Conférenciers invités

Age K. Smilde, University of Amsterdam

Age_K_SmildeAge K. Smilde is full professor of Biosystems Data Analysis at the Swammerdam Institute for Life Sciences atthe University of Amsterdam and as June 1, 2013 he holds a part-time as professor at the Department of Food Science at the University of Copenhagen. He has published more than 250 peer-reviewed papers of which more than 70 about metabolomics data analysis. He is Editor-in-Chief of the Journal of Chemometrics and he is co-founder of the Netherlands Metabolomics Centre, a large Public/Private Consortium devoted to all aspects of metabolomics. His research interest is data fusion and multiset methods.

  • Common and distinct components in data fusion

One of the active research topics in both chemometrics and systems biology is data fusion. This is also called data integration in systems biology and multiset analysis in chemometrics and psychometrics. Although this field has already a long history (especially in chemometrics and psychometrics), it has recently attracted a lot of attention in systems biology due to the availability of large genomics data sets. One of the topics within data fusion is to separate common from distinct components in these data sets. This is important since it cleans up the exploratory process by the ability to focus on these components. There are several methods to distinguish common from distinct components and some of these will be discussed. We will also address the fusion of data of different measurements scales and how to find common and distinct components in these types of data.

Julien Jacques, University of Lyon 2

J_JacquesJulien Jacques received his Ph.D degree in Applied Mathematics from University of Grenoble, France. In 2006, he joined University of Lille where he holds the position of Associate Professor. In 2014, he joined University of Lyon as Full Professor in Statistics. His current research in statistical learning concerns the design of clustering algorithm for functional data, ordinal data and mixed-type data. He is a member of the executive board of the French Society of Statistics.

  • Co-clustering model for three-way data  

Three-way data can be seen as a collection of two-way matrices, as we can meet when the same samples are measured several times in different conditions.

Considering such data as a matrix in which each element is the observation of a random function, we propose a new co-clustering methodology which aims to produce simultaneously a clustering of the rows and a clustering of the columns. The proposed functional latent block model (funLBM) extends the usual latent block model to the functional case by assuming that the curves of one block live into a low-dimensional functional subspace. Thus, funLBM is able to model and cluster large data set with high-frequency curves. A stochastic EM algorithm embedding a Gibbs sampler is proposed for model inference. An ICL model selection criterion is also derived to address the problem of choosing the number of row and column clusters. Numerical experiments on simulated data and application on electricity consumptions show the usefulness of the proposed methodology.

Thomas De Beer, Ghent University

Thomas_De_BeerThomas De Beer graduated in pharmaceutical sciences in 2002 at Ghent University in Belgium. He obtained his PhD at the same university in 2007. for his PhD research, he examined the suitability of Raman spectroscopy as a Process Analytical Technology tool for pharmaceutical production processes. Within his PhD research period, he worked four months at University of Copenhagen in Denmark, Department of Pharmaceutics and Analytical Chemistry (Prof. Jukka Rantanen). After his PhD, he was a FWO funded post-doctoral fellow at Ghent University (2007-2010). Within his post-doc mandate, he worked 9 months at the Department of Pharmacy, Pharmaceutical Technology and Biopharmaceutics from the Ludwig-Maximilians-University in Munich, Germany (Prof. Winter and Prof. Frieb). In February 2010, he became professor in Process Analytics & Technology at the Faculty of Pharmaceutical Sciences from the University of Ghent. His research goals include bringing innovation pharmaceutical production processes (freeze-drying, hot-melt extrusion, continuous from-powder-to-tablet processing etc.). More specifically, Prof. De Beer contributes to the development of continuous manufacturing processes for drug products such as solids, semi-solids, liquids and biologicals (continuous freeze-drying of unit doses). Thomas De Beer is also director of the Ghent University's Center of Excellence in Sustainable Pharmaceutical Engineering (CESPE) which is founded in 2016. In 2018, Thomas De Beer became co-founder and CTO of the Ghent University spin off company RheaVita which provides a continuous freeze-drying technology for the pharmaceutical market.  

