Minds of small things · Article

Gene regulatory network models of associative learning and bistable memory

Gene regulatory network models of associative learning ask whether a circuit of molecules can pair two stimuli, so that a neutral one comes to trigger the response to a significant one. Fernando and colleagues drew a genetic circuit by analogy with a Hebbian neural circuit². The Genetic Associative Memory model of Sorek and colleagues includes a memory module that keeps memories for long periods⁴.

Earliest held: 1896, Wilson, E. B.
Most discussed in: Conditioned Reflexes: An Investigation of the…, 1927
In the library: 334 passages in 22 works
Rewritten: 2026-10-20

The conditioning template

Bayliss (1915, Principles of General Physiology) described Pavlov's account of temporary association. A stimulus presented several times together with an unconditioned stimulus, such as food in the mouth, comes to excite salivary secretion when presented alone¹. Bayliss called the result a conditioned reflex and saw in it a way to test the effect of other stimuli on cerebral phenomena¹.

Circuits and chemical networks

Fernando and colleagues (2009) drew a genetic circuit in two halves, by analogy with a classical Hebbian neural circuit. One half represents the known response to the unconditioned stimulus. The other represents the response to the conditioned stimulus that is to be learned². They noted that circuits could hold more than two unconditioned stimuli, and that two stimuli are the minimum needed to show Hebbian learning². McGregor and colleagues (2012) worked with evolved chemical networks. Their training regimes did not explicitly require an unconditioned response, because they assumed a separate chemical reaction produces it. All their networks showed extinction, although none was explicitly evolved on an extinction paradigm³.

The GAM and bistable memory

Sorek, Balaban and Loewenstein (2013) proposed the Genetic Associative Memory model, called GAM. It has three components. A memory module supplies the long time-scale needed to keep memories. A second mechanism encodes the desired memories. A response mechanism reads them out when the relevant stimuli arrive⁴. They first treated a predefined association, then generalized to many associations and to a population of GAMs⁴. In the population, the signal-to-noise ratio of the response stayed above 1 for a much older conditioned stimulus than in a single GAM⁵.

Tests in living cells

Gagliano and colleagues (2016) trained pea seedlings in a Y-maze, using a fan as the conditioned stimulus and a blue LED as the unconditioned stimulus⁶. De la Fuente and colleagues (2019) reported conditioned behaviour in Amoeba proteus. They used a direct-current electric field as the conditioned stimulus and a chemotactic peptide as the unconditioned stimulus, and saw persistent locomotion lasting 44 minutes on average⁷. Markel (2020) answered the plant report with a paper titled as a lack of evidence⁸.

SourcesEach quotation was checked word for word against the passage it opens.

By the former, a stimulus presented several times in conjunction with an unconditioned stimulus, such as food in the mouth, has ultimately the effect of exciting salivary secretion when presented alone.Bayliss, W. M., 1915 · Principles of General Physiology · open at passage 2192
One-half (on the left in the diagram) represents the known response to the unconditioned stimulus, and the other (on the right) the response to the conditioned stimulus that is to be learned.Fernando CT, Liekens AM, Bingle LE, Beck C, Lenser T…, 2009 · Molecular circuits for associative learning in single-celled organisms · open at passage 9
All the networks presented here show extinction, even though they were not explicitly evolved on an extinction paradigmMcGregor S, Vasas V, Husbands P, Fernando C, 2012 · Evolution of associative learning in chemical networks · open at passage 8
Our Genetic Associative Memory model (GAM) for associative learning is based on three components: (1) a memory module that provides the long time-scale necessary for the maintenance of memories for long periods of time;Sorek M, Balaban NQ, Loewenstein Y, 2013 · Stochasticity, bistability and the wisdom of crowds: a model for associative… · open at passage 7
the SNR of the response of the population of GAMs is larger than 1 for much ‘older’ CS.Sorek M, Balaban NQ, Loewenstein Y, 2013 · Stochasticity, bistability and the wisdom of crowds: a model for associative… · open at passage 40
we employed a classical conditioning paradigm where a neutral environmental cue (a conditioned stimulus, CS) predicted the occurrence of light, which is biologically significant (an unconditioned stimulus, US).Gagliano M, Vyazovskiy VV, Borbély AA, Grimonprez M…, 2016 · Learning by Association in Plants · open at passage 3
We use a controlled direct-current electric field as the conditioned stimulus, and a specific chemotactic peptide as the unconditioned stimulus.De la Fuente IM, Bringas C, Malaina I, Fedetz M…, 2019 · Evidence of conditioned behavior in amoebae · open at passage 0
In December 2016, Gagliano et al. published a report claiming plants have the capacity to learn adaptive behaviors during individual lifetimesMarkel K, 2020 · Lack of evidence for associative learning in pea plants · open at passage 4
memory may be stored using a cell-intrinsic substrate, such as polynucleotide sequences (e.g. RNA), post-translational histone modifications, or DNA methylation patternsGershman SJ, Balbi PE, Gallistel CR, Gunawardena J, 2021 · Reconsidering the evidence for learning in single cells · open at passage 6

