Computational systems biology

Developing predictive models for precision medicine

Signaling network reconstruction

Inferring cell signaling networks from high-throughput data is a challenging problem in systems biology. Cell signaling plays a central role in cellular processes for constantly adapting environmental stimuli. Several cellular molecules can interact to form a complex signaling network, the subsets of which are associated with one or more signaling pathways, to maintain cellular tissue and organ health.

These networks are not entirely predictable based on the limited information available on the stochastic system. Statistical approaches play a significant role in “network estimation and inference” efforts. More specifically, causal inference is an effective tool for signaling network reconstruction to identify cause-effect relationships among biomolecules.

With the advancement of experimental techniques to quantify protein abundances and post-translational modifications of proteins, it has become feasible to reconstruct the wiring diagram of these networks by statistical means. In particular, mass cytometry allows some 50 different proteins to be monitored at single-cell resolution and hence provides unprecedented data for solving this inverse problem.

Dynamic Graphical Lasso (DynGLasso)

Recent advances in cytometric technology enable us to measure the abundance level of a large number of proteins at the single-cell level across time. Traditional network reconstruction approaches usually consider each time point separately, resulting in inferred networks that strongly vary across time.

In order to account for the possibly time-invariant physical coupling within the signaling network, we extend the traditional Graphical Lasso with an additional regularizer that penalizes network variations.

Graphical Lasso is a tool to learn conditional independence relationships among variables via estimating sparse inverse covariance matrix for a Gaussian graphical model.

Mutual information-based causal inference

The discovery of cause-and-effect relationships from the observation data is a fundamental problem in various scientific fields. Directed structure learning helps researchers design new hypotheses applicable to their disciplines.

Causal learning deals with the identification of cause-and-effect relationships among variables from their joint distribution and decides whether X causes Y or vice versa. Most causal discovery algorithms are based on the Gaussian graphical model assumption. These algorithms identify linear causal relationships among variables.

Here we consider mutual information as a statistical measure to identify nonlinear relationships among variables. We are focused on developing new tools for mutual information estimation.

Our methodology considers mutual information-based conditional independence tests or score functions to find directed networks in the observed data.

Ask the experts


Maria Rodriguez Martinez

María Rodríguez Martínez
IBM Research scientist

Sunil Kumar
Contact also our former colleague Sunil Kumar on LinkedIn or via email.

We gratefully acknowledge generous funding from