Saez-Rodriguez Group

Systems biomedicine

Modeling signaling networks from single cell data

We develop mechanistic models of signaling pathways based on single-cell data. The models are trained against time course data of signaling protein activity. This way cell subpopulation specific networks are identified, which reveal novel regulatory mechanisms and pathway cross-talks.

Signaling pathways, such as the ErbB pathway, link the cell to its environment through a series of protein phosphorylation. These pathways are known to be highly interconnected and regulated via feedback and feed-forward loops, further, cross-talk between pathways can enhance external signal transduction.  This way, external signals are translated to alteration of gene expression levels and let the cell adapt to changing environmental conditions. However, even isogenetic cells response differently to perturbation because of their different state, for example, due to different cell cycles or different signaling state.


Our goals in this project are

(1) To explain how the cell-to-cell variability can affect the variety of cell responses to growth factor stimuli and drug perturbation,

(2) To reveal signaling pathway interactions upon multi growth factor stimulation,

(3) To identify subpopulations of cells and reconstruct their signaling pathways,



Live cell imaging techniques, such as the Fluorescence Resonance Energy Transfer (FRET) based measurements are used (in collaboration with Carsten Schultz lab, at EMBL-HB) to monitor selected signaling proteins of the transduction networks in time on single cell level. Further, multiplexed mass cytometry (in collaboration with Bernd Bodenmiller lab, at UZH Zurich) provide a snapshot of phosphorylation activity of up to 40 proteins also on single cell level at specific time points.


Single cell time-course data is analyzed to identify groups of cells (subpopulations), which behave in similar way. Then pathway databases are used to create a Prior Knowledge Network (PKN), a map that specifies possible interaction among signaling proteins. Finally, the PKN is translated to dynamic models and then the models are trained to the data using optimization techniques. 

Working pipeline for signaling network identification. FRET and/or mass-Cytometry measurements provide time course protein activity upon single and combined treatments by growth factors. A prior knowledge network is assessed using curated pathway information based on OmniPath, a collection of several pathway databases. Finally the network is trained to the data using global optimization methods.