Understanding the 3D organization of DNA inside a cell’s nucleus is one of the grand challenges of modern genomics. Every human cell contains about two meters of DNA, folded into a compact nucleus in a way that deeply influences gene regulation, epigenetics, and disease mechanisms. Yet, most of what we know about genome structure comes from population-averaged data, obscuring the dynamic architecture that varies from cell to cell.
A new method, Tensor-FLAMINGO, now brings us closer than ever to visualizing the unique 3D genome of individual cells—at an unprecedented 10 kilobase resolution.
The Problem: Sparse and Noisy Data
Single-cell 3D genomics technologies like single-cell Hi-C and Dip-C offer a tantalizing glimpse into individual genome folding. But these methods suffer from extreme sparsity—over 99.9% of the possible contact points between genome regions are missing in each cell’s data. Reconstructing the full 3D structure from this limited information has been a persistent obstacle.
Previous methods often relied on strong biophysical assumptions or were restricted to coarse resolution reconstructions, limiting their accuracy and biological utility.
The Breakthrough: Low-Rank Tensor Completion
Tensor-FLAMINGO uses a data-driven approach based on low-rank tensor completion, an advanced technique originally developed for video compression. Each single-cell contact map is treated like a video frame; the full dataset becomes a multi-dimensional tensor. The algorithm infers the missing data by recognizing shared structural patterns across cells and within individual contact maps.
By combining this with the FLAMINGO framework—a method previously validated for reconstructing genome structures—Tensor-FLAMINGO can generate accurate 3D genome models for individual cells at high resolution.
What It Reveals: Chromatin Hubs, Long-Range Interactions, and Disease Links
The impact of this method is profound:
-
Cell-type-specific 3D structures: Tensor-FLAMINGO distinguishes cells of different types based purely on their 3D genome shapes.
-
Long-range regulatory links: It identifies spatial interactions between distant genomic regions, many of which are missed by bulk sequencing methods.
-
Functional insight into disease variants: The method shows that some mutations linked to leukemia (TCGA-LAML) reside in regions involved in unique single-cell chromatin loops—offering clues to their mechanism.
-
Multi-way interactions: Beyond pairwise loops, Tensor-FLAMINGO maps complex spatial “hubs” involving three or more regions, revealing higher-order genome architecture.
-
Predicting gene regulation: A LASSO-based model trained on the 3D data and DNA methylation identifies functional regulatory elements in single cells, outperforming traditional correlation-based methods.
Why It Matters
This research marks a leap forward in single-cell genomics. By treating high-resolution 3D genome architecture as a data recovery problem rather than a purely biophysical one, Tensor-FLAMINGO offers a scalable and accurate method to understand the molecular machinery of life—cell by cell.
As single-cell technologies continue to evolve, tools like Tensor-FLAMINGO will be essential in decoding how our genome’s spatial organization contributes to development, differentiation, and disease.
Source: Hao Wang, Jiaxin Yang, Xinrui Yu, Yu Zhang, Jianliang Qian & Jianrong Wang. Tensor-FLAMINGO unravels the complexity of single-cell spatial architectures of genomes at high-resolution, Nature Communications (2025) https://doi.org/10.1038/s41467-025-58674-w