Cell dynamics on energy landscape: Comparing attractor detection in Boolean network and diffusion-based models under in silico gene perturbations

Abstract

Steady-state attractor detection differs fundamentally between discrete Boolean network (BN) and continuous diffusion-based models because they operate on different state spaces and exhibit distinct dynamic behaviors. Here, we propose CellLand (Cell dynamics on energy Landscape) to systematically compare BN- and diffusion-based approaches for quantifying cellular dynamics on the energy landscape, with a particular focus on attractor identification under in silico gene perturbations. Although the energy landscape provides a unifying conceptual framework, BN-based models face specific challenges, including state-space explosion and pronounced sensitivity to network architecture, when assessing attractor stability and transition accessibility. Nevertheless, BN-based analyses can remain reliable and interpretable when the underlying network topology is suitable, while being comparatively less sensitive to parameter-estimation uncertainty and technical noise. Importantly, we show that attractor identification depends critically on network structure, and that no general theory currently exists for optimizing BN topology for dynamic cell-state modeling. Our findings deepen the comparative understanding of BN- and diffusion-based frameworks and provide practical guidelines for selecting and designing models for robust simulation of complex gene-regulatory systems under perturbations.

Lingyu Li

Postdoctoral Fellow

Focus on bioinformatics, including but not limited to spatial transcriptomics analysis, sparse statistical learning and biomarker identification.