@article{PfeufferBielowWeinetal.2024, author = {Pfeuffer, Julianus and Bielow, Chris and Wein, Samuel and Jeong, Kyowon and Netz, Eugen and Walter, Axel and Alka, Oliver and Nilse, Lars and Colaianni, Pasquale Domenico and McCloskey, Douglas and Kim, Jihyung and Rosenberger, George and Bichmann, Leon and Walzer, Mathias and Veit, Johannes and Boudaud, Bertrand and Bernt, Matthias and Patikas, Nikolaos and Pilz, Matteo and Startek, Michał Piotr and Kutuzova, Svetlana and Heumos, Lukas and Charkow, Joshua and Sing, Justin Cyril and Feroz, Ayesha and Siraj, Arslan and Weisser, Hendrik and Dijkstra, Tjeerd M. H. and Perez-Riverol, Yasset and R{\"o}st, Hannes and Kohlbacher, Oliver and Sachsenberg, Timo}, title = {OpenMS 3 enables reproducible analysis of large-scale mass spectrometry data}, volume = {21}, journal = {Nature Methods}, number = {3}, publisher = {Springer Science and Business Media LLC}, issn = {1548-7091}, doi = {10.1038/s41592-024-02197-7}, pages = {365 -- 367}, year = {2024}, language = {en} } @article{WeimannConrad2024, author = {Weimann, Kuba and Conrad, Tim}, title = {Federated Learning with Deep Neural Networks: A Privacy-Preserving Approach to Enhanced ECG Classification}, volume = {28}, journal = {IEEE Journal of Biomedical and Health Informatics}, number = {11}, doi = {10.1109/JBHI.2024.3427787}, year = {2024}, language = {en} } @misc{HajarolasvadiBaum2024, author = {Hajarolasvadi, Noushin and Baum, Daniel}, title = {Data for Training the DeepOrientation Model: Simulated cryo-ET tomogram patches}, doi = {10.12752/9686}, year = {2024}, abstract = {A major restriction to applying deep learning methods in cryo-electron tomography is the lack of annotated data. Many large learning-based models cannot be applied to these images due to the lack of adequate experimental ground truth. One appealing alternative solution to the time-consuming and expensive experimental data acquisition and annotation is the generation of simulated cryo-ET images. In this context, we exploit a public cryo-ET simulator called PolNet to generate three datasets of two macromolecular structures, namely the ribosomal complex 4v4r and Thermoplasma acidophilum 20S proteasome, 3j9i. We select these two specific particles to test whether our models work for macromolecular structures with and without rotational symmetry. The three datasets contain 50, 150, and 450 tomograms with a voxel size of 10 ̊A, respectively. Here, we publish patches of size 40 × 40 × 40 extracted from the medium-sized dataset with 26,703 samples of 4v4r and 40,671 samples of 3j9i. The original tomograms from which the samples were extracted are of size 500 × 500 × 250. Finally, it should be noted that the currently published test dataset is employed for reporting the results of our paper titled "DeepOrientation: Deep Orientation Estimation of Macromolecules in Cryo-electron tomography" paper.}, language = {en} } @article{MaierWeiserConrad2025, author = {Maier, Kristina and Weiser, Martin and Conrad, Tim}, title = {Hybrid PDE-ODE Models for Efficient Simulation of Infection Spread in Epidemiology}, volume = {481}, journal = {Proceedings of the Royal Society A}, number = {2306}, publisher = {Royal Society}, arxiv = {http://arxiv.org/abs/2405.12938}, doi = {10.1098/rspa.2024.0421}, year = {2025}, abstract = {This paper introduces a novel hybrid model combining Partial Differential Equations (PDEs) and Ordinary Differential Equations (ODEs) to simulate infectious disease dynamics across geographic regions. By leveraging the spatial detail of PDEs and the computational efficiency of ODEs, the model enables rapid evaluation of public health interventions. Applied to synthetic environments and real-world scenarios in Lombardy, Italy, and Berlin, Germany, the model highlights how interactions between PDE and ODE regions affect infection dynamics, especially in high-density areas. Key findings reveal that the placement of model boundaries in densely populated regions can lead to inaccuracies in infection spread, suggesting that boundaries should be positioned in areas of lower population density to better reflect transmission dynamics. Additionally, regions with low population density hinder infection flow, indicating a need for incorporating, e.g., jumps in the model to enhance its predictive capabilities. Results indicate that the hybrid model achieves a balance between computational speed and accuracy, making it a valuable tool for policymakers in real-time decision-making and scenario analysis in epidemiology and potentially in other fields requiring similar modeling approaches.