Looking for keys in model organisms: interpreting and keying-up in basal cognition

In this article, I analyse what epistemic operations are involved in extending model organisms to a novel research field. To this end, I draw on the DEKI account of scientific representation, and augment it by introducing the notions of “interpretation-” and “key-sketches”. These refer to the incomplete and heuristically-justified mappings between models and targets that are initially constructed while the model is being developed for a particular use. I substantiate this proposal by analysing three case studies coming from the emerging research area of basal cognition, where stablished model organisms are adopted for novel uses in the context of neuroscientific research.

Keywords: Model organisms, Basal cognition, Scientific representation, DEKI account, Key-sketches, Heuristic reasoning

The philosophical study of model organisms (henceforth, MOs) centres established MOs, and primarily addresses issues related to the justification of inferences drawn on the basis of model organism research and applied to other species or groups. However, a number of representational commitments is established already when a model organism is being introduced or adopted in a new area of research. The epistemic practices involved in the development of novel model organism or their extension to new research domains have been hitherto overlooked, with existing studies covering only other dimensions of practice (Dietrich et al., 2020; Nelson, 2018).

In the current paper, I mobilize the DEKI account of scientific representation (Frigg & Nguyen, 2020; Nguyen & Frigg, 2022), named so after the main components: denotation, exemplification, keying-up, and imputation. DEKI has recently become central to the discussion of the representational role of MOs (see Ankeny & Leonelli, 2020; Sartori, 2023; Prieto & Fábregas-Tejeda, 2025). Building on the account, I highlight the widespread use of heuristic reasoning in establishing novel model organisms or adopting them in new domains. I introduce the notions of “interpretation-” and “key-sketches” to analyze a number of cases from the emerging research field of basal cognition. Basal cognition seeks to extend the standard focus of neuroscience by emphasizing the organisms with no nervous systems, or only a simple one (see Lyon et al., 2021). In doing so, it adopts established MOs which previously have not been relevant to neuroscientific research, extending their domain.

Model organisms are non-human species (e.g., rats (Rattus norvegicus), invertebrates (e.g., Drosophila melanogaster), bacteria (e.g., Bacillus subtilis), or plants (Arabidopsis thaliana)) that are extensively studied across a range of biological phenomena with the hope of translating the results to “other organisms, particularly those that are in some way more complex than the original model, especially humans” (Ankeny & Leonelli, 2020, p. 2). They have been selected, standardized (through selective breeding and inbreeding, and in some cases by direct genetic control) and employed across a variety of biological subdisciplines. For example, D. melanogaster is commonly studied as a model of chromosomal crossover, and the conclusions of such studies are extended to the whole group of sexually reproducing organisms. For rats, one of the most widely employed MOs in biological research, some of its uses include studies of hippocampal activity during spatial navigation, extended to a number of other mammals, most importantly primates, or the genetic underpinnings of addiction, with results extended to humans (see Nelson, 2013).

Much of the translation from MOs to other organisms has been conceptualized by philosophers in terms of scientific representation. According to this line of thinking, MOs function analogously to theoretical models. While their representational status is contested,1 their representational uses aren’t. These have increasingly been viewed through the DEKI account of scientific representation (see Ankeny & Leonelli, 2020; Prieto & Fábregas-Tejeda, 2025; Sartori, 2023). DEKI is particularly useful for the study of model organisms as it explicitly aims to capture the role of model-target mismatches for model-based reasoning in science, which often escapes the alternative accounts of representation (such as the similarity (e.g., Giere, 2004; Weisberg, 2013) or isomorphism views (e.g., French, 2003; Van Fraassen, 2008)). Indeed, it is these mismatches—and the variety of epistemic practices they require from scientists—that will come to the fore in the case studies analysed later in the paper. In what follows, I focus on MOs as representations and adopt the DEKI lens.

The structure of the paper is following. In Sect. 2, I offer a primer on the DEKI account of scientific representation and review the previous literature which used DEKI to account for the use of model organisms in biological research. This allows me to also introduce the main innovations of the current paper: the notions of interpretation- and key-sketches. In Sects. 3 and 4, I introduce and analyse three case studies of how model organisms are adapted to a novel domain, in this case basal cognition and neuroscience, analyzing the role of heuristics in model organism research. I conclude in Sect. 5 by highlighting the epistemic benefits of the proposed perspective.

