Katerina Taškova
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
Katerina Taškova has a total of 82 co-authors. Here are the top co-authors by number of shared publications:
- Alfonso De Simone
Department of Pharmacy, University of Naples Federico II, Naples, Italy
3 shared publications - Davide Mercadante
School of Chemical Sciences, The University of Auckland
3 shared publications - Kaiqi Zhao
Shenzhen Key Laboratory of Internet Information Collaboration, Harbin Institute of Technology, Shenzhen, Shenzhen, China
3 shared publications - Miguel A. Andrade-Navarro
Johannes Gutenberg University of Mainz
3 shared publications - Patricia Riddle
School of Computer Science, University of Auckland, Auckland, New Zealand
3 shared publications - Peter Korošec
Computer Systems Department, Jožef Stefan Institute, Ljubljana, 1000, Slovenia
3 shared publications
Latest Publications
- Open Access
- 21 October 2026
- Published by National Academy of Sciences in Proceedings of the National Academy of Sciences
- Vol. 123 (43)
Enhancing simulations of disordered protein ensembles with chemical shift restraints in coarse-grained models
With more than one third of the human proteome being structurally disordered, intrinsically disordered proteins (IDPs) are key contributors to complex cellular functions, despite their highly dynamic conformational ensembles remaining elusive to conventional techniques in structural biology. Coarse-grained (CG) simulation methods thus serve as a gold standard to refine IDP ensembles. These methods, when relying on elastic network models or rigid body to sample secondary structure, are however inherently limited in their capability to model and sample transient secondary structure at a residue-level resolution as informed by experimental data. This fundamental limitation persists despite experimental evidence suggesting that transient structural propensities are a fundamental hallmark of IDP function. Through an artificial neural network capable of predicting backbone chemical shifts from C α -only protein conformations, we present a generalized framework which does not rely on dihedral parameterization to equip one bead-per-aminoacid CG force fields with the ability to sample functional elements of secondary structure. When applied on a large set of IDPs in isolation or condensates, our approach correctly recapitulates both local and global arrangements, yielding experimentally informed ensembles that can inform structure-mediated function and resolve force field limitations. This can yield a fruitful exploration of ensemble-to-function relationships in IDPs, allowing to explore phenomena such as secondary structure propensity and nucleation in condensates, conformationally selected binding motifs across disordered ensembles or conformational preferences of sequence-specific short-linear motifs.
- 21 October 2026
- Published by National Academy of Sciences in Proceedings of the National Academy of Sciences
- Vol. 123 (43)
Open AccessEnhancing simulations of disordered protein ensembles with chemical shift restraints in coarse-grained models
With more than one third of the human proteome being structurally disordered, intrinsically disordered proteins (IDPs) are key contributors to complex cellular functions, despite their highly dynamic conformational ensembles remaining elusive to conventional techniques in structural biology. Coarse-grained (CG) simulation methods thus serve as a gold standard to refine IDP ensembles. These methods, when relying on elastic network models or rigid body to sample secondary structure, are however inherently limited in their capability to model and sample transient secondary structure at a residue-level resolution as informed by experimental data. This fundamental limitation persists despite experimental evidence suggesting that transient structural propensities are a fundamental hallmark of IDP function. Through an artificial neural network capable of predicting backbone chemical shifts from C α -only protein conformations, we present a generalized framework which does not rely on dihedral parameterization to equip one bead-per-aminoacid CG force fields with the ability to sample functional elements of secondary structure. When applied on a large set of IDPs in isolation or condensates, our approach correctly recapitulates both local and global arrangements, yielding experimentally informed ensembles that can inform structure-mediated function and resolve force field limitations. This can yield a fruitful exploration of ensemble-to-function relationships in IDPs, allowing to explore phenomena such as secondary structure propensity and nucleation in condensates, conformationally selected binding motifs across disordered ensembles or conformational preferences of sequence-specific short-linear motifs.
