Michiel Van Dyck
Michiel Van Dyck has a total of 122 co-authors. Here are the top co-authors by number of shared publications:
Herbert Peremans
Department Milieu, Technologie, en Technologiemanagement, Universiteit Antwerpen, Antwerpen, Belgium
3 shared publications- Johan A. K. Suykens
ESAT-STADIUS, KU Leuven
2 shared publications - Joos Vandewalle
KU Leuven
2 shared publications - Samuel Xavier-De-Souza
UFRN
2 shared publications Adriana Borowa
Ardigen S.A, Krakow, Poland
1 shared publicationAmanda Lembke
Amgen, Thousand Oaks, CA, USA
1 shared publication
Latest Publications
- Open Access
- 29 October 2026
- Published by Springer Nature in Nature Methods
- Vol. 23 (11) , 2181-2194
Advancing morphological profiling with self-supervised learning through comparative insights into representation learning of cellular perturbations
Self-supervised learning (SSL) enables scalable, label-free foundation models (FMs) that learn task-agnostic representations directly from images, bypassing cell segmentation and feature engineering inherent to classical morphological profiling. We present a large-scale evaluation of SSL FMs for image-based phenotypic profiling, covering contrastive (DINO), masked-based generative (MAE, iTPN) and hybrid (DINOv2) learning mechanisms, pretrained on all unannotated chemical perturbations from the Joint Undertaking for Morphological Profiling–Cell Painting (JUMP–CP) dataset (119,876 compounds, ~1.77 million images, 7 laboratories). Here we show that top-performing SSL models outperformed CellProfiler in compound-replicate retrieval on the held-out JUMP–Target2 set (302 annotated compounds, P ≤ 0.001). DINO and DINOv2 generalized across diverse perturbations and cell types, outperforming CellProfiler with 31–37% gains in site-averaged mean average precision for mechanism-of-action retrieval on EU-OPENSCREEN Bioactives (2,464 compounds), and consistently surpassing masked-based models across all benchmarks. This study provides a systematic assessment of SSL approaches for advancing morphological profiling and offers pretrained FMs to accelerate drug discovery.
- 29 October 2026
- Published by Springer Nature in Nature Methods
- Vol. 23 (11) , 2181-2194
Open AccessAdvancing morphological profiling with self-supervised learning through comparative insights into representation learning of cellular perturbations
Self-supervised learning (SSL) enables scalable, label-free foundation models (FMs) that learn task-agnostic representations directly from images, bypassing cell segmentation and feature engineering inherent to classical morphological profiling. We present a large-scale evaluation of SSL FMs for image-based phenotypic profiling, covering contrastive (DINO), masked-based generative (MAE, iTPN) and hybrid (DINOv2) learning mechanisms, pretrained on all unannotated chemical perturbations from the Joint Undertaking for Morphological Profiling–Cell Painting (JUMP–CP) dataset (119,876 compounds, ~1.77 million images, 7 laboratories). Here we show that top-performing SSL models outperformed CellProfiler in compound-replicate retrieval on the held-out JUMP–Target2 set (302 annotated compounds, P ≤ 0.001). DINO and DINOv2 generalized across diverse perturbations and cell types, outperforming CellProfiler with 31–37% gains in site-averaged mean average precision for mechanism-of-action retrieval on EU-OPENSCREEN Bioactives (2,464 compounds), and consistently surpassing masked-based models across all benchmarks. This study provides a systematic assessment of SSL approaches for advancing morphological profiling and offers pretrained FMs to accelerate drug discovery.
- Open Access
JUMP Cell Painting chemical dataset: Morphological map of 115,796 compounds in human cells
Image-based profiling decodes complex cellular states for high-impact biological and therapeutic applications, yet a large, public reference dataset of chemical perturbations has been lacking. We present Cell Painting images and 1.53 billion single-cell profiles for 115,796 compounds in human osteosarcoma cells, generated by the Joint Undertaking for Morphological Profiling (JUMP)-Cell Painting Consortium of 18 industry and non-profit partners. Despite technical variability across 12 data-generating laboratories, more than 15% of compounds show statistically significant phenotypic activity, reproducibility drops only 34% versus single-laboratory data, and compounds sharing biological targets often show similar image-based profiles. JUMP establishes a validated public reference resource that brings image-based profiling on par with genomics and transcriptomics. The dataset, shared alongside previously released genetic perturbation data in the Cell Painting Gallery, is a powerful predictive engine for the impact, mechanism, targets and activities of compounds, fundamentally accelerating chemical and cell biology and early-stage drug development.
