNA FISH (Fluorescence In Situ Hybridization) directly measures physical distances between specific loci in individual cells, serving as the gold standard. Key validation metrics include:

• Spearman correlation between predicted and measured pairwise distances
• Distance distribution overlap for locus pairs across the cell population
• Stratum-Adjusted Correlation Coefficient (SCC) for genome-wide contact map similarity Strong concordance with FISH data confirms that the reconstruction algorithm captures genuine 3D organization rather than computational artifacts.

Standard Hi-C analysis collapses maternal and paternal chromosomes into a single signal, obscuring allele-specific folding patterns. Haplotype-resolved reconstruction leverages heterozygous single nucleotide polymorphisms to separate reads by parental origin, enabling independent 3D modeling of each allele. This reveals how genetic variation—including structural variants and single nucleotide polymorphisms—differentially influences chromatin architecture. The approach is critical for understanding imprinting and allele-specific gene regulation in development and disease.