Reviewer #1 (Public Review): AI-driven Automated Discovery Tools Reveal Diverse Behavioral Competencies of Biological Networks

Mayalen Etcheverry, Clément Moulin-Frier, Pierre-Yves Oudeyer, Michael LevinOpenalex Unknown 2024 · DOI · OpenAlex

Abstract

Many applications in biomedicine and synthetic bioengineering depend on the ability to understand, map, predict, and control the complex, context-sensitive behavior of chemical and genetic networks. This paper introduces a novel framework that treats biological gene regulatory networks (GRNs) as "agents" navigating a problem space. By applying curiosity-driven AI exploration, the researchers efficiently mapped the diverse behaviors of these networks, revealing that biological systems possess innate "navigation competencies"—the ability to reach robust goal states despite significant environmental perturbations—without needing any structural changes.

Introduction: The Behaviorist Perspective in Biology

In biomedicine and synthetic biology, traditional approaches often focus on "rewiring" networks—adding or deleting genes or altering connection strengths. However, the inherent complexity and non-linearity of these networks make predicting the outcomes of structural changes extremely difficult.

This research proposes a "behaviorist" perspective, drawing parallels between an animal navigating a physical environment and a GRN navigating a space of chemical concentrations to maintain homeostasis. The primary challenge lies in the vastness of the behavioral space, which makes traditional random screening infeasible. There is a critical need for automated tools to discover the full repertoire of a network's capabilities and assess their robustness against environmental "pushes" or "obstacles."

Methodology: Curiosity-Driven Exploration

The study combines dynamical systems theory with two major contributions from AI and behavioral science:

  1. Intrinsically-Motivated Goal Exploration Processes (IMGEP): Moving beyond random parameter sampling, the researchers employed a "curiosity-driven" algorithm. The AI selects a "goal" within the behavior space (e.g., specific protein concentrations) and identifies the interventions (initial conditions) required to reach it. This approach allows for an efficient mapping of a network's functional boundaries.
  2. Behaviorist Robustness Battery: To evaluate network "competency," the authors implemented tests inspired by animal behavior studies, including constant noise (fluctuations in gene expression), "pushes" (sudden displacements in transcriptional space), and "walls" (force fields blocking specific paths, requiring the network to find a detour).

The study utilized 30 biological ODE models from the BioModels repository, representing 432 unique systems across various species, alongside 432 random networks as a control group.

Key Insights: Evolution and Basal Cognition

The data revealed that biological networks are significantly more versatile than random ones, suggesting that evolution selects for networks that are easily "reprogrammed" by external stimuli. Furthermore, versatility appears to follow evolutionary strata; human and generic biological networks showed higher versatility than those of bacteria or plants. This hints at a multi-scale competency architecture where molecular flexibility supports organismal complexity.

Some networks demonstrated "basal cognition," exhibiting complex, non-linear trajectories to reach target states even when faced with environmental "walls." This navigation occurs without a nervous system, relying solely on the network's inherent dynamics.

Applications of Behavioral Catalogs

By mapping all reachable states, the researchers created "behavioral catalogs" that distinguish between "healthy" and "disease" regions of a network's behavior space. These catalogs have significant implications for drug discovery and synthetic biology.

Potential applications of behavioral catalogs in biological research and bioengineering.

In a proof-of-concept study using the RKIP-ERK signaling pathway model, the authors designed a timed, non-genetic drug pulse intervention. This intervention successfully reset a "diseased" network state to a "healthy" one, maintaining efficacy even in the presence of environmental obstacles.

Conclusion and Future Directions

The curiosity-driven IMGEP algorithm proved highly efficient, discovering a wider spectrum of states with half the experiments required by random search. While this study was computational (in silico) and focused on isolated models, it opens the path for integrating AI tools with automated lab equipment for in vitro exploration. Future work will also look into using unsupervised machine learning to discover meaningful descriptors of system behavior automatically.

Summary updated

Abstract

Many applications in biomedicine and synthetic bioengineering depend on the ability to understand, map, predict, and control the complex, context-sensitive behavior of chemical and genetic networks. The emerging field of diverse intelligence has offered frameworks with which to investigate and exploit surprising problem-solving capacities of unconventional agents. However, for systems that are not conventional animals used in behavior science, there are few quantitative tools that facilitate exploration of their competencies, especially when their complexity makes it infeasible to use unguided exploration. Here, we formalize and investigate a view of gene regulatory networks as agents navigating a problem space. We develop automated tools to efficiently map the repertoire of robust goal states that GRNs can reach despite perturbations. These tools rely on two main contributions that we make in this paper: (1) Using curiosity-driven exploration algorithms, originating from the AI community to explore the range of behavioral abilities of a given system, that we adapt and leverage to automatically discover the range of reachable goal states of GRNs and (2) Proposing a battery of empirical tests inspired by implementation-agnostic behaviorist approaches to assess their navigation competencies. Our data reveal that models inferred from real biological data can reach a surprisingly wide spectrum of steady states, while showcasing various competencies that living agents often exhibit, in physiological network dynamics and that do not require structural changes of network properties or connectivity. Furthermore, we investigate the applicability of the discovered “behavioral catalogs” for comparing the evolved competencies across classes of evolved biological networks, as well as for the design of drug interventions in biomedical contexts or for the design of synthetic gene networks in bioengineering. Altogether, these automated tools and the resulting emphasis on behavior-shaping and exploitation of innate competencies open the path to better interrogation platforms for exploring the complex behavior of biological networks in an efficient and cost-effective manner. To read the interactive version of this paper, please visit https://developmentalsystems.org/curious-exploration-of-grn-competencies.
BibTeX
@inproceedings{etcheverry2024reviewer,
  author = {Mayalen Etcheverry and Clément Moulin-Frier and Pierre-Yves Oudeyer and Michael Levin},
  title = {{Reviewer \#1 (Public Review): AI-driven Automated Discovery Tools Reveal Diverse Behavioral Competencies of Biological Networks}},
  booktitle = {Openalex Unknown},
  year = {2024},
  doi = {10.7554/elife.92683.2.sa1},
}

Loading related content...

Figure