- Open Access
- 12 August 2026
- Published by Elsevier in SSRN Electronic Journal
Habibot: A multi-model machine learning approach for rapid classification of temperate rocky reef ecosystems from underwater imagery
Rapid, scalable assessment of temperate rocky reef habitats is essential for detecting ecosystem change and guiding management, yet analysis of underwater imagery remains a major bottleneck due to the cost and limitations of human annotation. We present "Habibot", an open-source, multi-model machine learning pipeline that integrates (i) a frame-level habitat classifier, (ii) a patch-level classifier derived from point-based annotations, and (iii) an explicit sea urchin detector to produce transparent, ecologically grounded habitat assessments. Habibot targets common reef states across northeastern Aotearoa New Zealand and southeastern Australia, including kelp- and seaweed-dominated reefs, urchin barrens, and unconsolidated substrates, and additionally flags two heterogeneous "partial" (mixed) reef states designed to capture early or transitional conditions. Using transfer learning for both frame and patch classifiers and a rule-based fusion stage informed by ecological indicators (frame confidence, patch composition, and sea urchin presence), the system achieves strong per-model performance on the test sets (mean F1 score = 0.87 for frame classification; mean F1 score = 0.88 for patch classification). End-to-end evaluation on an independent set of 295 images assessed by three marine ecologists yielded 90.5% accuracy. We further demonstrate how Habibot outputs support spatially explicit transect mapping and label-agnostic region highlighting for downstream coverage estimation. Habibot provides a practical, interpretable pathway to accelerate habitat classification workflows and enable timely monitoring of kelp-forest change and sea urchin impacts from heterogeneous underwater imagery, with planned deployment through the Squidle+ platform to support scalable ecological assessment workflows.
- 12 August 2026
- Published by Elsevier in SSRN Electronic Journal
Open AccessHabibot: A multi-model machine learning approach for rapid classification of temperate rocky reef ecosystems from underwater imagery
Rapid, scalable assessment of temperate rocky reef habitats is essential for detecting ecosystem change and guiding management, yet analysis of underwater imagery remains a major bottleneck due to the cost and limitations of human annotation. We present "Habibot", an open-source, multi-model machine learning pipeline that integrates (i) a frame-level habitat classifier, (ii) a patch-level classifier derived from point-based annotations, and (iii) an explicit sea urchin detector to produce transparent, ecologically grounded habitat assessments. Habibot targets common reef states across northeastern Aotearoa New Zealand and southeastern Australia, including kelp- and seaweed-dominated reefs, urchin barrens, and unconsolidated substrates, and additionally flags two heterogeneous "partial" (mixed) reef states designed to capture early or transitional conditions. Using transfer learning for both frame and patch classifiers and a rule-based fusion stage informed by ecological indicators (frame confidence, patch composition, and sea urchin presence), the system achieves strong per-model performance on the test sets (mean F1 score = 0.87 for frame classification; mean F1 score = 0.88 for patch classification). End-to-end evaluation on an independent set of 295 images assessed by three marine ecologists yielded 90.5% accuracy. We further demonstrate how Habibot outputs support spatially explicit transect mapping and label-agnostic region highlighting for downstream coverage estimation. Habibot provides a practical, interpretable pathway to accelerate habitat classification workflows and enable timely monitoring of kelp-forest change and sea urchin impacts from heterogeneous underwater imagery, with planned deployment through the Squidle+ platform to support scalable ecological assessment workflows.
- Open Access
- 22 April 2026
- Published by Springer Nature in Journal of Cheminformatics
- Vol. 18 (1)
Predicting enzyme-compound associations for enzyme-catalysed reactions
Enzyme-catalysed reactions are common in many areas, including pharmaceutical metabolism and agricultural chemical biodegradation. Analysing and predicting
- 22 April 2026
- Published by Springer Nature in Journal of Cheminformatics
- Vol. 18 (1)
Open AccessPredicting enzyme-compound associations for enzyme-catalysed reactions
Enzyme-catalysed reactions are common in many areas, including pharmaceutical metabolism and agricultural chemical biodegradation. Analysing and predicting
- Open Access
- 20 April 2026
- Published by Association for Computing Machinery (ACM)
- p. 1251-1262
AnchorGK: Anchor-based Incremental and Stratified Graph Learning Framework for Inductive Spatio-Temporal Kriging