  • The development of a predictive tableting platform in the context of continuous manufacturing

The development of robust tableting processes in a timely manner is still challenging due to a lack of mechanistic process understanding, fundamental understanding of influence of raw material attributes, and limited usage of sophisticated process simulation tools. Experimentally determining the effects of involved process- and formulation parameters and thereafter optimization is labor-intensive, expensive and time-consuming. Therefore, raw material database management and process modeling using multivariate data analysis techniques and numerical simulations of the compaction process based on Finite Element Analysis (FEA) can provide a valuable and efficient alternative.

A set of 55 powders covering both excipients and active pharmaceutical ingredients (API) was characterized using over 20 techniques describing particle size and -shape, density, moisture content, powder flow, compressibility, aeration, surface area and triboelectric charging. Additionally, multiple mechanical properties, including plastic-, elastic, and brittle deformation, were determined based on in-die compaction tests. Principal Component Analysis (PCA) was then performed to elucidate correlations between the powders and their measured properties. Based on this analysis, formulation blends were selected containing multiple API’s and fillers covering a maximal area of the material variability space determined via the PCA. Disintegrant (Sodium Crosscarmellose; 5%), glidant (Colloidal Silicon Dioxide; 0,5%) and lubricant (Magnesium Stearate; 0,75%) were kept fixed in the selected blends. Formulation blend bulk properties were characterized with a minimum number of relevant tests, as derived from the PCA model of the raw materials. These blends were subsequently compacted using the press type simulation tool on a Huxley Bertram (HB) 1088-C compaction simulator to evaluate tablet quality attributes under different process conditions using the Design of Experiments (DoE) approach.

T-PLS models were developed to link raw material properties, blending ratio’s and process settings with the resulting tablet quality attributes. Based on these models it is possible to determine which formulation- and process parameters affect the direct compaction process and resulting tablet properties. By optimizing this model for new API’s, it can be used to predict an optimal formulation and finetune the process settings based on a minimum of relevant raw material characterization techniques and compaction tests. So far, this approach has only been tested on a compaction simulator, for further implementation scale-up experiments are required.

These results contribute to a better understanding of the impact of powder properties and process settings on a direct compaction process and final properties of the produced tablets. Applying this knowledge can reduce consumption of expensive API’s during product development phase. This predictive platform, which combines the use of PLS and FEA models, can be implemented for future development of formulations for direct compression and finding optimal process settings in a minimal amount of time. This platform can later be linked with additional unit operations such as feeding and granulation for implementation towards continuous manufacturing processes.

David Rousseau, Université d'Angers

David_RousseauDavid Rousseau is full Professor in Data Science at Université d'Angers, France. His research interests lay at the interface between physics of imaging, image processing and machine learning with applications in life sciences including cellular biology, plant science and biomedical imaging.

Lowering the cost of spectral imaging and machine learning: Application to plant disease detection 

In era of machine learning driven computer vision, unequaled performances are accessible with advanced algorithms such as deep learning. The bottleneck is no more the design of the algorithms but in the cost of computation devices and the time required for the creation of ground truth associated to the images to be processedIn this talk we propose some strategies to accelerate the creation of such ground truth or use deep architectures which require less hyperparameters. Also, we will provide an example of how deep learning can be used to decrease the cost of instrumentation when included in a computational imaging strategy. This is illustrated with recent applications in the domain of spectral plant imaging.

  • Douarre, C., Crispim-Junior, C. F., Gelibert, A., Tougne, L., Rousseau, D. (2019). Novel data augmentation strategies to boost supervised segmentation of plant disease. Computers and Electronics in Agriculture, 165, 104967.
  • Rasti, P., Ahmad, A., Samiei, S., Belin, E., Rousseau, D. (2019). Supervised image classification by scattering transform with application to weed detection in culture crops of high density. Remote sensing, 11(3), 249.
  • Sapoukhina, N., Samiei, S., Rasti, P., Rousseau, D. (2019). Data augmentation from RGB to chlorophyll fluorescence imaging application to leaf segmentation of Arabidopsis thaliana from top view images. In Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition Workshops.
  • Salma, S., Ahmad, A., Rasti, P., Belin, E., Rousseau, D. (2018). Low-cost image annotation for supervised machine learning. Application to the detection of weeds in dense culture. In BMVC Computer Vision Problems in Plant Phenotyping (CVPPP 2018).
Personnes connectées : 1