AI-driven Automated Discovery Tools Reveal Diverse Behavioral Competencies of Biological Networks

Mayalen Etcheverry, Clément Moulin-Frier, Pierre-Yves Oudeyer, Michael Levin

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 formalizes a view of gene regulatory networks (GRNs) as agents navigating a problem space. By adapting curiosity-driven exploration algorithms from the AI community, the researchers developed automated tools to efficiently map the repertoire of robust goal states that GRNs can reach despite perturbations. The data reveal that models inferred from real biological data can reach a surprisingly wide spectrum of steady states, showcasing various competencies that living agents often exhibit. This emphasis on behavior-shaping and exploitation of innate competencies opens the path to more efficient interrogation platforms for exploring biological network dynamics.

Introduction: Biological Networks as Agents

In biomedicine and synthetic biology, predicting or controlling the behavior of complex chemical and genetic networks is a significant challenge. Traditionally, these networks are viewed as "hardware" to be rewired (e.g., through gene therapy). However, biological systems are incredibly non-linear and redundant; simply changing one part doesn't always lead to the desired outcome.

Biological systems like slime molds and ciliates serve as examples of unconventional agents that exhibit complex problem-solving behaviors.

The emerging field of diverse intelligence suggests investigating the problem-solving capacities of unconventional agents. This paper addresses the gap in quantitative tools for exploring these competencies by treating a GRN as an agent with "basal cognition" and using AI to discover its innate navigation skills.

Methods: Curiosity-Driven Discovery

The authors introduce a two-part automated discovery pipeline:

  1. Curiosity-Driven Exploration (IMGEP): They adapted Intrinsically-Motivated Goal Exploration Processes (a type of AI used in robotics) to "poke" GRNs. Instead of random screening, the AI sets "goals" in a behavior space and learns which interventions (initial conditions) move the network toward those goals. This search prioritizes exploring novel, unmapped regions of the network's potential states.
  2. Behaviorist Robustness Battery: Once reachable states are found, the network is subjected to "displacement tests" inspired by animal behavior studies, including stochastic noise, sudden state shifts ("pushes"), and force fields ("walls") that prevent natural trajectories.
  3. Computational Implementation: The team developed SBMLtoODEjax, a library that converts standard biological models into high-performance JAX code, allowing for massive parallel simulation of these networks.

Key Insights and Findings

The study utilized a curated database of 432 systems derived from 30 continuous (ODE) models. Key results include:

  • Versatility is a Feature: Biological networks can reach a wide spectrum of steady states just by changing initial conditions, without any structural rewiring.
  • Evolutionary Advantage: Evolved biological networks are significantly more versatile and robust than random networks, suggesting that navigational competency is a trait shaped by evolution.
  • Efficiency of Curiosity: The AI-driven search was vastly more sample-efficient than random screening. In one oscillator design task, curiosity search found over 1,100 valid configurations, while random search found only 42.

Therapeutic Design and Intervention

By treating a GRN as an agent, the researchers could identify "healthy" and "disease" regions in the behavior space. They demonstrated that they could design drug-like interventions to "nudge" a network from a disease state back to a healthy one, even in the face of noise.

A conceptual landscape showing how drug interventions can guide a network agent toward a desired goal state (g) across a complex behavioral terrain.

Using the RKIP-ERK signaling pathway (linked to cancer) as a case study, the authors successfully identified a stimuli-based intervention that could robustly reset "disease" states to "healthy" ones despite environmental perturbations.

Limitations and Future Work

While the study relies on Ordinary Differential Equation (ODE) models, which are approximations, it provides a powerful proof-of-concept. Future work aims to use unsupervised learning to let the AI discover relevant behavior dimensions autonomously. The next major step is applying these "curiosity" algorithms to real-world, in vitro experiments where AI controls lab robots to discover the competencies of living tissues in real-time.

Summary updated

Abstract

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{etcheverry2025ai,
  author = {Mayalen Etcheverry and Clément Moulin-Frier and Pierre-Yves Oudeyer and Michael Levin},
  title = {{AI-driven Automated Discovery Tools Reveal Diverse Behavioral Competencies of Biological Networks}},
  booktitle = {eLife},
  year = {2025},
  doi = {10.7554/elife.92683.4},
}
Figure