Spatio-temporal kriging is an essential research problem in sensor networks due to the sparsity of deployed sensors. While recent studies consider spatial and temporal correlations, they often overlook the sparse spatial distribution of locations and the incomplete features across locations. To tackle these problems, we propose an Anchor-based Incremental and Stratified Graph Learning Framework for Inductive Spatio-Temporal Kriging (AnchorGK). AnchorGK introduces anchor locations to enable effective data stratification for accurate kriging. Anchor locations are constructed based on feature availability, and strata are subsequently established based on the an- chor locations. This stratification serves two purposes: 1) it ensures that the spatial correlations between unknown areas (no observations) and surrounding known locations are accurately represented and dynamically updated within the graph learning framework, and 2) it facilitates the use of all available features across different strata through a novel incremental representation method. Building on the data stratification, we propose a dual-view graph learning layer that integrates information from relevant features and locations and learns distinct representations for different strata. Finally, kriging is performed based on the obtained strata representations. Experimental results on multiple benchmark datasets demonstrate that AnchorGK consistently outperforms existing state-of-the-art methods. Our codes, datasets, and related materials are given in: https://github.com/xren451/Spatial-interpolation
- 20 April 2026
- Published by Association for Computing Machinery (ACM)
- p. 1251-1262
Open AccessAnchorGK: Anchor-based Incremental and Stratified Graph Learning Framework for Inductive Spatio-Temporal Kriging
Spatio-temporal kriging is an essential research problem in sensor networks due to the sparsity of deployed sensors. While recent studies consider spatial and temporal correlations, they often overlook the sparse spatial distribution of locations and the incomplete features across locations. To tackle these problems, we propose an Anchor-based Incremental and Stratified Graph Learning Framework for Inductive Spatio-Temporal Kriging (AnchorGK). AnchorGK introduces anchor locations to enable effective data stratification for accurate kriging. Anchor locations are constructed based on feature availability, and strata are subsequently established based on the an- chor locations. This stratification serves two purposes: 1) it ensures that the spatial correlations between unknown areas (no observations) and surrounding known locations are accurately represented and dynamically updated within the graph learning framework, and 2) it facilitates the use of all available features across different strata through a novel incremental representation method. Building on the data stratification, we propose a dual-view graph learning layer that integrates information from relevant features and locations and learns distinct representations for different strata. Finally, kriging is performed based on the obtained strata representations. Experimental results on multiple benchmark datasets demonstrate that AnchorGK consistently outperforms existing state-of-the-art methods. Our codes, datasets, and related materials are given in: https://github.com/xren451/Spatial-interpolation
- Open Access
Capturing secondary structure in coarse grained intrinsically disordered proteins with simulations driven by chemical shifts.
A major challenge when investigating intrinsically disordered proteins (IDPs) pertains to understanding how secondary structure formation across otherwise disordered ensembles, relates to function. While coarse-grained (CG) simulation methods are instrumental to describe ensemble-to-function relationships in IDPs, they remain largely incapable of sampling secondary structure. We here present NapshiftCα, an artificial neural network capable of modelling NMR chemical shifts from Cα-mapped CG protein conformations, leveraging its predictions to restrain simulations and yield experimentally-informed IDP ensembles that contain secondary structure. We applied NapshiftCα on a large set of proteins in isolation or condensed phase, and found that, when secondary structure is present, CS incorporation yields improved agreements with smFRET spectroscopy and resolves the over-reliance of CG models on near linear conformers. By incorporating observables that are routinely collected for IDPs but never coupled to CG simulations, NapshiftCα ultimately informs on both local and global conformational features of function-defining IDP ensembles.
Open AccessCapturing secondary structure in coarse grained intrinsically disordered proteins with simulations driven by chemical shifts.
A major challenge when investigating intrinsically disordered proteins (IDPs) pertains to understanding how secondary structure formation across otherwise disordered ensembles, relates to function. While coarse-grained (CG) simulation methods are instrumental to describe ensemble-to-function relationships in IDPs, they remain largely incapable of sampling secondary structure. We here present NapshiftCα, an artificial neural network capable of modelling NMR chemical shifts from Cα-mapped CG protein conformations, leveraging its predictions to restrain simulations and yield experimentally-informed IDP ensembles that contain secondary structure. We applied NapshiftCα on a large set of proteins in isolation or condensed phase, and found that, when secondary structure is present, CS incorporation yields improved agreements with smFRET spectroscopy and resolves the over-reliance of CG models on near linear conformers. By incorporating observables that are routinely collected for IDPs but never coupled to CG simulations, NapshiftCα ultimately informs on both local and global conformational features of function-defining IDP ensembles.