}, language = {en} } @article{WeimannConrad2024, author = {Weimann, Kuba and Conrad, Tim}, title = {FELRec: Efficient Handling of Item Cold-Start With Dynamic Representation Learning in Recommender Systems}, journal = {International Journal of Data Science and Analytics}, number = {2024}, publisher = {Springer Nature}, doi = {10.1007/s41060-024-00635-5}, year = {2024}, language = {en} } @article{AnteghiniGualdiOliva2025, author = {Anteghini, Marco and Gualdi, Francesco and Oliva, Baldo}, title = {How did we get there? AI applications to biological networks and sequences}, volume = {190}, journal = {Computers in Biology and Medicine}, publisher = {Elsevier BV}, issn = {0010-4825}, doi = {10.1016/j.compbiomed.2025.110064}, year = {2025}, abstract = {The rapidly advancing field of artificial intelligence (AI) has transformed numerous scientific domains, including biology, where a vast and complex volume of data is available for analysis. This paper provides a comprehensive overview of the current state of AI-driven methodologies in genomics, proteomics, and systems biology. We discuss how machine learning algorithms, particularly deep learning models, have enhanced the accuracy and efficiency of embedding sequences, motif discovery, and the prediction of gene expression and protein structure. Additionally, we explore the integration of AI in the embedding and analysis of biological networks, including protein-protein interaction networks and multi-layered networks. By leveraging large-scale biological data, AI techniques have enabled unprecedented insights into complex biological processes and disease mechanisms. This work underlines the potential of applying AI to complex biological data, highlighting current applications and suggesting directions for future research to further explore AI in this rapidly evolving field.}, language = {en} } @article{SenguptaZachow2025, author = {Sengupta, Agniva and Zachow, Stefan}, title = {Shape-from-Template with Generalised Camera}, volume = {162}, journal = {Image and Vision Computing}, doi = {10.1016/j.imavis.2025.105579}, year = {2025}, abstract = {This article presents a new method for non-rigidly registering a 3D shape to 2D keypoints observed by a constellation of multiple cameras. Non-rigid registration of a 3D shape to observed 2D keypoints, i.e., Shape-from-Template (SfT), has been widely studied using single images, but SfT with information from multiple-cameras jointly opens new directions for extending the scope of known use-cases such as 3D shape registration in medical imaging and registration from hand-held cameras, to name a few. We represent such multi-camera setup with the generalised camera model; therefore any collection of perspective or orthographic cameras observing any deforming object can be registered. We propose multiple approaches for such SfT: the first approach where the corresponded keypoints lie on a direction vector from a known 3D point in space, the second approach where the corresponded keypoints lie on a direction vector from an unknown 3D point in space but with known orientation w.r.t some local reference frame, and a third approach where, apart from correspondences, the silhouette of the imaged object is also known. Together, these form the first set of solutions to the SfT problem with generalised cameras. The key idea behind SfT with generalised camera is the improved reconstruction accuracy from estimating deformed shape while utilising the additional information from the mutual constraints between multiple views of a deformed object. The correspondence-based approaches are solved with convex programming while the silhouette-based approach is an iterative refinement of the results from the convex solutions. We demonstrate the accuracy of our proposed methods on many synthetic and real data.