BioelectricaAlterspan Labs
Goals and control  ·  Article

Perturbation-based testing of network robustness and versatility

Perturbation-based testing probes a model network by disturbing it and recording whether it still reaches its usual end state. Etcheverry, Moulin-Frier, Oudeyer and Levin (2025) defined three families of disturbance: noise, push and wall.45 They applied them to a database of gene regulatory network models. Low sensitivity was read as evidence of robustness alongside versatility.6

Earliest held
2014, Budnikova M, Habig JW, Lobo…
Most discussed in
AI-driven automated discovery tools reveal…, 2025
In the library
10 passages in 4 works
Rewritten
2026-10-29
18501875190019251950197520002025Budnikova M, Habig JW, Lobo D, Cornia…, 20141Budnikova M, Habig JW, Lobo D, Cornia…, 20142Lobo D, Levin M, 20153Etcheverry M, Moulin-Frier C, Oudeyer…, 20254Etcheverry M, Moulin-Frier C, Oudeyer…, 20255Etcheverry M, Moulin-Frier C, Oudeyer…, 20256Padalko A, Karavaeva V, Zamarreno Beas…, 20257
01

Automated model discovery

Related work by Budnikova and colleagues and by Lobo and Levin also concerns automated model discovery. Budnikova, Habig, Lobo, Cornia, Levin and Andersen (2014) laid the foundation for a system that draws on shape-based data from a repository of documented experiments.1 They showed that the graph difference evaluation could also serve as a fitness metric during evolutionary search.2 Lobo and Levin (2015) modelled gene, protein and metabolite regulatory networks with a system of nonlinear partial differential equations, as the basis for an automated discovery system.3

02

Three kinds of perturbation

Etcheverry and colleagues (2025) set out three perturbation families. The noise perturbation is parametrized by a standard deviation, scaled to the extent of the observed trajectory, and by a period in seconds.4 The push perturbation is parametrized by a magnitude and a number of occurrences. The wall perturbation is parametrized by a length and a number. Its walls are placed where the network would naturally go without the disturbance.5 Each family was also varied in degree, giving sub-families of increasing strength.

03

What the tests showed

Etcheverry and colleagues reported median sensitivity for the 432 systems in their database, across the three families. Sensitivity varied between systems. Variation between magnitudes of perturbation was expected. The median stayed relatively low. They took this to suggest that these networks can show versatility with respect to interventions and robustness with respect to perturbations.6 Some extreme examples took winding paths with detours and loops, yet reached the same endpoint despite several pushes or walls.

04

Use of the same pair of terms

The pairing of robustness and versatility also appears in work on the evolution of respiratory chains. Padalko, Karavaeva, Zamarreno Beas, Neukirchen and Sousa (2025) noted that microbes often carry more than one enzymatic solution in the same organism. They described this as giving redundancy, robustness and versatility in dealing with O2.7 Padalko and colleagues discuss enzyme redundancy rather than perturbation tests. It shows the terms are used in more than one setting.

SourcesEach quotation was checked word for word against the passage it opens.
  1. we laid the foundation for a system of automated model discovery and development that incorporates shape-based experimental data from a repository of documented experiments.Budnikova M, Habig JW, Lobo D, Cornia N, Levin M, Andersen T, 2014 · Design of a flexible component gathering algorithm for converting cell-based… · open at passage 32
  2. We showed that the utility of the graph difference evaluation could be further extended as a fitness evaluation metric during evolutionary search.Budnikova M, Habig JW, Lobo D, Cornia N, Levin M, Andersen T, 2014 · Design of a flexible component gathering algorithm for converting cell-based… · open at passage 32
  3. we modeled the behavior of gene, protein, and metabolite regulatory network with a system of nonlinear partial differential equations (PDE).Lobo D, Levin M, 2015 · Inferring regulatory networks from experimental morphological phenotypes: a… · open at passage 35
  4. the noise perturbation Unσn,pn∨y which is parametrized by its standard-deviation (scaled proportionally to the extent of the observed trajectory y prior perturbation) and period (secs)Etcheverry M, Moulin-Frier C, Oudeyer PY, Levin M, 2025 · AI-driven automated discovery tools reveal diverse behavioral competencies of… · open at passage 70
  5. parametrized by its length (proportional to the extent of y) and number, and where walls are generated in locations of the space that the GRN wouldEtcheverry M, Moulin-Frier C, Oudeyer PY, Levin M, 2025 · AI-driven automated discovery tools reveal diverse behavioral competencies of… · open at passage 70
  6. the median sensitivity remains relatively low, suggesting that GRNs could not only exhibit versatility (with respect to the considered interventions) but also robustness (with respect to the considered perturbations)Etcheverry M, Moulin-Frier C, Oudeyer PY, Levin M, 2025 · AI-driven automated discovery tools reveal diverse behavioral competencies of… · open at passage 35
  7. Frequently, more than one enzymatic solution is present in the same organism, providing them with redundancy, robustness and versatility to deal with O2.Padalko A, Karavaeva V, Zamarreno Beas J, Neukirchen S…, 2025 · Bioenergetics evolution: the link between Earth's and Life's history · open at passage 25
Linked ideas
Both treat gene regulatory networks as dynamical systems whose behaviour can be studied by computation rather than by listing parts.
Noise robustness is the property that the noise perturbation family is designed to measure in model networks.
Reaching the same form despite noise is the developmental counterpart of reaching the same endpoint despite pushes or walls.
Consistent arrival at one endpoint despite interference is the behavioural evidence on which goal-directedness is judged.
Agent-based models were noted in 2014 as tractable to simulated experimental manipulation, the idea behind later perturbation testing.
Recovery after a push or wall resembles regulation toward a set state, though the passages do not make this link themselves.
Evolvabilityrelated to
Passages don't discuss evolvability directly; Etcheverry tests GRN robustness/versatility via push, wall, noise perturbations, while Budnikova uses evolutionary search and Padalko cites redundancy-driven robustness/versatility in evolved respiratory chains.
Where it is discussedPassages matching noise perturbation, push perturbation, wall perturbation, robustness and versatility
2025AI-driven automated discovery tools reveal diverse behavioral competencies of… · Etcheverry M, Moulin-Frier C, Oudeyer…7
2025Bioenergetics evolution: the link between Earth's and Life's history · Padalko A, Karavaeva V, Zamarreno Beas…1
2015Inferring regulatory networks from experimental morphological phenotypes: a… · Lobo D, Levin M1
2014Design of a flexible component gathering algorithm for converting cell-based… · Budnikova M, Habig JW, Lobo D, Cornia…1