1. Functional redundancy is a key biological concept that offers critical insights into microbial ecosystem stability and responses to environmental perturbations. Existing methods to quantify redundancy typically rely on gene annotations as functional units, limiting their ability to capture complex or partially characterised traits and making analyses from metagenomic data-often fragmented or incomplete-challenging. 2. Here, we present miFRED (microbial Functional REDundancy), an accessible computational approach to quantify functional redundancy of microbial phenotypes and communities based on phenotypic rather than genetic annotations. Leveraging 86 machine-learning-predicted phenotypes as functional units, miFRED bridges the genotype-phenotype gap and enables interpretable and accurate assessment of redundancy at both community and single phenotype levels, as demonstrated on simulated communities. 3. As case studies, miFRED was applied to over 300 metagenomic samples from anaerobic systems, to a longitudinal dataset of a gut microbiome subject to antibiotic treatment and 100 marine and soil metagenomes. Across these applications, miFRED elucidated relationships between redundancy, community structure and functional organisation. Comparison to gene annotation-based analyses highlighted the improved interpretability of phenotype-based redundancy estimates, which allowed clearer resolution of functional changes associated with microbiome perturbation and recovery in the human gut dataset. In anaerobic digestion communities, analyses considering environmental conditions uncovered changes in redundancy, complexity and functional profiles with rising temperatures, with more extreme conditions favouring simpler, more redundant communities and shifts in redundant functions. 4. Using phenotype-based functional units, miFRED provides a practical and general framework for consistent functional redundancy analyses across environments and metagenomic datasets. By interpreting redundancy patterns in the context of community organisation and environmental factors, it enables exploration of microbial functional landscapes across environmental conditions in both natural and engineered ecosystems.
Environmental DNA (eDNA) is an emerging tool for surveying marine organisms. Advances in molecular assay designs and bioinformatic methods provide new opportunities for detecting rare species; however, overcoming methodological challenges of rare species detection remains an important obstacle for the widespread deployment of eDNA methods for biomonitoring. Here, we provide a rigorous evaluation of eDNA sampling and processing methods to optimize eDNA capture and detection of a rare delphinid species. We applied a hierarchical Bayesian occupancy model to separate the effects of presence, capture and detection of target eDNA from Delphinus delphis and its two associated subspecies. We used a highly replicated sampling effort in the California Current Ecosystem to assess (1) the effect of water depth on the probability of D. delphis eDNA presence, (2) the effects of filtered water volume and filtration method on the likelihood of capturing D. delphis eDNA given its presence and (3) the effect of primer choice on the probability of D. delphis eDNA detection given its capture. Our results show that D. delphis eDNA is most likely to occur in surface (0-10 m) waters, and that large volume (>4 L) sampling increased the probability of capturing D. delphis eDNA on a filter. Moreover, the use of primers designed for cetaceans increased detection probabilities relative to more general primers designed for vertebrates. Our intensive replication in field sampling and laboratory assays and subsequent hierarchical modelling indicates that D. delphis eDNA was present in 90% of surface samples. Yet, since it makes up a tiny fraction of the ambient DNA pool, reliable detection requires optimized experimental design at both the sampling and processing stages. The modelling approach applied here provides an experimental design framework for eDNA studies focused on capturing rare targets. We use the results of this application to provide specific guidance on optimal experimental design for detection of delphinid eDNA.
Trait-based ecology increasingly relies on measures of intraspecific trait variability to estimate how trait diversity within species varies across species and environmental gradients. Yet variance and related metrics are often interpreted without considering their intrinsic dependence on the mean, as described by the empirical Taylor's power law. Several approaches have been considered to remove the effect of trait mean, for which we hypothesize that their effectiveness will depend on the skewness in traits. Here we used a hierarchical plant trait dataset with intensive sampling campaign (>5500 individuals), including three key functional traits (leaf area, specific leaf area and flower size), to test mean-variance scaling across scales and regression methods, from mean and variance within individuals to across species within communities. We focussed on alternative variability metrics: variance and standard error, coefficient of variation metrics (CVs), slope of Taylor's power law. We assessed the relative influence of trait mean on trait variability metrics against predictors associated with environmental and biotic gradients. Results show a consistent mean-variance scaling across traits, scales and regression methods. Coherent with expectations for skewed distributions, log transformation already substantially reduced mean-variance dependence, but its effectiveness depends on the variability metric used. CVs on untransformed data and variance of log-transformed traits showed the lowest dependence on the mean. Our results show that the mean-variance scaling of the Taylor's power law controls patterns of intraspecific trait variability and should be taken into account when assessing trait diversity within species. However, results also highlight that the choice of variability metrics and transformation may help strengthening ecological conclusions.