}, language = {en} } @misc{Bautz2023, type = {Master Thesis}, author = {Bautz, Lisa}, title = {Unsupervised Shape Correspondence Estimation for Anatomical Shapes}, pages = {83}, year = {2023}, abstract = {The concept of shape correspondence describes a relation between two or more shapes of the same class. It often consists of a mapping between points on semantically similar locations of all shapes. One possible application for shape correspondence in medicine is the automatic location of anatomical landmarks. Another popular application is the construction of statistical shape models. These models are an established way to represent geometric variation of anatomical shapes in a compact way. Possible applications range from the generation of shapes and reconstruction tasks to disease classification. This thesis aims to investigate unsupervised methods that can be used to estimate such a correspondence on anatomical shapes. While most methods used in the medical domain focus on classical optimization algorithms to establish correspondence, the broader computer vision domain developed a versatile field of data-driven methods. Recently, the new shape model FlowSSM was introduced, which does not require predefined correspondences for training as it generates them itself. As the performance of the shape model is quite competitive, it is natural to assume that the generated correspondences are of high quality as well. For this reason, we evaluate the quality of the correspondences generated by FlowSSM within this thesis. Furthermore, we modify the method by adding a second loss term that minimizes geodesic distortions. This is done to favor isometric deformations which can lead to better correspondences. We compare the results with two established methods from the medical domain, LDDMM and Meshmonk. Furthermore, we investigate the performance of a fourth method called Neuromoph. This data-driven method comes from the wider computer vision field and was not tested on anatomical data yet. All methods are evaluated with a set of different metrics. This includes metrics to assess the quality of the resulting meshes, a sparse correspondence error on anatomical landmarks, and metrics to measure the quality of the resulting shape models. Furthermore, we test all methods on three datasets with different degrees of geometric variation, namely liver, distal femur and face. We show that FlowSSM produces correspondences with state-of-the-art quality. Moreover, our modification further improved the quality of correspondences at a global level. Nevertheless, there is no clear ranking between all methods, as the results differ between metrics and datasets. Thereby, we can show that there are different qualities to a proper correspondence which are reflected in the different metrics. It is therefore strongly recommendable to choose a correspondence estimation method specifically for the problem at hand.}, language = {en} } @article{FogalliPeresLineBaum2023, author = {Fogalli, Giovani Bressan and Peres Line, S{\´e}rgio Roberto and Baum, Daniel}, title = {Segmentation of tooth enamel microstructure images using classical image processing and U-Net approaches}, volume = {2}, journal = {Frontiers in Imaging}, doi = {10.3389/fimag.2023.1215764}, year = {2023}, abstract = {Tooth enamel is the hardest tissue in human organism, formed by prism layers in regularly alternating directions. These prisms form the Hunter-Schreger Bands (HSB) pattern when under side illumination, which is composed of light and dark stripes resembling fingerprints. We have shown in previous works that HSB pattern is highly variable, seems to be unique for each tooth and can be used as a biometric method for human identification. Since this pattern cannot be acquired with sensors, the HSB region in the digital photograph must be identified and correctly segmented from the rest of the tooth and background. Although these areas can be manually removed, this process is not reliable as excluded areas can vary according to the individual's subjective impression. Therefore, the aim of this work was to develop an algorithm that automatically selects the region of interest (ROI), thus, making the entire biometric process straightforward. We used two different approaches: a classical image processing method which we called anisotropy-based segmentation (ABS) and a machine learning method known as U-Net, a fully convolutional neural network. Both approaches were applied to a set of extracted tooth images. U-Net with some post processing outperformed ABS in the segmentation task with an Intersection Over Union (IOU) of 0.837 against 0.766. Even with a small dataset, U-Net proved to be a potential candidate for fully automated in-mouth application. However, the ABS technique has several parameters which allow a more flexible segmentation with interactive adjustments specific to image properties.}, language = {en} } @article{BaiDengDaietal.2023, author = {Bai, Mingze and Deng, Jingwen and Dai, Chengxin and Pfeuffer, Julianus and Sachsenberg, Timo and Perez-Riverol, Yasset}, title = {LFQ-Based Peptide and Protein Intensity Differential Expression Analysis}, volume = {22}, journal = {J. Proteome Res.