The DEKI account (Frigg & Nguyen, 2020; Nguyen & Frigg, 2022) proposes four main components of scientific representations, which give it its name: denotation, exemplification, keying-up, and imputation. The framework has initially been applied to model organisms by Ankeny and Leonelli (most importantly in 2020), and their account was later expanded by Sartori (2023). More recently, however, there’s been an alternative interpretation offered by Prieto & Fábregas-Tejeda (2025), who differ in important respects from these previous attempts. However, the current paper is agnostic with regard to main issues at hand in this debate, and for this reason I will attempt to synthesize these distinct positions into a coherent account, only marking the divergences in passing.

To give the overview of both the DEKI and its application to MOs, consider one of the standard case studies, the fruit fly (Drosophila melanogaster) used as a model in the study of chromosomal crossover during sexual reproduction. In this context, a specific interpretation function (I) is applied which allows the researchers to view some properties of the fly2—the biological processes involved in reproduction—as the mechanisms of chromosomal crossover (the domain Z). The actual, physical process of the exchange of genetic material (i.e., the properties the fly itself instantiates) is thus described in more abstract terms, as a particular “generalized conceptualization or theoretical description” (Prieto & Fábregas-Tejeda, 2025, Sect. 3.1) of the process, which constitutes the domain.3 In terms of the DEKI, we take the fly to exemplify (more precisely, I-exemplify) features of that mechanism—and this exemplification is possible, as Sartori notes (2023, Sect. 2.3) due to the large size of Drosophila’s chromosomes in salivary glands, which allowed for epistemic access to their properties early on in the history of genetics (see Levy & Currie, 2015 for some historical notes on this case).

While the DEKI account is also useful for cases of targetless modelling, if a particular model does denote a target—as happens with the D. melanogaster which is taken as a model for all sexually reproducing organisms—its use requires further epistemic steps. The exemplified features in such cases can be imputed to a target. This is done via a key, a crucial element, as the features of the model “rarely [...] hold directly in their target systems” (Frigg & Nguyen, 2020, p. 174).

Keys can have a number of different forms. In some cases they are relatively simple. Identity keys connect strictly equivalent features of Z-models and targets. Conventional keys are used for instance in the standard way of depicting subway connections—e.g., used in the London Underground. In that case, by convention the colours of individual lines map onto the distinct lines of the subway system, while the sequence of stops maps onto the sequence of stops of a given line. In science, conventional relations are at play in various methods which use colour coding of results (e.g., litmus papers). Simple mathematical keys include scaling keys, e.g., used to translate the distance on a map to the actual distance between the places. But other cases are much more complex, when non-linear transformations or mathematical limits are involved (Frigg & Nguyen, 2020, p. 175; Nguyen & Frigg, 2020).

In the case of model organisms, the keys are regarded as split into two categories: broad and local (Sartori, 2023, p. 5). For Ankeny & Leonelli (2020), the key “includes several factors such as principles (e.g., evolutionary conservation); the fit with other models such as simulations, diagrams, and mathematical models of development; and pragmatic factors such as the extent to which the objects chosen as models make the properties tractable and accessible” (Ankeny & Leonelli, 2020, p. 27). As Sartori notes (2023, p. 5), this diverges significantly from the standard treatment of keys in DEKI, and essentially equates the key with the repertoire: a framework for a particular way of doing science, embedded in a conceptual, material, logistical, and institutional setting (Ankeny & Leonelli, 2020, p. 40). Such a broad key loses the essential mapping component postulated in the DEKI. Hence, Sartori proposes to complement it with a local key, which precisely defines the mapping from exemplified features to the target, performing the primary role of the key within DEKI. However, in Sartori’s account of model organisms, the local and broad keys supplement each other, scaffolding the justifications of factual correctness of the model.4 Eventually, they should “reach a reflective equilibrium” (Sartori, 2023, p. 11), but continue to serve different epistemic functions.

