Junker and his team seek to reconstruct gene networks in normal embryonic development of zebrafish, a model organism for vertebrates, including humans.

Helmholtz AI project call showcase: From Single Cell Multiomics to Gene Regulatory Networks

Using machine learning approaches to better understand gene regulatory networks.

This week in the Helmholtz AI project showcase, we present the collaborative work from the Max Delbrück Center for Molecular Medicine and Helmholtz Center Munich. Titled "From single-cell multi-omics to gene regulatory networks – a machine learning approach," the project is a look at how cells make decisions.

Seen above: Reconstruction of gene regulatory networks. A. We combine to single-cell datatypes: Identification of potential regulatory regions and transcription factor (TF) binding sites by single-cell ATAC-seq, and about previously existing versus new mRNA molecules in single cells by single-cell SLAM-seq. B. Our goal is identify regulatory interactions with high accuracy by making use of the temporal information in scSLAM-seq: If a TF regulates a target gene, the TF should be transcribed earlier than the target.

Could you introduce yourself, giving your affiliation, area of work, and of course, the project title?

My name is Jan Philipp Junker. I'm a group leader at the Max Delbrück Center for Molecular Medicine, in the Berlin Institute for Medical Systems Biology. My lab combines experimental and computation work to understand how cells make decisions in health and disease. We mostly use zebrafish as a model system, and we develop and use methods in single-cell genomics to address our questions. Our joint Helmholtz AI project with Maria Colomé Tatché (Helmholtz Center Munich) is called SC-SLAM-ATAC, "From single-cell multi-omics to gene regulatory networks – a machine learning approach".

In simple words, what specifically is your project about? And, how and why do you think it is a high risk, high gain endeavour?

All cells in our body have almost exactly the same genome, but these cells still differ vastly in their appearance and function and can be grouped into different cell types, for instance neurons or muscle cells. Which fate a cell takes depends on the genes that are active in that particular cell, and this is in turn controlled by regulatory regions in the genome. Understanding how gene expression is regulated is a very active field of research. Here, we want to combine two data types, single-cell open chromatin profiling and single-cell RNA labeling, using machine learning approaches, in order to improve reconstruction of gene regulatory networks. If successful, this work may help us better understand how cell identity is misregulated in disease. An important challenge is that in vivo single-cell RNA labeling is a very new technique, and that we still need to understand this data type better. 

How important has the Helmholtz AI funding and platform been to carry out this project?

The Helmholtz AI funding is absolutely crucial for this project, since it allows us to build a computational infrastructure for a new data type and thereby establish a method that combines experimental and computational work. It would be difficult to find other funding sources for such interdisciplinary and exploratory method development. Importantly, the approaches developed here will be an excellent basis for future projects focused on biomedical applications of the technique.

Any other comments you wish to add? 

I particularly enjoy the collaborative aspect in this project, which will allow us to gain insights that my lab alone would not be able to achieve.