}, number = {6}, publisher = {American Chemical Society}, doi = {10.1021/acs.jproteome.2c00812}, pages = {2114 -- 2123}, year = {2023}, abstract = {Testing for significant differences in quantities at the protein level is a common goal of many LFQ-based mass spectrometry proteomics experiments. Starting from a table of protein and/or peptide quantities from a given proteomics quantification software, many tools and R packages exist to perform the final tasks of imputation, summarization, normalization, and statistical testing. To evaluate the effects of packages and settings in their substeps on the final list of significant proteins, we studied several packages on three public data sets with known expected protein fold changes. We found that the results between packages and even across different parameters of the same package can vary significantly. In addition to usability aspects and feature/compatibility lists of different packages, this paper highlights sensitivity and specificity trade-offs that come with specific packages and settings.}, language = {en} } @article{KontouWalterAlkaetal.2023, author = {Kontou, Eftychia E. and Walter, Axel and Alka, Oliver and Pfeuffer, Julianus and Sachsenberg, Timo and Mohite, Omkar and Nuhamunanda, Matin and Kohlbacher, Oliver and Weber, Tilmann}, title = {UmetaFlow: An untargeted metabolomics workflow for high-throughput data processing and analysis}, volume = {15}, journal = {Journal of Cheminformatics}, doi = {10.1186/s13321-023-00724-w}, year = {2023}, abstract = {Metabolomics experiments generate highly complex datasets, which are time and work-intensive, sometimes even error-prone if inspected manually. Therefore, new methods for automated, fast, reproducible, and accurate data processing and dereplication are required. Here, we present UmetaFlow, a computational workflow for untargeted metabolomics that combines algorithms for data pre-processing, spectral matching, molecular formula and structural predictions, and an integration to the GNPS workflows Feature-Based Molecular Networking and Ion Identity Molecular Networking for downstream analysis. UmetaFlow is implemented as a Snakemake workflow, making it easy to use, scalable, and reproducible. For more interactive computing, visualization, as well as development, the workflow is also implemented in Jupyter notebooks using the Python programming language and a set of Python bindings to the OpenMS algorithms (pyOpenMS). Finally, UmetaFlow is also offered as a web-based Graphical User Interface for parameter optimization and processing of smaller-sized datasets. UmetaFlow was validated with in-house LC-MS/MS datasets of actinomycetes producing known secondary metabolites, as well as commercial standards, and it detected all expected features and accurately annotated 76\% of the molecular formulas and 65\% of the structures. As a more generic validation, the publicly available MTBLS733 and MTBLS736 datasets were used for benchmarking, and UmetaFlow detected more than 90\% of all ground truth features and performed exceptionally well in quantification and discriminating marker selection.}, language = {en} } @article{PaltraConrad2024, author = {Paltra, Sydney and Conrad, Tim}, title = {Clinical Effectiveness of Ritonavir-Boosted Nirmatrelvir—A Literature Review}, volume = {92}, journal = {Advances in Respiratory Medicine}, number = {1}, doi = {10.3390/arm92010009}, year = {2024}, abstract = {Nirmatrelvir/Ritonavir is an oral treatment for mild to moderate COVID-19 cases with a high risk for a severe course of the disease. For this paper, a comprehensive literature review was performed, leading to a summary of currently available data on Nirmatrelvir/Ritonavir's ability to reduce the risk of progressing to a severe disease state. Herein, the focus lies on publications that include comparisons between patients receiving Nirmatrelvir/Ritonavir and a control group. The findings can be summarized as follows: Data from the time when the Delta-variant was dominant show that Nirmatrelvir/Ritonavir reduced the risk of hospitalization or death by 88.9\% for unvaccinated, non-hospitalized high-risk individuals. Data from the time when the Omicron variant was dominant found decreased relative risk reductions for various vaccination statuses: between 26\% and 65\% for hospitalization. The presented papers that differentiate between unvaccinated and vaccinated individuals agree that unvaccinated patients benefit more from treatment with Nirmatrelvir/Ritonavir. However, when it comes to the dependency of potential on age and comorbidities, further studies are necessary. From the available data, one can conclude that Nirmatrelvir/Ritonavir cannot substitute vaccinations; however, its low manufacturing cost and easy administration make it a valuable tool in fighting COVID-19, especially for countries with low vaccination rates.