Open AccessJUMP Cell Painting chemical dataset: Morphological map of 115,796 compounds in human cells
Image-based profiling decodes complex cellular states for high-impact biological and therapeutic applications, yet a large, public reference dataset of chemical perturbations has been lacking. We present Cell Painting images and 1.53 billion single-cell profiles for 115,796 compounds in human osteosarcoma cells, generated by the Joint Undertaking for Morphological Profiling (JUMP)-Cell Painting Consortium of 18 industry and non-profit partners. Despite technical variability across 12 data-generating laboratories, more than 15% of compounds show statistically significant phenotypic activity, reproducibility drops only 34% versus single-laboratory data, and compounds sharing biological targets often show similar image-based profiles. JUMP establishes a validated public reference resource that brings image-based profiling on par with genomics and transcriptomics. The dataset, shared alongside previously released genetic perturbation data in the Cell Painting Gallery, is a powerful predictive engine for the impact, mechanism, targets and activities of compounds, fundamentally accelerating chemical and cell biology and early-stage drug development.
- Open Access
- 10 September 2021
- Published by Springer Nature in Neuroinformatics
- Vol. 20 (3) , 825
Correction to: High‑throughput Analysis of Synaptic Activity in Electrically Stimulated Neuronal Cultures
- 10 September 2021
- Published by Springer Nature in Neuroinformatics
- Vol. 20 (3) , 825
Open AccessCorrection to: High‑throughput Analysis of Synaptic Activity in Electrically Stimulated Neuronal Cultures
- Open Access
- 27 August 2017
- Published by Institute of Mathematics and Informatics Bulgarian Academy of Sciences in Biomath Communications
- Vol. 4 (2)
Dynamic simulation and steady-state computation of 3D physiologically structured population models
The PSP modelling approach allows you to model biological/pharmaceutical behaviour by combining micro-scale ordinary differential equation (ODE) models with macro-scale ODE models and their bi-directional interaction. E.g.: based on the model of a single cell, billions of cells can be simulated to get the response of an entire organ (also incorporating the organ to cell reaction). The PSP approach allows to simulate this in a mathematically efficient way by characterising different cells by a set of physiologically relevant quantities [2], [3]. Our framework is capable of taking into account up to 3 physiological parameters resulting in a 3D structure for which a partial differential equation (PDE) should be solved. We achieve considerable speed-up by using a semi-Lagrangian PDE solver that allows big stable time stepping. Combining this with a low level C-language implementation, we achieve exceptional efficient usage of the computing processing unit (CPU). Furthermore we accelerate the computation of the steady-state by using a Newton-Krylov method. For ease of use, a PSP description language is constructed to allow straightforward input of different models. The simulator should empower future computational/mathematical biologists to create and evaluate more detailed models than currently is common practice.
- 27 August 2017
- Published by Institute of Mathematics and Informatics Bulgarian Academy of Sciences in Biomath Communications
- Vol. 4 (2)
Open AccessDynamic simulation and steady-state computation of 3D physiologically structured population models
The PSP modelling approach allows you to model biological/pharmaceutical behaviour by combining micro-scale ordinary differential equation (ODE) models with macro-scale ODE models and their bi-directional interaction. E.g.: based on the model of a single cell, billions of cells can be simulated to get the response of an entire organ (also incorporating the organ to cell reaction). The PSP approach allows to simulate this in a mathematically efficient way by characterising different cells by a set of physiologically relevant quantities [2], [3]. Our framework is capable of taking into account up to 3 physiological parameters resulting in a 3D structure for which a partial differential equation (PDE) should be solved. We achieve considerable speed-up by using a semi-Lagrangian PDE solver that allows big stable time stepping. Combining this with a low level C-language implementation, we achieve exceptional efficient usage of the computing processing unit (CPU). Furthermore we accelerate the computation of the steady-state by using a Newton-Krylov method. For ease of use, a PSP description language is constructed to allow straightforward input of different models. The simulator should empower future computational/mathematical biologists to create and evaluate more detailed models than currently is common practice.