Linked ideas

Transient versus asymptotic memory in bistable systemspart of / contains

The GAM memory module is a bistable element, so this article supplies the long time-scale storage the GAM model depends on.

Conditioned reflexprecursor of / follows

Pavlov's conditioned reflex gave the stimulus pairing and extinction that the gene circuit models try to reproduce.

Associative learning in plantsparallels

Covers the pea seedling experiment and its dispute, a test of the same conditioning logic in an organism without neurons.

Gene regulatory networks as Boolean networksrelated to

Another way to model gene circuits computationally, as networks whose states switch between discrete values.

Basal cognitionrelated to

Frames learning in single cells and aneural organisms as a question about cognition below the level of nervous systems.

Bacterial memory and history-dependent behaviorparallels

Another case of single cells holding a record of past conditions in their behaviour.

Homeostatic regulation and hierarchical reflex control as active inferenceparallels

Both treat conditioned/unconditioned stimulus association as a modelled mechanism: Fernando's gene circuits in single cells, Pezzulo's active inference in organism-level control. The passages show shared CS/US vocabulary, not a direct link.

Positional memory models of pattern regenerationrelated to

Auto-associative systems are among the non-neural network models of pattern regeneration that Manicka and Levin list.

Developmental memory capacity, robustness and attractor basins in gene networksrelated to

Concerns memory held in gene network interactions, the setting of the Szilágyi analysis of recovering adult states.

Auto- and heteroassociative memory in gene regulatory networksrelated to

Treats the same idea of gene networks acting as associative memories; read together for the modelling detail.

Where it is discussedPassages matching genetic associative memory, GAM, associative learning in gene networks, conditioned stimulus

1927	Conditioned Reflexes: An Investigation of the Physiological Activity of the… · Pavlov, I. P.	248
2013	Stochasticity, bistability and the wisdom of crowds: a model for associative… · Sorek M, Balaban NQ, Loewenstein Y	34
2009	Molecular circuits for associative learning in single-celled organisms · Fernando CT, Liekens AM, Bingle LE…	8
1915	Principles of General Physiology · Bayliss, W. M.	7
2015	Active Inference, homeostatic regulation and adaptive behavioural control · Pezzulo G, Rigoli F, Friston K	4
2021	Reconsidering the evidence for learning in single cells · Gershman SJ, Balbi PE, Gallistel CR…	4
2022	Learning in Transcriptional Network Models: Computational Discovery of… · Biswas S, Clawson W, Levin M	4
2020	Lack of evidence for associative learning in pea plants · Markel K	3
2016	Learning by Association in Plants · Gagliano M, Vyazovskiy VV, Borbély AA…	3
2021	Behaviorist approaches to investigating memory and learning: A primer for… · Abramson CI, Levin M	3