Reactionet Lasso - Structure Learning for Stochastic Reaction Networks with Sparse Regression
Reactionet lasso is a software package for structure learning of stochastic chemical reaction networks on the basis of high-throughput single-cell data (e.g. mass cytometry). Our work presents a solution that implicitly enables for exhaustive and systematic structure learning for stochastic reaction networks by translating this task into a sequence of efficiently solvable convex optimization tasks. The reactionet lasso takes advantage of the Chemical Master Equation as a formal link between model parameters and the empirically observed statistical moments of the measured molecular components. We use this formal link to originally formulate a convex relaxation of the exhaustive structure learning task, that can be computed efficiently.
The reactionet lasso software (including example data) is available on Github at: https://github.com/klanna/ReactionetLasso