- 6 September 2011
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Instrumentation and Measurement
- Vol. 61 (2) , 307-315
Structure From Flow for Calibration of an Array Air-Flow Sensor
We describe the automatic calibration of a set of air-flow sensitive sensors exposed to a set of unknown and randomly distributed air-flow stimuli. The algorithm makes use of the singular value decomposition of the sensor data and the known sensor model. The sensor array configuration consists of the maximum response direction vectors of the individual air-flow sensors. The relative orientations of the direction vectors of the array elements are automatically determined. The orientation of the entire array in a world reference frame has to be manually calibrated. The change in the air flow can be measured without any manual calibration.
- 6 September 2011
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Instrumentation and Measurement
- Vol. 61 (2) , 307-315
Structure From Flow for Calibration of an Array Air-Flow Sensor
We describe the automatic calibration of a set of air-flow sensitive sensors exposed to a set of unknown and randomly distributed air-flow stimuli. The algorithm makes use of the singular value decomposition of the sensor data and the known sensor model. The sensor array configuration consists of the maximum response direction vectors of the individual air-flow sensors. The relative orientations of the direction vectors of the array elements are automatically determined. The orientation of the entire array in a world reference frame has to be manually calibrated. The change in the air flow can be measured without any manual calibration.
- 26 May 2009
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Instrumentation and Measurement
- Vol. 58 (8) , 2426-2432
Biomimetic Mechanosensor Array Processing for 2-D Airflow Direction and Amplitude Sensing
This paper presents a computational study and preliminary experimental results of mechanosensor arrays for 2-D airflow sensing. In particular, the analysis presented here applies to a recently developed biomimetic microelectromechanical system (MEMS) inspired by the array of airflow-sensitive hairs found on the crickets' cercus. Based on the maximum-likelihood principle, an estimator that allows reconstruction of both the airflow direction and the airflow velocity amplitude is derived. Given this algorithm, different topologies of hair arrays are investigated in terms of sensitivity for measurement errors. In addition, it is analyzed how redundancy, i.e., the presence of many more than the minimum number of hairs, in the array can be used for lessening the effect of measurement noise. Multiple types of hardware failure common in biological and artificial array systems, as well as the occurrence of saturation, are analyzed, and it is shown how redundancy again increases the robustness of the sensor. Experimental results are shown to validate the theoretical model and its assumptions.
- 26 May 2009
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Instrumentation and Measurement
- Vol. 58 (8) , 2426-2432
Biomimetic Mechanosensor Array Processing for 2-D Airflow Direction and Amplitude Sensing
This paper presents a computational study and preliminary experimental results of mechanosensor arrays for 2-D airflow sensing. In particular, the analysis presented here applies to a recently developed biomimetic microelectromechanical system (MEMS) inspired by the array of airflow-sensitive hairs found on the crickets' cercus. Based on the maximum-likelihood principle, an estimator that allows reconstruction of both the airflow direction and the airflow velocity amplitude is derived. Given this algorithm, different topologies of hair arrays are investigated in terms of sensitivity for measurement errors. In addition, it is analyzed how redundancy, i.e., the presence of many more than the minimum number of hairs, in the array can be used for lessening the effect of measurement noise. Multiple types of hardware failure common in biological and artificial array systems, as well as the occurrence of saturation, are analyzed, and it is shown how redundancy again increases the robustness of the sensor. Experimental results are shown to validate the theoretical model and its assumptions.
- 1 January 2007
- Published by Institute of Electrical and Electronics Engineers (IEEE)
- p. 1-6
A computational study of biomimetic mechanosensor array processing for air flow sensing
We present a computational study of mechanosensor arrays for air flow sensing. The analysis presented here applies to air flow sensors consisting of arrays of hairs e.g., both the array of air flow sensitive hairs found on the crickets' cercus as well as a recently developed biomimetic MEMS system. Based on the maximum likelihood principle we derive an estimator which allows reconstruction of both the air flow direction in 2D and the air flow velocity amplitude assuming the air flow to be the same at all the hairs in the array. Given this algorithm, different topologies of hair arrays are investigated in terms of sensitivity for measurement errors. Also, we analyze how the redundancy i.e, the presence of many more than the minimum number of hairs, in the array can be used for lessening the effect of measurement noise. Multiple types of hardware failure common in biological and artificial array systems as well as the occurrence of saturation are analyzed and it is shown how redundancy again increases the robustness of the sensor.
