Sonja Georgievska
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Previous Organizations / Institutes
The organizations / institutes from the scholar's other articles in the past ten years
Research Subjects
Sonja Georgievska has a total of 75 co-authors. Here are the top co-authors by number of shared publications:
- Dario F. Marzella
Radboudumc
5 shared publications - Li C. Xue
Medical BioSciences Department, Radboud University Medical Center, 6525 GA Nijmegen, the Netherlands
5 shared publications ChiaYu Lin
Netherlands eScience Center, Amsterdam, The Netherlands
4 shared publicationsHeleen Severin
Medical BioSciences department, Radboudumc, Radboud University Medical Center, Nijmegen, The Netherlands
4 shared publications- Nicolas Renaud
Netherlands eScience Center, Amsterdam, The Netherlands
4 shared publications Rena Bakhshi
The Netherlands eScience Center, Science Park 140, Amsterdam, The Netherlands
4 shared publications
Latest Publications
- Open Access
- 9 October 2025
- Published by Elsevier in Computers in Biology and Medicine
- Vol. 198 (Pt A) , 111180
MyDigiTwin: A privacy-preserving framework for personalized cardiovascular risk prediction and scenario exploration
- 9 October 2025
- Published by Elsevier in Computers in Biology and Medicine
- Vol. 198 (Pt A) , 111180
Open AccessMyDigiTwin: A privacy-preserving framework for personalized cardiovascular risk prediction and scenario exploration
- Open Access
- 5 July 2025
- Published by Springer Nature in Communications Biology
- Vol. 8 (1) , 1-2
Author Correction: Geometric deep learning improves generalizability of MHC-bound peptide predictions
- 5 July 2025
- Published by Springer Nature in Communications Biology
- Vol. 8 (1) , 1-2
Open AccessAuthor Correction: Geometric deep learning improves generalizability of MHC-bound peptide predictions
- Open Access
A novel approach to identify the fingerprint of stroke gait using deep unsupervised learning
Background The gait pattern results from a complex interaction of several body parts, orchestrated by the (central) nervous system that controls the active and passive systems of the body. An impairment of gait due to a stroke results in a decline in quality of life and independence. Setting up efficient gait training requires an objective and wholesome assessment of the patient’s movement pattern to target individual gait alterations. However, current assessment tools are limited in their ability to capture the complexity of the movement and the amount of data acquired during gait analysis. Aims In this study, we explore the potential of variational autoencoders (VAE) to learn and recognise different gait patterns within both, pathologic and healthy gait. Methods For this purpose, the lower-limb joint angles of 71 participants (29 stroke survivors, 42 healthy controls) were used to train and test a VAE. Results The good reconstruction results (range r = 0.52 - 0.91, average normalized RMSE 23.36 % ± 4.13) indicate that VAEs extract meaningful information from the gait pattern. Furthermore, the extracted latent features are sensitive enough to distinguish between the gait patterns of stroke survivors and a healthy cohort (pConclusions The presented approach allows the assessment of gait data in an objective and wholesome manner, thereby integrating the individual characteristics of each person’s gait, making it a suitable tool for monitoring the progress of rehabilitation efforts.
Open AccessA novel approach to identify the fingerprint of stroke gait using deep unsupervised learning
Background The gait pattern results from a complex interaction of several body parts, orchestrated by the (central) nervous system that controls the active and passive systems of the body. An impairment of gait due to a stroke results in a decline in quality of life and independence. Setting up efficient gait training requires an objective and wholesome assessment of the patient’s movement pattern to target individual gait alterations. However, current assessment tools are limited in their ability to capture the complexity of the movement and the amount of data acquired during gait analysis. Aims In this study, we explore the potential of variational autoencoders (VAE) to learn and recognise different gait patterns within both, pathologic and healthy gait. Methods For this purpose, the lower-limb joint angles of 71 participants (29 stroke survivors, 42 healthy controls) were used to train and test a VAE. Results The good reconstruction results (range r = 0.52 - 0.91, average normalized RMSE 23.36 % ± 4.13) indicate that VAEs extract meaningful information from the gait pattern. Furthermore, the extracted latent features are sensitive enough to distinguish between the gait patterns of stroke survivors and a healthy cohort (pConclusions The presented approach allows the assessment of gait data in an objective and wholesome manner, thereby integrating the individual characteristics of each person’s gait, making it a suitable tool for monitoring the progress of rehabilitation efforts.
