Classifier construction in Boolean networks using algebraic methods

TL;DR

This paper presents an algebraic geometry-based method using Groebner bases to construct classifiers in Boolean networks, aiding in experimental design and understanding causal relations in regulatory networks.

Contribution

It introduces an algebraic approach to classifier construction in Boolean networks, leveraging Groebner bases to handle constraints and steady states.

Findings

Algorithm successfully applied to a 25-component model

Provides a systematic way to find classifiers under constraints

Facilitates linking molecular biomarkers with phenotypes

Abstract

We investigate how classifiers for Boolean networks (BNs) can be constructed and modified under constraints. A typical constraint is to observe only states in attractors or even more specifically steady states of BNs. Steady states of BNs are one of the most interesting features for application. Large models can possess many steady states. In the typical scenario motivating this paper we start from a Boolean model with a given classification of the state space into phenotypes defined by high-level readout components. In order to link molecular biomarkers with experimental design, we search for alternative components suitable for the given classification task. This is useful for modelers of regulatory networks for suggesting experiments and measurements based on their models. It can also help to explain causal relations between components and phenotypes. To tackle this problem we need to use the structure of the BN and the constraints. This calls for an algebraic approach. Indeed we demonstrate that this problem can be reformulated into the language of algebraic geometry. While already interesting in itself, this allows us to use Groebner bases to construct an algorithm for finding such classifiers. We demonstrate the usefulness of this algorithm as a proof of concept on a model with 25 components.