Tree crowns are complex, three-dimensional structures whose morphology varies among species, individuals and environments. Although light detection and ranging (LiDAR) provides high-resolution, single-tree point clouds that advance species discrimination and the assessment of intraspecific variation in situ, crown shape is still commonly reduced to low-dimensional metrics (e.g. crown diameter or crown base height), losing much of its three-dimensional geometric complexity. We introduce a fully 3D geometric morphometric framework that captures crown shape directly from LiDAR point clouds at both species and individual levels. Pre-segmented LiDAR single-tree point clouds of eight temperate forest species were converted into three-dimensional shape representations using radial bounding volumes (RBVs), which partitioned each crown into a standardized set of vertical layers and radial sectors. Surface points automatically digitized from each RBV formed geospatially aligned, 3D pseudolandmark configurations representing geometric morphometric crown shapes. These configurations served as the input data for multivariate analyses of crown shape variation within and between species. Twelve structural traits, including crown and stem dimensions, were extracted from the same RBVs and integrated into analyses of trait-shape associations. The morphospace of crown shape was structured along different axes of variation in broadleaf species than in conifers. Within these groups, species pairs-such as Fagus versus Quercus and Picea versus Pinus-exhibited contrasting intraspecific morphological gradients, with different structural traits driving shape variation in each. Crown base height and total crown height emerged as the strongest predictors of crown shape. Differences in crown shape among species were primarily captured by symmetric components, with asymmetry providing a negligible signal. Interspecific differentiation was largely driven by architectural variation rather than pure size differences. Morphological differences derived from pseudolandmarks and convolutional neural network features exhibited stronger correlations in conifers than in broadleaf species. We present a reproducible, LiDAR-native framework for quantifying and comparing 3D crown morphology within and across species. Using the RBV approach, geospatially aligned pseudolandmarks can be derived from any pre-segmented, single-tree LiDAR point cloud, enabling scalable, multi-regional analyses of intraspecific variability. This framework provides a robust foundation for integrating crown shape into ecological, evolutionary, silvicultural and modelling studies, including assessments of environmental effects and architectural constraints.
Abstract Mechanistic models, expressed as simulations that can be run and visualised, are a powerful way to understand biological and other systems. Building them, however, requires formulating mathematical or rule‐based descriptions and translating them into working code, skills that many ecologists and other disciplinary researchers do not have. Generative AI can now perform much of this translation, making mechanistic modelling accessible to researchers, educators and students previously excluded by its technical demands. But delegating this translation to AI also carries real risks: modelling assumptions can become hidden, plausible but incorrect code can pass unnoticed, and the skills and critical habits that model building once developed may never form. We examine the benefits and risks of using generative AI as a modelling companion at each stage of model development (conceptual specification, formalisation, implementation and exploration), drawing on recent experimental evidence on AI‐assisted learning and on a worked predator–prey example built with AI, with a rule‐based epidemic model as a parallel case. At each stage we identify where consequential decisions are silently delegated to the AI and set out practical mitigations that keep human judgement engaged without requiring the technical skills AI has made optional: specifying the conceptual model explicitly before engaging AI, having the AI translate its formalisation back into plain language, testing the implementation against outcomes known from domain knowledge and designing the interface to support interrogation. This new opportunity is substantial: researchers, educators and students who understand their systems but have never been able to build models of them are now able to explore these systems mechanistically, testing how components interact and how the system responds to conditions it has not yet experienced, questions that correlative approaches alone cannot answer. Used carelessly, AI risks eroding modelling skill for little lasting benefit, but used deliberately, in the ways set out here, it opens a new pathway to mechanistic understanding without removing the critical thinking the old technical barrier once enforced.