Some simple local keys in model organism research are at play when a dosage of a drug is adapted from an animal model, where it was tested, to humans, where it is to be used therapeutically (known as “human equivalent dose”). In such cases, an initial step of human testing of the substance is to translate the “no observed adverse effect level” dose from the animal to the human (as recommended e.g., by Food and Drug Administration, 2005). This often can be simply done by multiplying the animal dose in mg/kg of body mass by a scaling factor—e.g., for rats, the dose is to be multiplied by approximately 0.16 (see Food and Drug Administration 2005). However, most uses of model organisms require more complex keys. For some such cases, Sartori (2023, p. 9) introduces the “functional identity key”, which maps elements of mechanisms with the same overarching function in virtue of the identity of those elements’ contributions to the mechanism. For instance, Sartori argues that such a key is at play in the study of 3D organization of genome in Drosophila as translated to vertebrates. The largely different sets of architectural proteins in the two groups perform the same function (three-dimensional organization of the genome) and as such a functional identity key can be constructed to justify inferences between the model and the target (Sartori, 2023, pp. 9-10; see also Moretti et al., 2020).

In the case of the fly as a chromosomal crossover-model, consider how the model was used to establish the “centromere effect”, i.e., an inhibition of crossovers at and close to the centromere, the point where the two chromatids constituting the chromosome are linked (Beadle, 1932). This effect was then identified in a number of other species, e.g., in humans (Ottolini et al., 2015). The inferences between the model (the fly) and the target (humans) are often implicit. For instance, Ottolini and colleagues note: “Centromeric crossovers interfere with segregation of sister chromatids in Drosophila melanogaster and budding yeast and are associated with an increased risk of aneuploidy in humans” (Ottolini et al., 2015, p. 7, references removed). This draws a parallel between the known mechanism studied in the model and its likely consequence that is observed in humans. Since the description is relatively abstract, this can be taken as an identity key, where the identity of outcomes (abnormal number of chromosomes in a cell) is suggested to result from the same mechanism (the occurrence of centromeric crossovers resulting in the incorrect segregation of sister chromatids during cell division). If the description was offered on a more detailed level, the key would be better characterized as a functional identity key, as it is likely that the actual mechanism is different (e.g., involves homologous or analogous genes in different species).

The DEKI is primarily geared towards capturing scientific representation, and aims to explain how we can reason about target phenomena on the basis of their models. In this sense the historical process of model development goes beyond its intended scope. Nevertheless, the account provides conceptual tools that allow for analyzing the (epistemic) development of novel representations, or the extension of existing models to novel domains.

Crucially, I propose to distinguish interpretation-sketches and key-sketches from the interpretation function and the key that are the central components of DEKI. As discussed above, an interpretation function connects features or properties of the model carrier to features or properties in the domain Z. Frigg & Nguyen (2020, pp. 167-168) require it to be a bijective, potentially non-linear function which provides, for variables, an additional mapping allowing the translation of values of those variables. A key is defined more broadly, as an “association” rather than a strict function (see Frigg & Nguyen, 2020, p. 175). This allows for some of the less formal keys discussed above, for instance grounded in similarity between the Z-representation and the target. The key is repeatedly called a “blank” to be filled by researchers and the only criterion Nguyen & Frigg (2022, p. 61) posit is that it is to be a “systematic ‘translation manual’ ”. Despite this broad specification, there is an intuitive sense of “completeness” the key must meet that follows from the systematicity. Using a metaphor of a map legend, if a colour of a subway line on the map is omitted from the legend, or the marking of some stops differs from others without any description, the legend is incomplete. For scientific representations, a scaling key specifying a drug dosage that would predicate multiplying the dosage by 0.2 if under 1 mg/kg and by 0.3 if over 5 mg/kg and omit the doses between 1 and 5 mg/kg, would be similarly incomplete and in a sense “bad”. Importantly, this is different from the hypothetical or conjectural keys that DEKI allows, which are complete but may lack sufficient evidence base.5

The interpretation function composed with the key establishes the connection between model and target, and allows to justify the derivational correctness of the model. Frigg & Nguyen (2022, p. 296) distinguish between “derivational” and “factual” correctness:

An inference drawn from a representation is derivationally correct if the inferential steps that lead to the conclusion are correct with respect to the rules of the representation and only use premises that form part of the representation. The conclusion of an inference is factually correct if the conclusion is true of the representation’s target.

