Biology subjects

Nolte, K.

Publications and source records attributed to Nolte, K..

6 recordsLinked to original sources

bioRxiv2026

Automated landmark and semilandmark annotation for wing geometric morphometrics in Diptera using deep learning

1. Diptera represent a diverse insect order, including vectors of human and animal pathogens. Their accurate species identification remains a major bottleneck in ecological and epidemiological studies. Morphological identification requires taxonomic expertise, while molecular methods are costly and not universally reliable. Wing geometric morphometrics offers an alternative, but manual landmark annotation is time-consuming and introduces observer bias. 2. We developed ITHILDIN, an automated pipeline for landmark and semilandmark annotation of Diptera wings, combining UNet++ segmentation and an Hourglass landmark prediction model. Using mosquitoes as the primary model system, we extended an existing repository with 5,793 additional images. Models were trained on 5991 annotations of landmarks and segmentations and then evaluated on 12,522 images across 34 taxa. We assessed landmark prediction accuracy against human observers and ML-morph, evaluated species identification using Linear Discriminant Analysis on 17 homologous landmarks and 52 semilandmarks, and tested out-of-distribution generalisation by reproducing an independent study. Transferability was demonstrated by adapting the pipeline to the Dipteran families Drosophilidae and Glossinidae. 3. The Hourglass model achieved a mean landmark error of 4.5 pixels (95% CI: 4.3-4.6), within human observer variability (4.7 pixels, 95% CI: 4.4-5.0) and substantially outperforming ML-Morph (12.7 pixels, 95% CI: 11.1-14.2). The semilandmark-based approach for species identification achieved 91% balanced accuracy across 34 taxa, comparable to CNN performance (94%). On out-of-distribution data, the landmark pipeline generalised substantially better than the CNN and a soft-voting ensemble of the landmark and CNN classifiers achieved 88% balanced accuracy on a replicated study. 4. Combining geometric morphometrics with deep learning provides a reproducible, interpretable, and generalisable alternative to black-box CNN classifiers for Diptera wing analysis. By acting as a consistent single observer comparable to human annotation, the system eliminates inter-observer bias, enabling large-scale and cross-study morphometric analyses of Dipteran wings. The system is publicly available at www.ithildin.bnitm.de and transferable to other Diptera families with moderate retraining effort. Data availabilityImages used in this study are accessible under CC BY 4.0 license at https://doi.org/10.6019/S-BIAD1478. Downloadable and installable docker application can be accessed on the applications git page: https://anonymous.4open.science/r/ITHILDIN-4313/

bioinformatics↗

bioRxiv2026

Accurate identification of invasive Aedes mosquito species using low-cost imaging and geometric wing morphometrics

Accurate species identification is crucial to assess the medical and veterinary relevance of a mosquito specimen, but it requires high experience of the observers and well-equipped laboratories. This study aimed to evaluate whether low-cost imaging in combination with geometric wing morphometrics can provide accurate identification of invasive, morphologically similar Aedes species. The right wings of 670 female specimens covering 184 Ae. aegypti, 156 Ae. albopictus, 166 Ae. j. japonicus and 164 Ae. koreicus, were removed, mounted and photographed with a professional stereomicroscope (Olympus SZ61, Olympus, Tokyo, Japan) and a macro lens (Apexel-24XMH, Apexel, Shenzhen, China) attached to a smartphone. The coordinates of 18 landmarks on the vein crosses were digitalized by a single observer for each image. In addition, the landmarks of 20 specimens per species and imaging device were digitalized by six different observers to assess the degree of the observer error. The superimposed shape variables were used to compare the species classification accuracy of linear discriminant analysis (LDA), support vector machine (SVM), Random Forest (RF), and XGBoost. In the single-observer landmark data, the LDA achieved the best classification results with a mean accuracy of 95 % for landmarks from microscope images and 92 % for those obtained from smartphone images. In the multi-observer landmark data, LDA consistently performed worse than the other three classifiers, and the reduction in the accuracy was more pronounced for smartphone images than for microscope images. This pattern was associated with a higher degree of observer error for smartphone images, as confirmed by a landmark-wise comparison across all landmarks. Geometric wing morphometrics provides a reliable method to distinguish the most common invasive Aedes species in Europe. Thereby, the image quality obtained by smartphones equipped with a macro lens is sufficient and represents a cost-effective alternative to professional microscopes. However, due to the greater degree of observer variation for smartphone images, landmark coordinates for such images should ideally be collected by a single observer.

