Biology subjects

Sievers, M.

Publications and source records attributed to Sievers, M..

9 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

Shedding light on day-night habitat use of oyster reefs by fish

Oyster reefs provide key ecosystem functions, including the provision of habitat for fish. While their importance in supporting fish assemblages is widely accepted, an assessment of intertidal oyster reefs at night remains a knowledge gap. Documenting habitat use at night could reveal new species reliant on these systems and help uncover diel fish movements among structured and unstructured habitats. We hypothesise that habitat use will interact with time and habitat, whereby differences in fish assemblages between oyster reef and unstructured habitats would be greater during the day. In this study, we used infrared-capable remote underwater video stations (RUVS) to sample paired intertidal oyster reefs and unstructured habitat during day and night. Assemblages using these habitats during the day do not match those at night, consistent with diel shifts observed in similar habitats. Consistent with our hypothesis, during the day oyster reefs had higher diversity and richness than unstructured habitat, whereas both habitats had low diversity and richness at night. The high relative abundance of piscivores in oyster reefs during the day suggests reefs create food webs that extend beyond benthic matter and sustain higher trophic levels. Nighttime abundances and diversity in oyster reefs dropped to levels similar to those of unstructured habitats during the day and night, likely a product of temporal niche partitioning and dynamic predation risks. Nighttime studies were important in documenting a wider suite of species and can thus provide a more complete understanding of fish-habitat interactions. Furthermore, this study used novel methods that enabled depth and size estimation from a single camera unit. While still in its infancy, this can serve as a force multiplier in monitoring intertidal habitats and lead towards a cost-effective method for standardising day/night abundance measurements and obtaining fish size estimates.

ecology↗

bioRxiv2025

Optimizing seagrass planting arrangements for animal benefits in a multi-habitat restoration seascape

Restoring lost and degraded ecosystems to enhance biodiversity and ecosystem services is a global priority, and animal responses to the restoration of habitats are a critical but undervalued component. Identifying the key drivers of animal colonization in restored habitats provides critical insights for restoration practitioners seeking to maximize ecological outcomes. When integrated into predictive frameworks and spatial decision- support tools, this knowledge becomes especially valuable for strategic planning, particularly in complex multi-habitat restoration projects where spatial configuration remains a crucial yet understudied dimension influencing ecosystem recovery trajectories. We collect and analyze animal data from one of the worlds largest multi- habitat coastal restoration systems in Denmark, comprising restored seagrass (Zostera marina), boulder reefs and mussel reefs. Using fine-scale spatial patterns in population abundances, we develop spatially explicit predictions across the seascape for various seagrass restoration scenarios and produce a series of optimizations, showing that it is practical to configure restoration to optimize biodiversity objectives, including those linked with fished species. Species-specific responses translated to variable outcomes across restoration scenarios and optimizations. While the optimal number and arrangement of restored patches varied depending on the target species or species group (e.g., fisheries species or seagrass specialists), one near-ubiquitous arrangement was patchy seagrass planting. This aligns with current practice, maximizes restoration efficiency, and highlights the importance of not homogenizing seascapes for biodiversity. Our approach provides a practical framework for incorporating animal monitoring data into restoration planning, helping practitioners design and optimize spatial planting configurations to achieve specific ecological objectives. Open Research StatementAll data and code/scripts (R language), including a README file, are freely available at: https://github.com/msievers100/DenmarkSpatial

ecology↗

bioRxiv2024

Connectomic reconstruction of a cortical column

The cerebral cortex of mammals has long been proposed to comprise unit-modules, so-called cortical columns. The detailed synaptic-level circuitry of such a neuronal network of about 104 neurons is still unknown. Here, using 3-dimensional electron microscopy, AI-based image processing and automated proofreading, we report the connectomic reconstruction of a defined cortical column in mouse barrel cortex. The cortical column appears as a structural feature in the connectome, without need for geometrical or morphological landmarks. We then used the connectome for definition of neuronal cell types in the column, to determine intracolumnar circuit modules, analyze the logic of inhibitory circuits, investigate the circuits for combination of bottom-up and top-down signals in the column and the specificity of bottom-up and top-down cortical input, search for higher-order circuit structure within homogeneous neuronal populations, and estimate the degree and symmetry of Hebbian learning in the various connection types. With this, we provide a first column-level connectomic description of the cerebral cortex, the likely substrate for a synaptic-level mechanistic understanding of sensory-conceptual integration and learning.

neuroscience↗

bioRxiv2022

RoboEM: automated 3D flight tracing for synaptic-resolution connectomics

Mapping neuronal networks from 3-dimensional electron microscopy data still poses substantial reconstruction challenges, in particular for thin axons. Currently available automated image segmentation methods, while substantially progressed, still require human proof reading for many types of connectomic analyses. RoboEM, an AI-based self-steering 3D flight system trained to navigate along neurites using only EM data as input, substantially improves automated state-of-the-art segmentations and replaces human proof reading for more complex connectomic analysis problems, yielding computational annotation cost for cortical connectomes about 400-fold lower than the cost of manual error correction.

neuroscience↗

bioRxiv2021

From the Modern Synthesis to the Molecular Synthesis: updating how we teach and assess evolution by natural selection