}, language = {de} } @phdthesis{Pfeuffer2023, author = {Pfeuffer, Julianus}, title = {Computational Methods for Protein Inference in Shotgun Proteomics Experiments}, year = {2023}, abstract = {Since the beginning of this millennium, the advent of high-throughput methods in numerous fields of the life sciences led to a shift in paradigms. A broad variety of technologies emerged that allow comprehensive quantification of molecules involved in biological processes. Simultaneously, a major increase in data volume has been recorded with these techniques through enhanced instrumentation and other technical advances. By supplying computational methods that automatically process raw data to obtain biological information, the field of bioinformatics plays an increasingly important role in the analysis of the ever-growing mass of data. Computational mass spectrometry in particular, is a bioinformatics field of research which provides means to gather, analyze and visualize data from high-throughput mass spectrometric experiments. For the study of the entirety of proteins in a cell or an environmental sample, even current techniques reach limitations that need to be circumvented by simplifying the samples subjected to the mass spectrometer. These pre-digested (so-called bottom-up) proteomics experiments then pose an even bigger computational burden during analysis since complex ambiguities need to be resolved during protein inference, grouping and quantification. In this thesis, we present several developments in the pursuit of our goal to provide means for a fully automated analysis of complex and large-scale bottom-up proteomics experiments. Firstly, due to prohibitive computational complexities in state-of-the-art Bayesian protein inference techniques, a refined, more stable technique for performing inference on sums of random variables was developed to enable a variation of standard Bayesian inference for the problem. nextflow and part of a set of standardized, well-tested, and community-maintained workflows by the nf-core collective. Our workflow runs on large-scale data with complex experimental designs and allows a one-command analysis of local and publicly available data sets with state-of-the-art accuracy on various high-performance computing environments or the cloud.}, language = {en} } @article{WeimannConrad2025, author = {Weimann, Kuba and Conrad, Tim}, title = {Self-supervised pre-training with joint-embedding predictive architecture boosts ECG classification performance}, volume = {196}, journal = {Computers in Biology and Medicine}, publisher = {Elsevier BV}, issn = {0010-4825}, doi = {10.1016/j.compbiomed.2025.110809}, year = {2025}, abstract = {Accurate diagnosis of heart arrhythmias requires the interpretation of electrocardiograms (ECG), which capture the electrical activity of the heart. Automating this process through machine learning is challenging due to the need for large annotated datasets, which are difficult and costly to collect. To address this issue, transfer learning is often employed, where models are pre-trained on large datasets and fine-tuned for specific ECG classification tasks with limited labeled data. Self-supervised learning has become a widely adopted pre-training method, enabling models to learn meaningful representations from unlabeled datasets. In this work, we explore the joint-embedding predictive architecture (JEPA) for self-supervised learning from ECG data. Unlike invariance-based methods, JEPA does not rely on hand-crafted data augmentations, and unlike generative methods, it predicts latent features rather than reconstructing input data. We create a large unsupervised pre-training dataset by combining ten public ECG databases, amounting to over one million records. We pre-train Vision Transformers using JEPA on this dataset and fine-tune them on various PTB-XL benchmarks. Our results show that JEPA outperforms existing invariance-based and generative approaches, achieving an AUC of 0.945 on the PTB-XL all statements task. JEPA consistently learns the highest quality representations, as demonstrated in frozen evaluations, and proves advantageous for pre-training even in the absence of additional data.}, language = {en} }