NEWS
中文Research Highlights
- Publish Date：2025-07-09
Untangling the Epigenomic Universe: NYCU Uses AI to Reconstruct the 3D World Inside Our Cells
AI-generated illustration of the “yarn-ball universe” of chromatin. EpiVerse blends “Epi” (epigenome) and “Verse” (metaverse) to represent an AI-created virtual epigenomic space.
Edited by Chance Lai
______
______
Imagine cramming two meters of yarn into a space just 5–10 microns wide. That’s the extraordinary feat our DNA accomplishes—folding itself into a dense, dynamic structure inside the nucleus of every cell. Now, researchers from the Department of Computer Science at National Yang Ming Chiao Tung University (NYCU) have developed a groundbreaking AI-powered tool to decode this mysterious “yarn-ball universe.”
Named EpiVerse, the new platform offers scientists an entirely new lens to explore how our genome is organized—and how that organization influences health and disease. The study “Unveiling chromatin dynamics with virtual epigenome” was recently published in the prestigious journal Nature Communications.
From Experiment-Heavy to AI-Driven Biology
“Traditionally, exploring chromatin structure required months of complex and costly experiments,” said Professor Jui-Hung Hung, lead researcher and faculty member in NYCU’s Department of Computer Science. “EpiVerse can simulate 3D chromatin folding in different cell states using only computational models. This saves time and resources and opens new paths for understanding gene regulation.”
EpiVerse enables the analysis of the entire human chromatin structure. Shown here is a dendrogram illustrating the hierarchical clustering of 39 different tissue types based on similarities in their Hi-C features, revealing potential functional or developmental relationships.
A New AI Frontier for Life Sciences
EpiVerse harnesses deep learning and virtual reconstruction to model chromatin dynamics across various tissues and cell types, even in sparse experimental data. At its core, the system combines HiConformer multi-task learning and MIRNet multi-scale image reconstruction, enabling more accurate, high-resolution simulations.
Named EpiVerse, the new platform offers scientists an entirely new lens to explore how our genome is organized—and how that organization influences health and disease. The study “Unveiling chromatin dynamics with virtual epigenome” was recently published in the prestigious journal Nature Communications.
From Experiment-Heavy to AI-Driven Biology
“Traditionally, exploring chromatin structure required months of complex and costly experiments,” said Professor Jui-Hung Hung, lead researcher and faculty member in NYCU’s Department of Computer Science. “EpiVerse can simulate 3D chromatin folding in different cell states using only computational models. This saves time and resources and opens new paths for understanding gene regulation.”
EpiVerse enables the analysis of the entire human chromatin structure. Shown here is a dendrogram illustrating the hierarchical clustering of 39 different tissue types based on similarities in their Hi-C features, revealing potential functional or developmental relationships.
A New AI Frontier for Life Sciences
EpiVerse harnesses deep learning and virtual reconstruction to model chromatin dynamics across various tissues and cell types, even in sparse experimental data. At its core, the system combines HiConformer multi-task learning and MIRNet multi-scale image reconstruction, enabling more accurate, high-resolution simulations.
But EpiVerse goes beyond static models. One of its most powerful features is its ability to conduct in silico perturbation experiments, predicting how chromatin structures might shift in response to environmental stimuli, genetic mutations, drug treatments, or disease states like cancer. This capability allows scientists to map regulatory gene networks and explore potential therapeutic targets with unprecedented speed and scale.
Open-Source, Open Science
“Before EpiVerse, running a single perturbation experiment could take months and cost millions,” Prof. Hung explained. “Now, researchers can test hypotheses rapidly, iterate with flexibility, and design smarter follow-up experiments. This is AI’s true potential in transforming life sciences.”
The complete EpiVerse codebase is open-sourced and freely available, providing a powerful new toolset for scientists worldwide studying the epigenome and chromatin structure.
Prof. Hung led this pioneering work jointly developed by two graduate students, Yu-Cheng Lo and Ming-Yu Lin, from the Institute of Data Science and Engineering. The project highlights NYCU’s leadership in training next-generation talent at the intersection of AI, bioinformatics, and biomedical science.
