Study Benchmarks 13 Single-Cell Hi-C Embedding Tools

DNA forms a complex three-dimensional structure inside every cell, influencing how genes function and interact. To study this intricate architecture, scientists require advanced computational tools capable of decoding not only DNA’s sequence but also how it folds and interacts inside individual cells. A research team at Case Western Reserve University has now compared single-cell Hi-C (scHi-C) embedding tools to determine which perform best under varying conditions. 

Led by Fulai Jin, the team evaluated 13 embedding tools that analyze DNA’s 3D organization using 10 datasets from mouse and human cells. The study, published in Nature Communications, revealed that results from these tools often differ, much like translators offering conflicting interpretations of the same text. The researchers found that each program has strengths and limitations depending on the data type, and that pre-processing methods—how data are prepared before analysis—can strongly affect accuracy. Artificial intelligence approaches proved especially adept at handling lower-quality or complex data that other methods struggled to interpret. 

“The 3D structure of DNA affects how genes interact with each other, just like the layout of a house affects how people move through it,” Jin explained. “Understanding this structure is crucial for figuring out how diseases develop and how we might treat them.”

The team likened their work to designing better microscopes for visualizing DNA activity inside cells. By refining and benchmarking computational methods, they hope to help researchers discern which genes switch on or off in disease contexts, why treatments vary in effectiveness between patients, and how cells transition during development.

To make their findings actionable, the researchers created a software package that identifies the most effective analysis approach for a particular dataset—similar to how a GPS application finds an optimal route. “Instead of researchers having to guess which tool might work best, our software can test multiple approaches and recommend the optimal one,” Jin said. The tools are freely available through GitHub, where scientists worldwide can access, test, and adapt them.