(C) PLOS One
This story was originally published by PLOS One and is unaltered.
 . . . . . . . . . .



Algorithmic reconstruction of trophic networks from open-access species lists reveals key organisms in real ecosystems [1]

['Miguel Brun-Usan', 'Departamento Paleobiología', 'Cipb', 'Centre For The Integration Of Paleobiology . Universidad Autónoma De Madrid', 'Uam', 'Madrid', 'Roberto Latorre', 'Departamento Ingeniería Informática. Escuela Politécnica Superior. Universidad Autónoma De Madrid', 'Ángela D. Buscalioni', 'Paloma Alcorlo']

Date: 2026-03 

Biotic interactions, crucial for understanding the ecology and evolution of species, are often conceptualized as ecological networks. However, the complexity of real ecosystems poses challenges for empirical inference, and theoretical interaction models, while informative, frequently fail to undergo empirical validation. This dual limitation creates a gap between theoretical and empirical approaches in portraying ecosystem dynamics and identifying (and protecting) key species, which are critical for conservation efforts and ecosystem management. In order to bridge this operational gap, we present a novel automated protocol capable of generating realistic trophic networks, including multilayer ones, using non-curated, freely-available species lists from real ecosystems as input data. As a proof-of-concept, we applied this method to the species lists contained in the RAMSAR database of wetland ecosystems. Our data mining algorithm enriches these species lists with functional traits, such as body size, habitat, and diet, by integrating information directly sourced from online biodiversity databases. Subsequently, a modified version of the Allometric Niche Model is used to sort species within the trophic network according to their functional traits and ecological roles. After demonstrating the algorithmic robustness of our method and the biological plausibility of the resulting ecological networks, we illustrate its potential to characterize, in a cost-efficient manner, the structure of real-world ecosystems and to identify the organisms that are crucial for maintaining that structure. In this case study, our findings indicate that the robustness of wetland ecosystems often depends on medium-sized, highly mobile organisms occupying intermediate trophic levels.

Understanding who eats whom in nature is key to conserving ecosystems, but inferring these complex food webs (trophic networks) in real environments is resource-intensive and sometimes unfeasible. Although theoretical models can generate plausible networks, they often lack direct links to real-world data. To address this, we developed a fully automated protocol that uses openly available species lists to reconstruct realistic ecological networks. Our method enriches these raw species lists with traits like body size, diet, and habitat, mined from public databases. Then, it applies a well-tested theoretical model to generate trophic networks where links represent biologically plausible interactions. We validate the networks against known ecological patterns and show that they capture the essential structural features of ecosystems. This approach offers a scalable, cost-efficient way to estimate interaction networks, identify keystone species, and assess ecosystem robustness, potentially informing biodiversity conservation and the management of vulnerable habitats worldwide.

Funding: This work was supported by the project PID2019-105546GB-I00 (REDEK, Spanish National Research Council, Spain), which provided funding to MBU, ADB, PA, and JML; by project PID2024-155923NB-I00 (MCIN/AEI and ERDF – “A way of making Europe”, Spain), which provided funding to RL; and by project PID2023-147427NB-I00 (DETECKT), awarded to JML, which provided additional funding for the final stages of the project, including the publication costs (APCs) for this article. The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript.

Data Availability: All code and data supporting the findings of this study are openly available. The software used to generate the results is archived at Zenodo (DOI: https://doi.org/10.5281/zenodo.18405661 ). A public development mirror is available at GitHub: https://github.com/rlatorre-uam/ec-networks . The dataset containing the enriched species lists and the network output files used to generate Fig 4 (Ramsar wetland Orihuela del Tremedal, Spain; five independent replicates) is archived at Zenodo: https://doi.org/10.5281/zenodo.18405396 .

1-2. Inputs consist of non-curated species lists. In our case study, we compiled lists from ≈160 RAMSAR wetland communities worldwide. 3. These raw species lists were enriched with functional traits (e.g., size, diet) automatically mined from public databases. 4-5. Missing data were estimated using two complementary strategies: (i) previously known allometric body mass/size relationships, and (ii) taxonomic inference, where missing values were inferred from the closest phylogenetic relatives. 6. The resulting enriched lists containing trait values were introduced as input parameters in the allometric niche model, a theoretical method capable of generating realistic food webs. 7. In the resulting (automatically constructed) networks, nodes correspond to the species found in the original ecosystem, and links represent highly plausible trophic interactions. 8. This procedure can create large ensemble of homogeneous and realistic trophic networks amenable for quantitative and comparative analysis. Credits: Panel 1 map: Panel 1 world map showing Ramsar sites sourced from the Ramsar Sites Information Service (RSIS, rsis.ramsar.org), licensed under Creative Commons Attribution 4.0 International (CC BY 4.0). Animal silhouettes in other panels were obtained from OpenClipart (openclipart.org), which releases all content under CC0 (public domain).

