Speaker
Description
In Boolean networks, biological phenotypes are traditionally mapped to system attractors. However, attractors are often sensitive to the chosen update scheme and the granularity of the influence graph. Applying a more permissive update scheme or adding mediator nodes can easily disrupt plausible attractors, forcing modelers to carefully tune model properties to achieve biologically relevant outcomes.
In this talk, we introduce long-lived sets, a robust generalization of attractors. Long-lived sets are strongly connected sets of states that can be escaped, but never by value percolation (updating a single system variable across all states of the set). Theoretically, long-lived sets are immune to disruptions caused by more permissive update schemes or refinements of the influence graph. Practically, we demonstrate their utility in real-world models: they successfully capture known oscillatory behaviors that standard attractors miss. Furthermore, the number of long-lived sets in a model is often close to the number of attractors, even in models with millions of other "short-lived" strongly connected components. This makes long lived sets not only robust, but highly specific generalization of known attractor-based phenotypes.