- Open Access
- 19 December 2024
- Published by Springer Nature in Communications Biology
- Vol. 7 (1) , 1-12
Geometric deep learning improves generalizability of MHC-bound peptide predictions
The interaction between peptides and major histocompatibility complex (MHC) molecules is pivotal in autoimmunity, pathogen recognition and tumor immunity. Recent advances in cancer immunotherapies demand for more accurate computational prediction of MHC-bound peptides. We address the generalizability challenge of MHC-bound peptide predictions, revealing limitations in current sequence-based approaches. Our structure-based methods leveraging geometric deep learning (GDL) demonstrate promising improvement in generalizability across unseen MHC alleles. Further, we tackle data efficiency by introducing a self-supervised learning approach on structures (3D-SSL). Without being exposed to any binding affinity data, our 3D-SSL outperforms sequence-based methods trained on ~90 times more data points. Finally, we demonstrate the resilience of structure-based GDL methods to biases in binding data on an Hepatitis B virus vaccine immunopeptidomics case study. This proof-of-concept study highlights structure-based methods’ potential to enhance generalizability and data efficiency, with possible implications for data-intensive fields like T-cell receptor specificity predictions.
- 19 December 2024
- Published by Springer Nature in Communications Biology
- Vol. 7 (1) , 1-12
Open AccessGeometric deep learning improves generalizability of MHC-bound peptide predictions
The interaction between peptides and major histocompatibility complex (MHC) molecules is pivotal in autoimmunity, pathogen recognition and tumor immunity. Recent advances in cancer immunotherapies demand for more accurate computational prediction of MHC-bound peptides. We address the generalizability challenge of MHC-bound peptide predictions, revealing limitations in current sequence-based approaches. Our structure-based methods leveraging geometric deep learning (GDL) demonstrate promising improvement in generalizability across unseen MHC alleles. Further, we tackle data efficiency by introducing a self-supervised learning approach on structures (3D-SSL). Without being exposed to any binding affinity data, our 3D-SSL outperforms sequence-based methods trained on ~90 times more data points. Finally, we demonstrate the resilience of structure-based GDL methods to biases in binding data on an Hepatitis B virus vaccine immunopeptidomics case study. This proof-of-concept study highlights structure-based methods’ potential to enhance generalizability and data efficiency, with possible implications for data-intensive fields like T-cell receptor specificity predictions.
- Open Access
- 7 November 2024
- Published by Copernicus GmbH in The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences
- Vol. XLVIII-3-2, 387-392
De Olho na Mata: monitoring Atlantic Forests with drones and few-shot learning
The expansion of invasive species is a global challenge that leads to the loss of biodiversity habitat, and there are few tools to control it. In São Paulo, identification of invasive species is done through field inspections, in parts of Conservation Units and parks, making it difficult to map all tree individuals for adequate management and coping strategies. This manuscript presents a workflow that combines Unmanned Aerial Vehicles (UAVs), or drones, with Artificial Intelligence (AI) to accurately map invasive species in the Atlantic Forest. It describes best practices on how to conduct drone flights to map the forests, exponentially expanding the range of identification and efficiency in invasive tree species management. It also presents an AI workflow that uses few-shot learning and Explainable AI techniques (to guarantee transparency and understanding of the decisions made by the algorithms). Preliminary results indicate that the method obtains acceptable results in the range of 70 percent accuracy for Archontophoenix cunninghamiana (popular name: Seafórtia), an invasive Australian palm.
- 7 November 2024
- Published by Copernicus GmbH in The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences
- Vol. XLVIII-3-2, 387-392
Open AccessDe Olho na Mata: monitoring Atlantic Forests with drones and few-shot learning
The expansion of invasive species is a global challenge that leads to the loss of biodiversity habitat, and there are few tools to control it. In São Paulo, identification of invasive species is done through field inspections, in parts of Conservation Units and parks, making it difficult to map all tree individuals for adequate management and coping strategies. This manuscript presents a workflow that combines Unmanned Aerial Vehicles (UAVs), or drones, with Artificial Intelligence (AI) to accurately map invasive species in the Atlantic Forest. It describes best practices on how to conduct drone flights to map the forests, exponentially expanding the range of identification and efficiency in invasive tree species management. It also presents an AI workflow that uses few-shot learning and Explainable AI techniques (to guarantee transparency and understanding of the decisions made by the algorithms). Preliminary results indicate that the method obtains acceptable results in the range of 70 percent accuracy for Archontophoenix cunninghamiana (popular name: Seafórtia), an invasive Australian palm.