Evolution by natural selection is recognized as both the most important concept in undergraduate biology and the most difficult to teach. Unfortunately, teaching and assessment of evolution have been impaired by legacy approaches that focus on Darwins original insights and the Modern Synthesis integration of Mendelian genetics, but ignore or downplay advances from what we term the Molecular Synthesis. To create better alignment between instructional approaches and contemporary research in the biosciences, we propose that the primary learning goal in teaching evolution should be for students to connect genotypes, phenotypes, and fitness. To support this approach, we developed and tested assessment questions and scoring rubrics called the Extended Assessing Conceptual Reasoning of Natural Selection (E-ACORNS) instrument. Initial E-ACORNS data suggest that after traditional instruction, few students recognize the molecular synthesis--prompting us to propose that introductory course sequences be re-organized with the molecular synthesis as their central theme.

scientific communication and education↗

bioRxiv2020

Future carbon emissions from global mangrove forest loss

Mangroves have among the highest carbon densities of any tropical forest. These "blue carbon" ecosystems can store large amounts of carbon for long periods, and their protection reduces greenhouse gas emissions and supports climate change mitigation. The incorporation of mangroves into Nationally Determined Contributions to the Paris Agreement and their valuation on carbon markets requires predicting how the management of different land-uses can prevent future greenhouse gas emissions and increase CO2 sequestration. Management actions can reduce CO2 emissions and enhance sequestration, but should be guided by predictions of future emissions, not just carbon storage. We project emissions and forgone soil carbon sequestration potential caused by mangrove loss with comprehensive global datasets for carbon stocks, mangrove distribution, deforestation rates, and drivers of land-use change. Emissions from mangrove loss could reach 2,397 Tg CO2eq by the end of the century, or 3,401 Tg CO2eq when considering forgone carbon sequestration. The highest emissions were predicted in southeast and south Asia (West Coral Triangle, Sunda Shelf, and the Bay of Bengal) due to conversion to aquaculture or agriculture, followed by the Caribbean (Tropical Northwest Atlantic) due to clearing and erosion, and the Andaman coast (West Myanmar) and north Brazil due to erosion. Together, these six regions accounted for 90% of the total potential CO2eq future emissions. We highlight hotspots for future emissions and the land-use specfic management actions that could avoid them with appropriate policies and regulation.

biochemistry↗

bioRxiv2020

Deep learning for automated analysis of fish abundance: the benefits of training across multiple habitats

Environmental monitoring guides conservation, and is thus particularly important for coastal aquatic habitats, which are heavily impacted by human activities. Underwater cameras and unmanned devices monitor aquatic wildlife, but manual processing of footage is a significant bottleneck to rapid data processing and dissemination of results. Deep learning has emerged as a solution, but its ability to accurately detect animals across habitat types and locations is largely untested for coastal environments. Here, we produce three deep learning models using an object detection framework to detect an ecologically important fish, luderick (Girella tricuspidata). Two were trained on footage from single habitats (seagrass or reef), and one on footage from both habitats. All models were subjected to tests from both habitat types. Models performed well on test data from the same habitat type (object detection measure: mAP50: 91.7 and 86.9% performance for seagrass and reef, respectively), but poorly on test sets from a different habitat type (73.3 and 58.4%, respectively). The model trained on a combination of both habitats produced the highest object detection results for both tests (92.4 and 87.8%, respectively). Performance in terms of the ability for models to correctly estimate the ecological metric, MaxN, showed similar patterns. The findings demonstrate that deep learning models extract ecologically useful information from video footage accurately and consistently, and can perform across habitat types when trained on footage from the variety of habitat types.

ecology↗

bioRxiv2019

Automating the analysis of fish abundance using object detection: optimising animal ecology with deep learning

Aquatic ecologists routinely count animals to provide critical information for conservation and management. Increased accessibility to underwater recording equipment such as cameras and unmanned underwater devices have allowed footage to be captured efficiently and safely. It has, however, led to immense volumes of data being collected that require manual processing, and thus significant time, labour and money. The use of deep learning to automate image processing has substantial benefits, but has rarely been adopted within the field of aquatic ecology. To test its efficacy and utility, we compared the accuracy and speed of deep learning techniques against human counterparts for quantifying fish abundance in underwater images and video footage. We collected footage of fish assemblages in seagrass meadows in Queensland, Australia. We produced three models using a MaskR-CNN object detection framework to detect the target species, an ecologically important fish, luderick (Girella tricuspidata). Our models were trained on three randomised 80:20 ratios of training:validation data-sets from a total of 6,080 annotations. The computer accurately determined abundance from videos with high performance using unseen footage from the same estuary as the training data (F1 = 92.4%, mAP50 = 92.5%), and from novel footage collected from a different estuary (F1 = 92.3%, mAP50 = 93.4%). The computers performance in determining MaxN was 7.1% better than human marine experts, and 13.4% better than citizen scientists in single image test data-sets, and 1.5% and 7.8% higher in video data-sets, respectively. We show that deep learning is a more accurate tool than humans at determining abundance, and that results are consistent and transferable across survey locations. Deep learning methods provide a faster, cheaper and more accurate alternative to manual data analysis methods currently used to monitor and assess animal abundance. Deep learning techniques have much to offer the field of aquatic ecology.

ecology↗