Causal Structure Learning in Biological Systems

Specialeforsvar ved Margit Schilling Riis

Titel: Causal Structure Learning in Biological Systems

 

 

Abstract: In this thesis we apply current methods for causal structure learning on simulated data from the dynamical system modelling the Maillard reaction. The goal is to use the causal structure of the derivative values of a target variable for out-of-sample prediction of the target trajectory. To improve the prediction result, a new method, that we call stabilizing regression, is proposed as an extension to the invariant prediction methods for dynamical systems Causal KinetiX and AIM. Stabilizing regression averages the _t of several models instead of relying on only one model of the structure of the derivative values of a target variable. On simulated data we see an improvement of out-of-sample predictions using the extension of stabilizing regression for both Causal KinetiX and AIM.

 

 

Vejleder:  Jonas Peters
Censor:    Alexander Sokol, Nordea