- Open Access
sv-channels: filtering genomic deletions using one-dimensional convolutional neural networks
Structural variant (SV) detection in human genomes using short-read sequencing data is hindered by false positives, arising from sequencing and mapping artifacts that mimic genuine SV signals. Despite advances, state-of-the-art SV callers like GRIDSS and Manta exhibit trade-offs between precision and recall, with GRIDSS offering the highest precision and Manta excelling in recall. To address these limitations, we introduce sv-channels, a novel deep learning model designed to improve the precision of SV detection by leveraging read information at call sites. Our method effectively reduces false positives in Manta’s deletion callsets, achieving precision that surpasses GRIDSS while maintaining a recall rate comparable to Manta. This represents a significant improvement in SV detection, leveraging Manta’s high recall through deep learning and paving the way for more accurate genomic analyses. The sv-channels codebase is openly accessible on GitHub at https://github.com/GooglingTheCancerGenome/sv-channels enabling further research and application in the field.
Open Accesssv-channels: filtering genomic deletions using one-dimensional convolutional neural networks
Structural variant (SV) detection in human genomes using short-read sequencing data is hindered by false positives, arising from sequencing and mapping artifacts that mimic genuine SV signals. Despite advances, state-of-the-art SV callers like GRIDSS and Manta exhibit trade-offs between precision and recall, with GRIDSS offering the highest precision and Manta excelling in recall. To address these limitations, we introduce sv-channels, a novel deep learning model designed to improve the precision of SV detection by leveraging read information at call sites. Our method effectively reduces false positives in Manta’s deletion callsets, achieving precision that surpasses GRIDSS while maintaining a recall rate comparable to Manta. This represents a significant improvement in SV detection, leveraging Manta’s high recall through deep learning and paving the way for more accurate genomic analyses. The sv-channels codebase is openly accessible on GitHub at https://github.com/GooglingTheCancerGenome/sv-channels enabling further research and application in the field.
- Open Access
- 22 August 2024
- Published by The Open Journal in Journal of Open Source Software
- Vol. 9 (100) , 6591
MOTrainer: Distributed Measurement Operator Trainer for Data Assimilation Applications
- 22 August 2024
- Published by The Open Journal in Journal of Open Source Software
- Vol. 9 (100) , 6591
Open AccessMOTrainer: Distributed Measurement Operator Trainer for Data Assimilation Applications
- Open Access
- 26 March 2024
- Published by Springer Nature in Research Square
Improving generalizability and data efficiency for MHC-I binding peptide predictions through structure-based geometric deep learning
The interaction between peptides and major histocompatibility complex (MHC) molecules is pivotal for tissue transplantation, pathogen recognition and autoimmune disease treatments. Recent advances in cancer immunotherapies demand for more accurate computational prediction of MHC-bound peptides. We address the generalizability challenge of MHC-bound peptide predictions, revealing limitations in current sequence-based approaches. Our solution employs structure-based methods leveraging geometric deep learning (GDL), yielding up to 8% improvement in generalizability across unseen MHC alleles. We tackle data efficiency by introducing a self-supervised learning approach surpassing sequence-based methods, even without being trained on binding affinity data. Finally, we demonstrate the resilience of structure-based GDL methods to biases in binding data on an Hepatitis B virus vaccine design case study. This study highlights structure-based methods’ potential to enhance generalizability and data efficiency, with implications for data-intensive fields like T-cell receptor specificity predictions, paving the way for enhanced comprehension and manipulation of immune responses.
- 26 March 2024
- Published by Springer Nature in Research Square
Open AccessImproving generalizability and data efficiency for MHC-I binding peptide predictions through structure-based geometric deep learning
The interaction between peptides and major histocompatibility complex (MHC) molecules is pivotal for tissue transplantation, pathogen recognition and autoimmune disease treatments. Recent advances in cancer immunotherapies demand for more accurate computational prediction of MHC-bound peptides. We address the generalizability challenge of MHC-bound peptide predictions, revealing limitations in current sequence-based approaches. Our solution employs structure-based methods leveraging geometric deep learning (GDL), yielding up to 8% improvement in generalizability across unseen MHC alleles. We tackle data efficiency by introducing a self-supervised learning approach surpassing sequence-based methods, even without being trained on binding affinity data. Finally, we demonstrate the resilience of structure-based GDL methods to biases in binding data on an Hepatitis B virus vaccine design case study. This study highlights structure-based methods’ potential to enhance generalizability and data efficiency, with implications for data-intensive fields like T-cell receptor specificity predictions, paving the way for enhanced comprehension and manipulation of immune responses.