While the first relies completely on the structure of the representation, including the interpretation and key, the latter requires going beyond the representation, e.g., studying the target phenomenon through experiment or observation.

In the development of a novel model, or when an established one is adopted in a new domain, there is no representation (yet) to ground derivational correctness. Instead of relying on intepretation functions or keys, researchers construct initial versions of these mappings which are used to decide what counts as a part of the representation, and which are themselves incrementally refined in the process. They are supported by different heuristics that guide the development of the model and establish an initial relation to the target which allows for the model to be used as such.

Due to this circularity, interpretation- and key-sketches are incomplete (in the intuitive sense outlined above), possibly arbitrary mappings that fail the inferential job assigned to interpretation functions and keys proper. They may lack sufficient empirical backing, for instance when the full range of values a model variable takes is unknown, and so—cannot be systematically translated into the Z-domain. Further, they may contain some placeholders or be outright incorrect. Nevertheless, they serve an important representational role by allowing the construction, evaluation, and further tuning of the model. As visible in the examples discussed below, they can take forms analogous to complete interpretation-functions and keys, and if the model is accepted, they are filled in and eventually serve their epistemic role.

This notion of “sketches” draws on the concept of a mechanism-sketch proposed by the neo-mechanist account of scientific explanation (Darden, 2002; for mechanistic view of explanation, especially in the context of neuro- and cognitive science see Bechtel & Abrahamsen, 2005; Craver et al., 2007; Machamer et al., 2024; Machamer et al., 2000; Miłkowski, 2013). Mechanism-sketches are incomplete representations of the target mechanism that specify only some of its features. Nevertheless, mechanism-sketches, as well as interpretation- and key-sketches, already provide some understanding of the target and significantly constrain further research.

Establishing interpretation functions and keys is a complex process which “depends on a myriad of factors: the scientific discipline, the context, the aims, and purposes for which the model is used, the theoretical backdrop against which the model operates, etc.” (Frigg & Nguyen, 2020, p. 175). These factors are primarily associated with the repertoire, but through the mutual dependence between the broad and local keys, noted by Sartori (see above in Sect. 2), epistemic factors contribute to key construction as well. In the case of MOs, this process intersects in interesting ways with the selection of an appropriate organism for a particular research purpose (a topic discussed in detail by Dietrich et al., 2020; see also the social aspects analyzed by Lewis et al., 2013).

To highlight this interdependence, consider the following three case studies drawn from the emerging basal cognition research tradition (e.g., Levin & Dennett, 2020; Levin et al., 2021; Lyon, 2020; Lyon et al., 2021). Basal cognition extends the standard subject of cognitive science and neuroscience to include the study of organisms with a limited or no nervous system. In what follows, I look specifically at the studies of signalling across plants, developmental systems, and bacteria. Within the framework, signalling is suggested as one of the basic organizing principles underpinning cognition—and biological information processing more broadly (e.g., Lyon, 2020)—with the most obvious example being neural signalling.6 The first case centres on the development of an interpretation-sketch, as an established model is adopted in a new research context, and hence—in a novel Z-domain. In the second, researchers develop both a novel interpretation-sketch and a key-sketch, altering the interpretation and target of an existing model. The last case adopts a widely accepted interpretation function, but proposes a new key-sketch to extend the scope of properties mapped onto the target.

At the beginning of the twenty-first century, a group of researchers began to call for the development of the field of “plant neurobiology” (Brenner et al., 2006). The point these authors underscore is the presence of systems for long-distance electrical signalling, of vesicle-mediated transport of auxin, a plant-growth hormone, in specialized vascular tissues (analogous to neurotransmitter transport), and production of chemicals “known to be neuronal in animals” (Brenner et al., 2006). This proposal was met with significant scepticism (Alpi et al., 2007; Kingsland & Taiz, 2025; Mallatt et al., 2021; see below), but persisted and later joined forces with the broader basal cognition research programme (e.g., Baluška & Levin, 2016; Baluška & Mancuso, 2021; see also Calvo et al., 2025).