zoology↗

bioRxiv2025

Potentials and limitations in the application of Convolutional Neural Networks for mosquito species identification using wing images

1. This study addresses the pressing global health burden of mosquito-borne diseases by investigating the application of Convolutional Neural Networks (CNNs) for mosquito species identification using wing images. Conventional identification methods are hampered by the need for significant expertise and resources, while CNNs offer a promising alternative. Our research aimed to develop a reliable and applicable classification system that can be used under real-world conditions, with a focus on improving model adaptability to unencountered devices, mitigating dataset biases, and ensuring usability across different users without standardized protocols. 2. We utilized a large, diverse dataset of mosquito wing images of 21 taxa and three imagecapturing devices and an optimized preprocessing pipeline to standardize images and remove undesirable image features. 3. The developed CNN models demonstrated high performance, with an average balanced accuracy of 98.3% and a macro F1-score of 97.6%, effectively distinguishing between the 21 mosquito taxa, including morphologically similar pairs. The preprocessing pipeline improved the models robustness, reducing performance drops on unfamiliar devices effectively. However, the study also highlights the persistence of inherent dataset biases, which the preprocessing steps could only partially mitigate. The classification systems practical usability was demonstrated through a feasibility study, showing high inter-rater reliability. 4. The results underscore the potential of the proposed workflow to enhance vector surveillance, especially in resource-constrained settings, and suggest its applicability to other winged insect species. The classification system developed in this study is available for public use, providing a valuable tool for vector surveillance and research, supporting efforts to mitigate the spread of mosquito-borne diseases.

bioinformatics↗

bioRxiv2024

Mosquito Wing Image Repository for Advancing Research on Geometric Morphometric- and AI-Based Identification

Accurate identification of mosquito species is essential for effective vector control and mitigation of mosquito-borne disease outbreaks. Traditional morphological identification requires highly specialized personnel and is time-consuming, while molecular techniques can be cost-effective and dependent on comprehensive genetic information. Wing geometric morphometry has emerged as a promising alternative, leveraging detailed geometric measurements of wing shapes and vein patterns to distinguish between species and detect intraspecies variations. This paper presents a curated dataset of 18,104 mosquito wing images, collected from 10,500 mosquito specimens, annotated with extensive meta-information, designed to support research in wing geometric morphometry and the development of machine learning models, ultimately supporting efforts in vector surveillance and research.

ecology↗

bioRxiv2021

Detecting ice artefacts in processed macromolecular diffraction data with machine learning

Contamination with diffraction from ice crystals can negatively affect, or even impede macromolecular structure determination and therefore, detecting the resulting artefacts in diffraction data is crucial. However, once the data have been processed, it can be very difficult to automatically recognize this problem. To address this, a set of convolutional neural networks named Helcaraxe has been developed which can detect ice diffraction artefacts in processed diffraction data from macromolecular crystals. The networks outperform previous algorithms and will be available as part of the AUSPEX webserver and CCP4-distributed software. SynopsisA program utilizing artificial learning and convolutional neural networks, named Helcaraxe, has been developed which can detect ice crystal artefacts in processed macromolecular diffraction data with unprecedented accuracy.

biophysics↗

bioRxiv2020

Making the invisible enemy visible

During the COVID-19 pandemic, structural biologists rushed to solve the structures of the 28 proteins encoded by the SARS-CoV-2 genome in order to understand the viral life cycle and enable structure-based drug design. In addition to the 204 previously solved structures from SARS-CoV-1, 548 structures covering 16 of the SARS-CoV-2 viral proteins have been released in a span of only 6 months. These structural models serve as the basis for research to understand how the virus hijacks human cells, for structure-based drug design, and to aid in the development of vaccines. However, errors often occur in even the most careful structure determination - and may be even more common among these structures, which were solved quickly and under immense pressure. The Coronavirus Structural Task Force has responded to this challenge by rapidly categorizing, evaluating and reviewing all of these experimental protein structures in order to help downstream users and original authors. In addition, the Task Force provided improved models for key structures online, which have been used by Folding@Home, OpenPandemics, the EU JEDI COVID-19 challenge and others.

molecular biology↗