Computational exploration of high-dimensional stochastic biochemical reaction networks: A machine learning-assisted approach

One of the transformational insights in systems biology is that inherent molecular noise in gene regulatory networks leads to substantial single-cell variability and that the regulatory designs of gene regulatory networks are critically influenced by stochasticity.

These realizations were to a large part enabled by discrete stochastic models of the chemical interactions between molecules in networks of interacting genes, mRNA and protein.

As a consequence, stochastic simulation is today one of the most important tools in systems biology, and a large number of black-box simulators are readily available to the modeler. An efficient simulator is only one of many components in the model analysis process. Often, no or only a limited amount of experimental data is available early in a project.

The challenge is then to use the simulator to go from uncertainties over 3-4 orders of magnitude in parameters to robust qualitative predictions.

Such model exploration is constrained both by computational cost and by thee extensive manual input needed from the modeler. Today there are no tools for model exploration that scale to complex, non-linear and stochastic models. We propose to develop black-box machine-learning assisted methods to scale to high-dimensional stochastic models.

By fusing ideas from scientific computing, machine learning and systems biology we will develop software that automates the analysis of stochastic biochemical pathway models under large uncertainty.