One of the key points in the development of plant neurobiology have been the studies from František Baluška and his lab, which have argued that the vesicle-mediated transport of auxin in Arabidopsis and maize root apices (two standard MOs) bears important structural similarities to how neurotransmitters are transported in nervous systems (Mettbach et al., 2017; Schlicht et al., 2006). The studies reported in Schlicht et al. (2006) developed novel immunolocalization and visualization techniques (based on a novel antibody specific to a single hormone from the auxin class, indole-3-acetic acid), which allowed for discovering accumulation of auxin at the end-poles of root apex cells. They interpreted these results as indicating the limitations of a standard, chemiosmotic theory of auxin transport and support for the view that auxin is transported in a manner structurally similar to the neurotransmitters in the brain, and that the cellular end-poles can be conceptualized as “plant synapses” (Schlicht et al., 2006). Functionally, they take auxin vesicles to be involved in rapid (for plants) signalling, coordinating plant movements and tropisms, and integrating plant bodies (Baluška & Mancuso, 2021; Mettbach et al., 2017).

The inference in this case consists of two important steps: first, the imaging evidence is used to construct an interpretation-sketch which maps the properties of auxin transport mechanisms in plants to the properties from the domain of biological signalling. Second, these authors construct the Z-domain to include properties analogous to those of neurotransmitters in the brains (which figure as the representation target from our current perspective, though no explicit key-sketch is constructed). The inference begins with the structural similarities between the auxin-transport system in Arabidopsis root apices and the neurotransmitter vesicles in the nervous system. Notably, the neurotransmitters are well-studied and their dynamical and functional properties are established. Hence, the authors construct a tentative interpretation function based on those structural similarities, following an inverse form of reasoning, from the target to the model.

The putative vesicle-mediated transport of auxin is thus taken to follow the same dynamics as neurotransmitter vesicles do. This includes rapid export and uptake of auxin between cells. Based on these activity patterns, they are ascribed an analogous function to neurotransmitters in cell-to-cell communication. The interpretation function maps the dynamics between the different cellular systems. The interpretation connects some coarser-grained functional properties across the different organisms, analogously to how functional identity keys work. This step essentially proceeds from target to model, and concludes in establishing the Arabidopsis as a model of cell-to-cell communication, and eventually of information processing (which is the domain of interest for plant neurobiology and basal cognition alike).

But this proposal was met with strong skepticism. Two primary arguments of Alpi and colleagues (2007) highlight the normative dimension of model development. They claim (Alpi et al., 2007, p. 136): “There are two immediate difficulties with this reasoning. (i) Neurotransmitters are not transported from cell to cell over long distances. (ii) The evidence that auxin is sequestered within exocytic vesicles is weak.” The first point aims at the selection of the domain. The putative signalling domain is at odds with the alternative transport interpretation. Long-distance transport of auxin, not present in neurotransmitters, for Alpi and colleagues is an argument against the proposed interpretation (i.e., the signalling domain). The second point raises doubts whether the model actually I-exemplifies the properties postulated (the vesicular structure of auxin transport).

Nevertheless, the potential benefits of Arabidopsis as an information processing-model are wide-ranging. Baluška underscores that the presence of signalling mechanisms, and their structuring analogous to the ones in neural organisms, motivates applying a “cognitive lens” to plants (specifically, flowering plants, Baluška & Levin, 2016). Hence, the fact that the model exemplifies some features of information processing is taken (through a realistic lens) as a novel piece of information about the model organism, and the group that is the standard target of that MO (flowering plants). In turn, this can inform research in neuroscience by identifying basic structure of biological information processing, abstracting away from any peculiarities that may have arisen in the animal part of the tree of life, a feat of crucial importance to astrobiology, for example. Effectively, plants offer the most salient example of a system which allows for the direct study of patterns of difference across putatively cognitive systems. In a sense, “plant neurobiology” furnishes a novel set of boundary cases for neuroscience, in a way similar to how optimality modelling (e.g., Rice, 2015), broadly used in mainstream neuroscience, does.