Speeding up taxonomy in the digital age: A deep learning approach for identifying cryptic freshwater snails
6 affiliations
- 1Department of Computer Science, Goethe University Frankfurt, Frankfurt am Main, Germany
- 2The Hessian Center for Artificial Intelligence (hessian.AI), Darmstadt, Germany
- 3Museo Nacional de Ciencias Naturales CSIC, Madrid, Spain
- 4SNSB – Bavarian State Collection for Palaeontology and Geology, Munich, Germany
- 5Department of Earth and Environmental Sciences, Palaeontology & Geobiology, Ludwig-Maximilians-Universität München, Munich, Germany
- 6Department of Animal Ecology & Systematics, Justus Liebig University, Giessen, Germany
Abstract
Cryptic species complexes pose fundamental challenges to biologists, as species exhibit minimal morphological differences that require integrating morphology, genetics, and biogeography for identification. Here, we present a deep learning approach to support species identification in the freshwater snail genus Radomaniola (Hydrobiidae), a morphologically cryptic group from the Balkans. Our approach mirrors the integrative workflow of expert taxonomists by combining shell images, morphometric measurements, and collection‑site metadata, with optional phylogenetic information. Despite being trained on fewer than 700 specimens across 20 visually similar species with strongly imbalanced class sizes, the system achieved high identification performance. Careful control of spurious correlations, such as those arising from site‑specific imaging conditions or overly precise geographic metadata, was essential to ensure that the network learned biologically meaningful features. Across all experiments, integrating multiple data types and jointly optimizing meaningful embeddings and classification consistently improved performance over image‑only and classification‑only baselines. On specimens from collection sites seen during training we achieved a macro-averaged F1 score of 0.93. Even though this dropped as low as 0.14 when evaluating on specimens from previously unsampled localities, it could be rapidly recovered by retraining with 2–3 newly labeled specimens. Additionally, model top-3 accuracy stayed consistently above 80% in all settings. These results show that relatively lightweight deep learning models can provide practical decision support in real taxonomic workflows. Author summary: Taxonomy, the identification and classification of species, is a time‑consuming task that relies heavily on scarce expert knowledge. In this study, we use Siamese neural networks to support the classification of freshwater snails of the genus Radomaniola, a group in which species are notoriously difficult to distinguish based on shell morphology alone. Unlike many machine learning applications that depend on thousands of labeled examples, our system learns effectively from fewer than 700 specimens distributed across 20 visually similar species with very uneven sample sizes. This is achieved by combining high‑quality shell photographs with morphometric measurements and collection‑site information, and by explicitly reducing confounding factors such as site‑specific backgrounds and imaging conditions. We further test the system in realistic scenarios, including cases where specimens originate from previously unsampled sites and where expert labels become available only gradually. These experiments show that the method can provide useful candidate species suggestions and quickly improve as new labeled specimens are added. Overall, our work illustrates how deep learning can be integrated into taxonomic workflows as a practical aid, while leaving final decisions and species delimitation to human experts.
Related
- Standardizing freshwater macroinvertebrate taxonomy: a Lithuanian operational taxa list for ecological quality assessments and biodiversity research.
- Fitting dynamics is not identifying causal edges: a white-box masked ODE benchmark for trans-omics digital twins of drug action mechanisms
- Development of Taksonify Mobile Application for Learning Biological Taxonomy with Gamification and Computer Vision
- An explainable machine learning model based on multidimensional task-state electroencephalography for identifying attention-deficit/hyperactivity disorder comorbid developmental dyslexia
- Identifying facilitators of and barriers to digital health literacy in pediatric rheumatic disease: A scoping review
BibTeX
@article{vetter2026speeding,
title = {{Speeding up taxonomy in the digital age: A deep learning approach for identifying cryptic freshwater snails}},
author = {Dennis Vetter and Muhammad Ahsan and Diana Delicado and Thomas A. Neubauer and Thomas Wilke and Gemma Roig},
journal = {PLOS Computational Biology},
year = {2026},
month = sep,
doi = {10.1371/journal.pcbi.1014733},
url = {https://journals.plos.org/ploscompbiol/article?id=10.1371/journal.pcbi.1014733},
note = {CC BY 4.0},
}
RIS
TY - JOUR
TI - Speeding up taxonomy in the digital age: A deep learning approach for identifying cryptic freshwater snails
AU - Dennis Vetter
AU - Muhammad Ahsan
AU - Diana Delicado
AU - Thomas A. Neubauer
AU - Thomas Wilke
AU - Gemma Roig
PY - 2026
DA - 2026/09/02
JO - PLOS Computational Biology
DO - 10.1371/journal.pcbi.1014733
UR - https://journals.plos.org/ploscompbiol/article?id=10.1371/journal.pcbi.1014733
AB - Cryptic species complexes pose fundamental challenges to biologists, as species exhibit minimal morphological differences that require integrating morphology, genetics, and biogeography for identification. Here, we present a deep learning approach to support species identification in the freshwater snail genus Radomaniola (Hydrobiidae), a morphologically cryptic group from the Balkans. Our approach mirrors the integrative workflow of expert taxonomists by combining shell images, morphometric measurements, and collection‑site metadata, with optional phylogenetic information. Despite being trained on fewer than 700 specimens across 20 visually similar species with strongly imbalanced class sizes, the system achieved high identification performance. Careful control of spurious correlations, such as those arising from site‑specific imaging conditions or overly precise geographic metadata, was essential to ensure that the network learned biologically meaningful features. Across all experiments, integrating multiple data types and jointly optimizing meaningful embeddings and classification consistently improved performance over image‑only and classification‑only baselines. On specimens from collection sites seen during training we achieved a macro-averaged F1 score of 0.93. Even though this dropped as low as 0.14 when evaluating on specimens from previously unsampled localities, it could be rapidly recovered by retraining with 2–3 newly labeled specimens. Additionally, model top-3 accuracy stayed consistently above 80% in all settings. These results show that relatively lightweight deep learning models can provide practical decision support in real taxonomic workflows. Author summary: Taxonomy, the identification and classification of species, is a time‑consuming task that relies heavily on scarce expert knowledge. In this study, we use Siamese neural networks to support the classification of freshwater snails of the genus Radomaniola, a group in which species are notoriously difficult to distinguish based on shell morphology alone. Unlike many machine learning applications that depend on thousands of labeled examples, our system learns effectively from fewer than 700 specimens distributed across 20 visually similar species with very uneven sample sizes. This is achieved by combining high‑quality shell photographs with morphometric measurements and collection‑site information, and by explicitly reducing confounding factors such as site‑specific backgrounds and imaging conditions. We further test the system in realistic scenarios, including cases where specimens originate from previously unsampled sites and where expert labels become available only gradually. These experiments show that the method can provide useful candidate species suggestions and quickly improve as new labeled specimens are added. Overall, our work illustrates how deep learning can be integrated into taxonomic workflows as a practical aid, while leaving final decisions and species delimitation to human experts.
SN - 1553-7358
ER -