- Open Access
Integrating NMR restraints into coarse-grained simulations: toward accurate conformational ensembles of complex protein systems
Structural dynamics play critical roles for the biological activity of protein molecules. Characterising the inherent conformational landscapes of these macromolecules remains a major experimental and computational challenge, particularly for heterogeneous and transient systems such as intrinsically disordered proteins, membrane-associated assemblies and disordered fuzzy coats of amyloid aggregates. In this context, coarse-grained (CG) molecular dynamics simulations have enabled accessing to extended timescales and large system sizes, however, their reduced resolution and simplified interaction potentials often limit the structural accuracy. Here, we introduce Martini3-NMR, an integrative framework that incorporates nuclear magnetic resonance (NMR) observables directly into CG protein force fields. Using artificial neural networks to model NMR chemical shifts at the CG level, and integrating these data with NOE restraints, we define an approach to significantly enhance the accuracy of CG simulations while maintaining their elevated sampling efficiency, thereby resulting in a substantially improved description of protein conformational ensembles. We demonstrate the broad applicability of Martini3-NMR by generating CG ensembles for a diverse range of systems and complex processes such as protein folding from misfolded states, oligomer disassembly within lipid bilayers and the topology of disordered fuzzy regions at amyloid fibril surfaces, which were found to display condensate-like structural and dynamical properties. By enabling an experimentally driven and computationally efficient exploration of protein conformational landscapes, Martini3-NMR provides a novel general framework for investigating dynamic, heterogeneous and multiscale biomolecular processes. This approach opens to significant new opportunities for extending CG simulations towards a more quantitative understanding of the relationship between molecular structure, dynamics and biological function.
Open AccessIntegrating NMR restraints into coarse-grained simulations: toward accurate conformational ensembles of complex protein systems
Structural dynamics play critical roles for the biological activity of protein molecules. Characterising the inherent conformational landscapes of these macromolecules remains a major experimental and computational challenge, particularly for heterogeneous and transient systems such as intrinsically disordered proteins, membrane-associated assemblies and disordered fuzzy coats of amyloid aggregates. In this context, coarse-grained (CG) molecular dynamics simulations have enabled accessing to extended timescales and large system sizes, however, their reduced resolution and simplified interaction potentials often limit the structural accuracy. Here, we introduce Martini3-NMR, an integrative framework that incorporates nuclear magnetic resonance (NMR) observables directly into CG protein force fields. Using artificial neural networks to model NMR chemical shifts at the CG level, and integrating these data with NOE restraints, we define an approach to significantly enhance the accuracy of CG simulations while maintaining their elevated sampling efficiency, thereby resulting in a substantially improved description of protein conformational ensembles. We demonstrate the broad applicability of Martini3-NMR by generating CG ensembles for a diverse range of systems and complex processes such as protein folding from misfolded states, oligomer disassembly within lipid bilayers and the topology of disordered fuzzy regions at amyloid fibril surfaces, which were found to display condensate-like structural and dynamical properties. By enabling an experimentally driven and computationally efficient exploration of protein conformational landscapes, Martini3-NMR provides a novel general framework for investigating dynamic, heterogeneous and multiscale biomolecular processes. This approach opens to significant new opportunities for extending CG simulations towards a more quantitative understanding of the relationship between molecular structure, dynamics and biological function.
- Open Access
- 25 October 2025
- Published by Elsevier in Marine Environmental Research
- Vol. 213, 107662
Urchinbot: An open-source model for the rapid detection and classification of habitat-modifying sea urchin species
- 25 October 2025
- Published by Elsevier in Marine Environmental Research
- Vol. 213, 107662
Open AccessUrchinbot: An open-source model for the rapid detection and classification of habitat-modifying sea urchin species
- Open Access
- 21 October 2024
- Published by Association for Computing Machinery (ACM)
- p. 1973-1982
Periormer: Periodic Transformer for Seasonal and Irregularly Sampled Time Series
- 21 October 2024
- Published by Association for Computing Machinery (ACM)
- p. 1973-1982
Open AccessPeriormer: Periodic Transformer for Seasonal and Irregularly Sampled Time Series
- Top Cited PapersOpen Access
- 17 September 2024
- Published by Springer Nature in Artificial Intelligence Review
- Vol. 57 (11) , 1-51
A systematic review of aspect-based sentiment analysis: domains, methods, and trends
Aspect-based sentiment analysis (ABSA) is a fine-grained type of sentiment analysis that identifies aspects and their associated opinions from a given text. With the surge of digital opinionated text data, ABSA gained increasing popularity for its ability to mine more detailed and targeted insights. Many review papers on ABSA subtasks and solution methodologies exist, however, few focus on trends over time or systemic issues relating to research application domains, datasets, and solution approaches. To fill the gap, this paper presents a systematic literature review (SLR) of ABSA studies with a focus on trends and high-level relationships among these fundamental components. This review is one of the largest SLRs on ABSA. To our knowledge, it is also the first to systematically examine the interrelations among ABSA research and data distribution across domains, as well as trends in solution paradigms and approaches. Our sample includes 727 primary studies screened from 8550 search results without time constraints via an innovative automatic filtering process. Our quantitative analysis not only identifies trends in nearly two decades of ABSA research development but also unveils a systemic lack of dataset and domain diversity as well as domain mismatch that may hinder the development of future ABSA research. We discuss these findings and their implications and propose suggestions for future research.
