Modeling the 4D Genome

Modeling the 4D genome with deep learning and stochastic simulations.


Recent studies have revealed the importance of 3D chromatin structure in the regulation of vital biological processes. Our team has devised different deep learning models to predict 3D chromatin structure from chromatin conformation capture (Hi-C) experiments:

  • In REACH-3D we employ LSTM-based autoencoders to model ensembles of 3D structures and their dynamics over time.
  • In TECH-3D we explore the idea of transfer learning combined with creative 3D genome structure simulations to reconstruct chromatin structure.

See open-source implementations of our models on GitHub.


Our team is interested in understanding how 3D genome structure and nuclear architecture affects vital biological processes. We develop in silico spatiotemporal models that simulate DNA and protein interactions within the 3D cell nucleus and capture the complex stochastic hybrid dynamics that govern these processes. Using this approach, we were able to realistically simulate DNA replication across a full fission yeast genome (see animation), elucidating mechanisms governing this vital cellular process.

Access the code on GitHub.





Related projects


AI Methods for Precision Therapies

Developing AI methods for heterogeneity-aware precision therapies.

Modeling Spatial Heterogeneity of the Tumor Microenvironment

Combining spatial single-cell omics with AI to model the complexity of the tumor micorenvironment and enable novel spatial biomarker discovery.

Quantifying Biological Heterogeneity from Single-cell Data

Understanding, modeling and quantifying different sources of heterogeneity from single-cell measurements.

AI for Single-cell Research

Understanding spatiotemporal heterogeneity across different scales of biological organization.