Here, we present a general automated protocol for generating realistic trophic networks, including multilayer ones, based on raw, open access species lists. Our method utilizes automatically recorded functional attributes of species as a bridge between theoretical models and real-world ecosystems, allowing for a massive generation of ecological networks and serial data analysis (see Fig 1 for a schematic conceptualization). To illustrate the ability of the method to address features of real-world ecosystems, such as spatial compartmentalization, temporal dynamics and high biodiversity indexes [ 28 ], we apply it to the RAMSAR database, an online resource of species composition of wetlands for their conservation and sustainable management. We describe first the algorithm and statistically demonstrate its robustness and biological plausibility. Later, demonstrate its ability to identify functional modules and key species, contributing to the development of practical solutions and guide best practices in applied ecology.

A potential strategy to solve this situation relies on the development of automated methods capable of estimating networks of species interactions in a realistic, efficient and standardized manner, based on accessible data such as online species lists [ 5 , 22 ]. Such procedures would be well-suited for bridging the application of theoretical ecology and network models to real-world case-studies, including the conservation of endangered habitats or the reconstruction of extinct ecosystems. Approaches following this logic have been proposed for restricted scenarios (e.g., fisheries, [ 22 , 26 ]), yet a general procedure to automatically resolve large-scale ecological networks is still lacking [ 27 ].

However, the mismatch between algorithmically derived nodes in abstract networks and species in real-world ecosystems hinders the operational reconciliation between theoretical and empirical approaches [ 20 , 21 ]. A wide range of methods has been developed to estimate ecological networks, including trait- and allometry-based approaches [ 22 ], proxy-based inference of biotic interactions [ 5 ], and niche- or size-structured food-web models linking foraging behavior to network structure [ 23 ]. Despite these advances, many existing models rely on abstract functional groupings rather than explicitly resolved species, which can limit their applicability to empirical systems [ 24 – 25 ].

The study of species interactions constitutes an active research field [ 1 – 3 ], and network-based representations are a powerful tool to this end [ 4 ]. However, species interactions are very complex, diverse and numerous (scaling to the square of the number of species), and characterizing ecosystems demands tremendous sampling efforts [ 4 , 5 ]. This posits a significant challenge for deducing the structure of real-world ecological networks and explains why publicly available databases (e.g., Web-of-life) often provide oversimplified representations of the real, functional ecosystems they aim to describe [ 6 , 7 ]. Despite such limitations, analyses of ecological networks have consistently revealed a highly structured architecture similar to that found in genetic, social or linguistic networks [ 8 – 12 ]. Such commonalities have paved the way to theoretical investigations where, by applying different algorithmic rules, in silico trophic networks with realistic topologies can be derived, enabling the simulation of eco-evolutionary dynamics [ 1 , 13 – 20 ].

Results

The Results are structured into two blocks. The first aims to validate the protocol by evaluating its performance in capturing features of real-world ecosystems. The second analyzes its capacity to recover key ecological features under different scenarios, some of which are highly relevant to conservation ecology. The quantitative descriptors used for this purpose, including their acronyms and brief definitions, are summarized in Box 1, which serves as a quick reference for conceptual and computational details.

