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. 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. 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. 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.

Introduction: From Micromanagement to Behavior Shaping

In biomedicine and synthetic biology, chemical and genetic networks are frequently treated as simple machines. When a system malfunctions, the standard approach is to "rewire" it through structural changes like gene therapy or permanent knockouts. However, these networks are inherently non-linear and redundant, making it difficult to identify the precise interventions needed to reach a healthy state.

This research shifts the perspective from micromanaging network structure to shaping the behavior of the network. By treating a Gene Regulatory Network (GRN) as an unconventional "intelligent" agent navigating its own state space, researchers can identify latent competencies that allow the system to reach desired phenotypes without permanent genetic alteration. To facilitate this, the authors propose a formal isomorphism between dynamical systems and behavioral science.

Proposed isomorphism between dynamical systems and behavioral science terminology.

Methods: The Automated Discovery Pipeline

The study introduces a two-step automated discovery pipeline designed to map the "behavioral catalogs" of biological systems efficiently:

  1. Autotelic Curiosity Search (IMGEP): Adapted from AI and robotics, Intrinsically-Motivated Goal Exploration Processes (IMGEP) allow the system to "set its own goals." Instead of random sampling, the AI attempts to reach unexplored regions of the behavior space, efficiently mapping reachable steady states (phenotypes) within a limited experimental budget.
  2. Behaviorist Empirical Tests: Once behaviors are discovered, their robustness is assessed using tests inspired by animal behavior science, including noise (fluctuations), "pushes" (temporary state shifts), and "walls" (energy barriers or constraints).
Overview of the automated discovery framework, from motivation to behavioral catalog reuse.

The pipeline was implemented using JAX for high-performance parallelization of differential equation (ODE) simulations, testing 432 unique systems derived from the BioModels repository and a baseline of random networks.

Key Results and Insights

Efficiency and Discovery

Curiosity-driven search proved significantly more efficient than traditional methods. In mapping behavior spaces, it outperformed random search using only one-third of the experimental budget. In synthetic oscillator design, curiosity search identified 1,167 successful configurations, whereas random search found only 42, and gradient descent found zero due to local minima.

Innate Robustness and Versatility

Biological networks demonstrated a high degree of "navigation competency." Many were naturally robust to perturbations, taking complex "detours" through their state space to reach target goals. Furthermore, biological networks were found to be significantly more versatile—capable of reaching more diverse states—than random networks of similar size, suggesting that versatility is a trait shaped by evolution.

Therapeutic and Engineering Potential

The resulting "behavioral catalogs" can be used to identify specific drug-like stimuli that reset a network from a disease state to a healthy state. This approach exploits the innate competencies of the network rather than forcing structural changes, offering a new paradigm for medical intervention and synthetic circuit design.

Conclusion and Future Directions

While currently limited to in silico ODE models, this framework provides a powerful quantitative toolbox for exploring biological complexity. Future work aims to apply these tools to in vitro experiments and test for higher-level cognitive competencies, such as memory and creative problem-solving, within molecular networks. By treating biological networks as agents, we can move beyond simple repair toward sophisticated behavioral guidance.

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{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