- Open Access
Improving generalizability for MHC-I binding peptide predictions through geometric deep learning
Understanding the peptide presentation mechanism of Major Histocompatibility Complex (MHC) is crucial to study the recognition of pathogens, the treatment of autoimmune diseases and in developing cancer immunotherapies. To comprehend these mechanisms and design effective therapies at affordable costs, it is fundamental to be able to precisely predict in silico which peptides can be bound by each MHC. Many computational approaches have been developed to identify MHC-binding peptides based on their sequences, however, they present biases and their performance on less-studied MHC alleles with limited experimental binding data is severely restricted by their sequence-based (SeqB) nature. We reason that structure-based (StrB) methods could learn the physico-chemical and geometrical rules of peptide-MHC (pMHC) complexes conserved among the alleles, thus generalizing better. We designed three supervised StrB geometric deep learning (GDL) methods, showing superior generalization across all StrB methods over two SeqB methods when the test data is distant from the training data. To further explore data efficiency, we designed a novel self-supervised GDL approach trained only on ∼1K x-ray structures, 3D-SSL, which predicts binding affinities without seeing any during training. Finally, we showcase our StrB method on a case study from the HPV vaccine design to show how StrB methods are robust to biases in the binding data. These findings highlight the potentials of structure-based methods in efficiently utilizing limited data, enhancing generalizability. This study bears important implications on data hungry fields, for example, the long-standing T cell receptor (TCR) specificity challenge.
Open AccessImproving generalizability for MHC-I binding peptide predictions through geometric deep learning
Understanding the peptide presentation mechanism of Major Histocompatibility Complex (MHC) is crucial to study the recognition of pathogens, the treatment of autoimmune diseases and in developing cancer immunotherapies. To comprehend these mechanisms and design effective therapies at affordable costs, it is fundamental to be able to precisely predict in silico which peptides can be bound by each MHC. Many computational approaches have been developed to identify MHC-binding peptides based on their sequences, however, they present biases and their performance on less-studied MHC alleles with limited experimental binding data is severely restricted by their sequence-based (SeqB) nature. We reason that structure-based (StrB) methods could learn the physico-chemical and geometrical rules of peptide-MHC (pMHC) complexes conserved among the alleles, thus generalizing better. We designed three supervised StrB geometric deep learning (GDL) methods, showing superior generalization across all StrB methods over two SeqB methods when the test data is distant from the training data. To further explore data efficiency, we designed a novel self-supervised GDL approach trained only on ∼1K x-ray structures, 3D-SSL, which predicts binding affinities without seeing any during training. Finally, we showcase our StrB method on a case study from the HPV vaccine design to show how StrB methods are robust to biases in the binding data. These findings highlight the potentials of structure-based methods in efficiently utilizing limited data, enhancing generalizability. This study bears important implications on data hungry fields, for example, the long-standing T cell receptor (TCR) specificity challenge.
- Open Access
- 3 December 2021
- Published by Springer Nature in Nature Communications
- Vol. 12 (1) , 1-8
DeepRank: a deep learning framework for data mining 3D protein-protein interfaces
Three-dimensional (3D) structures of protein complexes provide fundamental information to decipher biological processes at the molecular scale. The vast amount of experimentally and computationally resolved protein-protein interfaces (PPIs) offers the possibility of training deep learning models to aid the predictions of their biological relevance. We present here DeepRank, a general, configurable deep learning framework for data mining PPIs using 3D convolutional neural networks (CNNs). DeepRank maps features of PPIs onto 3D grids and trains a user-specified CNN on these 3D grids. DeepRank allows for efficient training of 3D CNNs with data sets containing millions of PPIs and supports both classification and regression. We demonstrate the performance of DeepRank on two distinct challenges: The classification of biological versus crystallographic PPIs, and the ranking of docking models. For both problems DeepRank is competitive with, or outperforms, state-of-the-art methods, demonstrating the versatility of the framework for research in structural biology.
- 3 December 2021
- Published by Springer Nature in Nature Communications
- Vol. 12 (1) , 1-8
Open AccessDeepRank: a deep learning framework for data mining 3D protein-protein interfaces
Three-dimensional (3D) structures of protein complexes provide fundamental information to decipher biological processes at the molecular scale. The vast amount of experimentally and computationally resolved protein-protein interfaces (PPIs) offers the possibility of training deep learning models to aid the predictions of their biological relevance. We present here DeepRank, a general, configurable deep learning framework for data mining PPIs using 3D convolutional neural networks (CNNs). DeepRank maps features of PPIs onto 3D grids and trains a user-specified CNN on these 3D grids. DeepRank allows for efficient training of 3D CNNs with data sets containing millions of PPIs and supports both classification and regression. We demonstrate the performance of DeepRank on two distinct challenges: The classification of biological versus crystallographic PPIs, and the ranking of docking models. For both problems DeepRank is competitive with, or outperforms, state-of-the-art methods, demonstrating the versatility of the framework for research in structural biology.