Validating the model II: Simulated networks contain realistic predator-prey interactions Beyond apparent or structural similarities with empirica food webs, we assessed whether the specific trophic links generated by our algorithm actually correspond to empirically recorded interactions. As an initial qualitative validation, we manually inspected a sub-set of the > 3000 generated networks (replicates 1 and 2 of all sites) to determine whether the position of their nodes was consistent with the lifestyle of the represented species. We found that the algorithm consistently sorts the species according to the trophic position that they occupy in nature, reflecting what could be recorded under direct observation (see Fig 4 for a representative example). Interestingly, due to algorithmic stochasticity, and that the c i and r i parameters of the allometric niche model are randomly drawn from uniform (U) and β distributions (see Methods), each different run for the same RAMSAR wetland produced distinct networks. A closer analysis shows that indeed only a “core subnetwork” (≈30%) of interactions remains relatively constant across replicates, while >50% of the edges vary (Fig 2A). A sensible interpretation of this result is that our algorithm does not produce a network in which each species is fully connected to everything it can potentially eat, as classical food web representations do. Instead, each replicate shows a “snapshot” of what species might be potentially feeding on at a given time, analogous to a field sample (see Discussion for detailed treatment of potentiality). In other words, the networks resulting from our approach do not represent the cumulative record of all possible interactions for each species, nor do they restrict species to a single interaction per replicate; rather, each replicate draws a subset of interactions that a species (or an individual) could experience within a discrete timeframe, sampled from the species’ full ecological potential. This effectively mirrors the balance between deterministic and stochastic processes observed in the biological realm. On one hand, the stable core reflects highly deterministic, obligate interactions driven by trophic specialization and other functional constraints. Within our framework, such a stable core of interactions emerges from the feeding preferences recorded for specialist taxa, which impose a highly stringent filter on potential links. As a consequence, even across replicates with varying parameterizations of the allometric niche model, only a limited and consistent set of interactions is realized. In contrast, the variable fraction of links captures stochastic influences such as opportunistic feeding, transient prey availability, seasonality, or simple chance, which can lead species to feed on different prey in different spatiotemporal contexts. Importantly, this result is not only conceptually relevant, but also allows for more detailed identification of key species and functional groups within ecosystems, with implications for conservation biology (see below and Discussion). PPT PowerPoint slide

PNG larger image

TIFF original image Download: Fig 4. Example of a biologically plausible trophic network automatically generated by our algorithm. (A). Case example: the RAMSAR wetland Orihuela del Tremedal (Spain) whose original species list (N = 110) included un-resolved names. (B). An example (1st replicate) network reveals that simulated trophic interactions are generally consistent with the known lifestyles of the species present in the ecosystem. For illustrative purposes, the silhouettes of representative species of the main trophic levels have been juxtaposed to the network. Node size reflects relative body mass; while colors indicate functional guilds (blue = plants, green = herbivores, orange = carnivores). Notice that some highly specific relationships such as insect-eater plants (circle in the central, lower part of the Figure), are adequately captured by the automated algorithm. This is achieved by introducing a few algorithmic rules devoted to handle these highly specific interactions (see Materials and Methods and SI). Credits: Panel A (left to right): world map from Natural Earth (naturalearthdata.com, public domain); map of Spain from the Instituto Geográfico Nacional (IGN), licensed under Creative Commons Attribution 4.0 International (CC BY 4.0); original photograph taken by M. Brun-Usan. Panel B: animal silhouettes obtained from OpenClipart (openclipart.org) and released under the CC0 (public domain) dedication. https://doi.org/10.1371/journal.pcbi.1014061.g004 To conduct a more rigorous, quantitative test, we next used the basic setting of our protocol to generate 100 virtual replicates of five well-resolved and widely referenced empirical networks found in the literature [6], and checked, one by one, if the predicted interactions were present in the reference (empirical) networks. An interaction (or its absence) was considered correctly predicted only when both the interacting species and the direction of the interaction matched the empirical record. As shown in Fig 5A, our protocol typically predicts over 80% of the empirically recorded interactions, and that performance is relatively insensitive to the species richness or overall network connectivity. A closer inspection of the interactions that are not correctly predicted by the model shows that they are not fully attributable to errors of the model, but to the assumption of shared habitat, namely, that any two organisms satisfying the allometric and trophic criteria can potentially interact. For example, if a species list includes both a marine dolphin (carnivore) and a rabbit, having an adequate allometric relationship, the basic algorithm may allow a potential interaction despite their spatial segregation. Fortunately, those exceptions to this rule can be easily handled using different buffering strategies (see SI and the section on Multilayer Networks below). PPT PowerPoint slide

