Biology subjects

Spillias, S.

Publications and source records attributed to Spillias, S..

3 recordsLinked to original sources

bioRxiv2026

Operationalising LLM-assisted screening of literature to support systematic reviews

Large language models (LLMs) can ease the work of screening titles and abstracts for systematic reviews, but obtaining reliable results requires researchers to make practical choices about which LLMs to use, how to combine their scores into a ranking, and how far down that ranking to read. We aimed to identify a general- purpose workflow that screens accurately, minimises human review effort, and generalises across environmental literature corpora. We ran an ensemble of five open-source LLMs across ten human-annotated systematic reviews from the field of ecology and environmental science spanning 19,155 studies. We then asked: (1) how well an ensemble of LLMs ranks relevant papers above irrelevant ones, and (2) where a human reviewer should stop working down that ranked list. A four-LLM ensemble chosen without any labels came close, on every review, to the best ranking achievable with that review's annotations (mean Average Precision 0.66 versus 0.68). We tested different rules for when to stop human review, finding that a single SAFE stopping rule chosen in advance recovered greater than 95% of relevant records on all ten reviews while requiring a human to screen 56% of the corpus on average. Against the established open-source active-learning tool ASReview, this label-free ranking reached the same recall target at lower human workload on seven of the ten reviews (39% versus 42% of the corpus on average). The paper offers a complete workflow that can be adopted for new, unlabelled reviews, using open-source LLMs small enough to run on a high-end consumer laptop, and we provide it as an open-source R package.

ecology↗

bioRxiv2025

Data-Driven Discovery of Mechanistic Ecosystem Models with LLMs

Ecosystem models are essential for ecosystem management, but their development traditionally requires significant time and expertise, creating bottlenecks in addressing urgent environmental challenges. We present LEMMA (LLM Enabled Mechanistic Modelling for ecosystem Assessment), a framework that programmatically generates and iteratively refines mechanistic ecosystem models by combining large language models (LLMs) for equation synthesis and parameter search, evolutionary algorithms for structural optimization, and Template Model Builder (TMB) for efficient parameter estimation. We critically review LEMMAs ability to recover known ecological relationships through two complementary marine case studies: (1) a nutrient-phytoplankton-zooplankton model, and (2) a Crown-of-Thorns starfish (COTS) model. In the first case, our best models displayed almost perfect recovery of known ecological dynamics while maintaining strong predictive performance across multivariate time-series. In the second case, best LEMMA generated models approached human expert models in terms of their ability to successfully capture COTS outbreak dynamics and demonstrated strong out-of-sample predictive power. LEMMA produces interpretable models with meaningful parameters that capture real biological processes, facilitating scientific insight and potentially accelerating management applications. By dramatically accelerating model development while offering ecological interpretability, LEMMA offers a powerful new tool for addressing urgent ecological challenges in a changing world.

ecology↗

bioRxiv2025

Automated Diet Matrix Construction for Marine Ecosystem Models Using Generative AI

We introduce a proof-of-concept framework, Synthesising Parameters for Ecosystem modelling with LLMs (SPELL), that automates species grouping and diet matrix generation to accelerate food web construction for ecosystem models. SPELL retrieves species lists, classifies them into functional groups, and synthesizes trophic interactions by integrating global biodiversity databases (e.g., FishBase, GLOBI), species interaction repositories, and optionally curated local knowledge using Large Language Models (LLMs). We validate the approach through a marine case study across four Australian regions, achieving high reproducibility in species grouping (>99.7%) and moderate consistency in trophic interactions (51-59%). Comparison with an expert-derived food web for the Great Australian Bight indicates strong but incomplete ecological accuracy: 92.6% of group assignments were at least partially correct and 82% of trophic links were identified. Specialized groups such as benthic organisms, parasites, and taxa with variable feeding strategies remain challenging. These findings highlight the importance of expert review for fine-scale accuracy and suggest SPELL is a generalizable tool for rapid prototyping of trophic structures in marine and potentially non-marine ecosystems. HighlightsO_LILLM-based framework automates species grouping and diet matrix creation with >99.7% consistency C_LIO_LI51-59% of trophic interactions show high stability (stability score > 0.7) across iterations C_LIO_LIIn expert comparison, SPELL achieved 81.6% agreement and 80% of diet differences < 0.2 C_LIO_LILLM-driven synthesis integrates global databases with unstructured local knowledge C_LIO_LIReduces ecosystem model development time from months to hours C_LI

ecology↗