- 17 September 2024
- Published by Springer Nature in Artificial Intelligence Review
- Vol. 57 (11) , 1-51
Top Cited PapersOpen AccessA systematic review of aspect-based sentiment analysis: domains, methods, and trends
Aspect-based sentiment analysis (ABSA) is a fine-grained type of sentiment analysis that identifies aspects and their associated opinions from a given text. With the surge of digital opinionated text data, ABSA gained increasing popularity for its ability to mine more detailed and targeted insights. Many review papers on ABSA subtasks and solution methodologies exist, however, few focus on trends over time or systemic issues relating to research application domains, datasets, and solution approaches. To fill the gap, this paper presents a systematic literature review (SLR) of ABSA studies with a focus on trends and high-level relationships among these fundamental components. This review is one of the largest SLRs on ABSA. To our knowledge, it is also the first to systematically examine the interrelations among ABSA research and data distribution across domains, as well as trends in solution paradigms and approaches. Our sample includes 727 primary studies screened from 8550 search results without time constraints via an innovative automatic filtering process. Our quantitative analysis not only identifies trends in nearly two decades of ABSA research development but also unveils a systemic lack of dataset and domain diversity as well as domain mismatch that may hinder the development of future ABSA research. We discuss these findings and their implications and propose suggestions for future research.
- Open Access
- 13 June 2023
- Published by Association for Computing Machinery (ACM) in Proceedings of the ACM on Management of Data
- Vol. 1 (2) , 1-25
DAMR: Dynamic Adjacency Matrix Representation Learning for Multivariate Time Series Imputation
Missing data imputation for location-based sensor data has attracted much attention in recent years. The state-of-the-art imputation methods based on graph neural networks have a priori assumption that the spatial correlations between sensor locations are static. However, real-world data sets often exhibit dynamic spatial correlations. This paper proposes a novel approach to capturing the dynamics of spatial correlations between geographical locations as a composition of the constant, long-term trends and periodic patterns. To this end, we design a new method called Dynamic Adjacency Matrix Representation (DAMR) that extracts various dynamic patterns of spatial correlations and represents them as adjacency matrices. The adjacency matrices are then aggregated and fed into a well-designed graph representation learning layer for predicting the missing values. Through extensive experiments on six real-world data sets, we demonstrate that DAMR reduces the MAE by up to 19.4% compared with the state-of-the-art methods for the missing value imputation task
- 13 June 2023
- Published by Association for Computing Machinery (ACM) in Proceedings of the ACM on Management of Data
- Vol. 1 (2) , 1-25
Open AccessDAMR: Dynamic Adjacency Matrix Representation Learning for Multivariate Time Series Imputation
Missing data imputation for location-based sensor data has attracted much attention in recent years. The state-of-the-art imputation methods based on graph neural networks have a priori assumption that the spatial correlations between sensor locations are static. However, real-world data sets often exhibit dynamic spatial correlations. This paper proposes a novel approach to capturing the dynamics of spatial correlations between geographical locations as a composition of the constant, long-term trends and periodic patterns. To this end, we design a new method called Dynamic Adjacency Matrix Representation (DAMR) that extracts various dynamic patterns of spatial correlations and represents them as adjacency matrices. The adjacency matrices are then aggregated and fed into a well-designed graph representation learning layer for predicting the missing values. Through extensive experiments on six real-world data sets, we demonstrate that DAMR reduces the MAE by up to 19.4% compared with the state-of-the-art methods for the missing value imputation task