- 1 January 2007
- Published by Institute of Electrical and Electronics Engineers (IEEE)
- p. 1-6
A computational study of biomimetic mechanosensor array processing for air flow sensing
We present a computational study of mechanosensor arrays for air flow sensing. The analysis presented here applies to air flow sensors consisting of arrays of hairs e.g., both the array of air flow sensitive hairs found on the crickets' cercus as well as a recently developed biomimetic MEMS system. Based on the maximum likelihood principle we derive an estimator which allows reconstruction of both the air flow direction in 2D and the air flow velocity amplitude assuming the air flow to be the same at all the hairs in the array. Given this algorithm, different topologies of hair arrays are investigated in terms of sensitivity for measurement errors. Also, we analyze how the redundancy i.e, the presence of many more than the minimum number of hairs, in the array can be used for lessening the effect of measurement noise. Multiple types of hardware failure common in biological and artificial array systems as well as the occurrence of saturation are analyzed and it is shown how redundancy again increases the robustness of the sensor.
- 1 August 2006
- Published by Institute of Electrical and Electronics Engineers (IEEE)
- p. 1
Fast and roust face tracking applieed to wheelchair driving
In this demonstration we will present an application of a fast and robust face tracking algorithm implemented with a CNN Universal Machine chip. The application is the driving of a wheelchair according to predefined movements of the user's head. A 3D model of a wheelchair is used in order to demonstrate the driving mechanism.
- 1 August 2006
- Published by Institute of Electrical and Electronics Engineers (IEEE)
- p. 1
Fast and roust face tracking applieed to wheelchair driving
In this demonstration we will present an application of a fast and robust face tracking algorithm implemented with a CNN Universal Machine chip. The application is the driving of a wheelchair according to predefined movements of the user's head. A 3D model of a wheelchair is used in order to demonstrate the driving mechanism.
- 1 August 2006
- Published by Institute of Electrical and Electronics Engineers (IEEE)
- No. 21650144,p. 1-6
Fast an Robust Face Tracking for CNN chips: application to wheelchair driving
An algorithm for fast and robust face tracking with the CNN universal machine is proposed in this paper. It is applied to a driving mechanism for a wheelchair with an on-chip implementation. A novel object tracking CNN visual algorithm is introduced and employed in the tracking of multiple face features. The speed and robustness of this method are achieved due to the parallelism in the visual algorithm, and the tracking of multiple face features. The tracking algorithm is designed to achieve a high frame rate and exploit the specific properties of face features. The face tracking method proposed here was implemented on a Bi-I stand-alone cellular vision system and applied to a wheelchair driving mechanism. The template operations were trained and/or fine-tuned in order to generate chip-specific robust templates. In order to improve performance in environments with varying illumination, an adaptive image capture procedure was also introduced. Our simulations with a 3D model wheelchair showed that the final algorithm is capable of performing tracking with a frame rate of 92 frames/sec, which is supposedly enough for real-time driving in most of the real life situations
- 1 August 2006
- Published by Institute of Electrical and Electronics Engineers (IEEE)
- No. 21650144,p. 1-6
Fast an Robust Face Tracking for CNN chips: application to wheelchair driving
An algorithm for fast and robust face tracking with the CNN universal machine is proposed in this paper. It is applied to a driving mechanism for a wheelchair with an on-chip implementation. A novel object tracking CNN visual algorithm is introduced and employed in the tracking of multiple face features. The speed and robustness of this method are achieved due to the parallelism in the visual algorithm, and the tracking of multiple face features. The tracking algorithm is designed to achieve a high frame rate and exploit the specific properties of face features. The face tracking method proposed here was implemented on a Bi-I stand-alone cellular vision system and applied to a wheelchair driving mechanism. The template operations were trained and/or fine-tuned in order to generate chip-specific robust templates. In order to improve performance in environments with varying illumination, an adaptive image capture procedure was also introduced. Our simulations with a 3D model wheelchair showed that the final algorithm is capable of performing tracking with a frame rate of 92 frames/sec, which is supposedly enough for real-time driving in most of the real life situations