PNG larger image

TIFF original image Download: Fig 5. Quantitative comparison between empirically measured trophic interactions and those predicted by the automated protocol. The Figure compares, on a link-by-link basis, 100 automatically generated network replicates with the corresponding empirical food webs reported for five wetland-like ecosystems with varying species richness and connectivity [6]: Benguela Coastal Ecosystem (N = 29), Ythan Estuary (N = 92), Tuesday Lake (N = 56), Skipwith Pond (N = 37), and Broadstone Stream (N = 37). Underestimated links (red) correspond to interactions present in the empirical network but missing in the reconstructed networks, whereas overestimated links (yellow) are interactions present in the reconstructed networks but absent from the empirical data. (A) Comparison using a default estimated average connectivity of ≈0.12. (B) Comparison using the empirically measured connectivity for each ecosystem. (C) Comparison with equally connected random networks, in which each possible link is realized with a probability equal to the empirical connectivity. https://doi.org/10.1371/journal.pcbi.1014061.g005 Finally, to assess whether the level of agreement could arise by chance, we compared our results with equally connected random networks. In this case, prediction accuracy dropped to approximately 20% (Fig 5C). This marked contrast between random and reconstructed networks demonstrates that the correspondence between predicted and empirical interactions reflects genuine biological signal rather than a statistical artifact.

Identifying key organisms for maintaining ecosystem structure Modularity analysis (based on the Louvain method, see SI) reveals positive modularity in all simulated networks (0.1 < MOD < 0.5, Fig 6A), indicating the existence of functional subgroups (communities). These values are fully consistent with the few available estimates of modularity in empirical food webs (e.g., [35]), which typically report MOD values in the range of ≈0.2-0.4. Detailed inspection of these communities further reveals that the most densely (intra-connected) modules tend to be enriched with carnivores, while the remaining communities show a more hierarchical and sparser structure, composed by cohorts of functionally similar herbivores and the plants they preferably consume (Fig 6E-6G). This structure roughly corresponds to that typically found in actual ecosystems (Fig 6G), where herbivores engage in specific interactions due to plant defences (i.e., toxins) [24,36,37]. In this context, modularity provides a heuristic framework for identifying species that are critical for maintaining ecosystem connectivity and structure (in the sense that their extinction may elicit a fragmentation of the whole network, triggering extinction cascades and a loss of ecosystem functioning). According to network theory [10,15], those species are not necessarily those with the highest degrees, but rather those acting as functional links between modules. Our analysis reveals that those key species acting as “bridges” between modules tend to be medium-sized, high-mobility, generalist animals, such as frogs, rodents or crayfish that feed in one ecological compartment (e.g., water) but serve as prey in others (e.g., land) (Fig 6E-6G). PPT PowerPoint slide

PNG larger image

TIFF original image Download: Fig 6. Modularity analysis. (A). The number of semi-independent modules remains relatively low (around 4) and constant for most ecosystems. (B). The average clustering (CLUST) increases in large ecosystems. (C). The average modularity (MOD) decreases in large ecosystems. (D). Some of the modularity-related measures are linearly correlated as indicated by the corresponding Pearson’s correlation coefficient. Anticorrelation between CLUST and MOD suggests that modules become less independent as ecosystems grow in size. (E-F). Two case examples of modular structure in our dataset: (E). Naikaikemi (Japan, N = 28), and (F). Whooping Crane (USA, N = 36). Colors represent the optimal division maximizing modularity, dashed line separates carnivores-scavengers from the remaining organisms and dashed nodes represent “bridge” species between modules. (G). An idealized network showing how the modules are, in general, organized across networks. https://doi.org/10.1371/journal.pcbi.1014061.g006 Additionally, the high modularity (MOD) values observed in smaller ecosystems (N < 150) and the negative correlation between MOD and the number of partitions PART further suggest that, while the number of modules remains relatively constant (around PART = 4) across ecosystems (Fig 6A and 6D, similar to empirical values reported in [35]), the modules become less independent as the number of species increases (Fig 6B-6C). This is also evidenced by a consistent increase in the average clustering CLUST, which indicates that inter-module connections become denser in larger networks (Fig 6D). This scaling implies that ecosystem connectivity increasingly relies on a growing set of bridge species, whose loss may disproportionately affect network integrity and functioning. In other words, larger ecosystems may require more key bridge species to maintain their structure and ecological functions, with clear implications for ecosystem management (see Discussion).

[END]
---
[1] Url: https://journals.plos.org/ploscompbiol/article?id=10.1371/journal.pcbi.1014061

Published and (C) by PLOS One
Content appears here under this condition or license: Creative Commons - Attribution BY 4.0.

via Magical.Fish Gopher News Feeds:
gopher://magical.fish/1/feeds/news/plosone/