Joint model-based distance embedding of multi-track Hi-C data for chromosomal conformation learning

Motivated by the problem of reconstructing chromatin conformation from multi-track Hi-C data, we develop a data-integration method named Joint Model-based Distance Embedding (JMDE). JMDE enables probabilistic modeling for data from multiple sources, and learns the underlying shared Euclidean distance embedding in a unified framework. The practical merits of JMDE is demonstrated by simulations and real applications for reconstructing chromatin conformations of human chromosomes 14 and 22 in human lymphoblastoid cell line using two tracks of Hi-C data where the assays were performed with two restriction enzymes HindIII and NcoI, respectively. The proposed JMDE method can be applied to other fields to learn low-dimensional manifold latent structures from multiple related high-dimensional data where pairwise distances are not directly observed.

Keywords

manifold learning, data integration, chromatin conformation, distance embedding

Full Text (PDF format)

Received 26 November 2022

Published 19 July 2024