HICTDA: A Persistent Homology Framework for Multiscale Identification of Topologically Associating Domains (TADS) in Hi-C Data

Background High-throughput chromosome conformation capture (Hi-C) provides genome-wide measurements of chromatin interactions and enables characterization of three-dimensional genome organization. However, identifying and quantifying Topologically Associating Domains (TADs) from Hi-C contact maps remains challenging because contact matrices are inherently noisy and sensitive to resolution, genomic window, analytical scale, and methodological choices. Methods This study presents HICTDA, an open-source computational framework for the multiscale topological analysis of Hi-C data and a possible identification of TAD-associated structural features using persistent homology. Hi-C data from the GM12878 lymphoblastoid cell line were analyzed for human chromosomes 3 and 4. Normalized contact matrices were transformed into distance-like representations, followed by construction of Vietoris–Rips filtrations across multiple spatial scales. Eigenvector decomposition of the normalized matrices was used to characterize A/B chromatin compartments, while persistent homology was applied to quantify H 0 and H 1 topological features across filtration scales. Persistence diagrams and barcodes were used to represent and track these topological features. Results Multiscale persistent homology demonstrated a strong dependence of TAD-associated topological features on genomic window size. At 50-kb resolution, 271 H 1 features were detected across the 50-Mb chr4 region, decreasing to 65, 15, 2, and 1 across progressively smaller windows, with no features detected in the 2-Mb region. This progressive loss was consistent with a boundary-truncation effect, whereby insufficient genomic context may disrupt the topological representation of complete TAD structures. At the 50-Mb scale, chr4 also exhibited greater H 1 features abundance than chr3. Conclusions HICTDA provides an open-source framework for basic statistical evaluation and quantifying multiscale topological features of Hi-C data that can serve as inputs for machine-learning analysis. The correspondence of H 1 features to TADs and their separation from noise remain to be established and benchmarked against known algorithms.