Researchers to benefit from cutting-edge research software in 25 newly awarded projects

Dr. B. Kranstauber, Universiteit van Amsterdam

Monitoring the daily and seasonal movements of birds through space and time is challenging without technological solutions. Meteorological radars, used to measure precipitation, can also monitor birds’ aerial movements. Up till now, this has mostly been restricted to studying average movements of birds during migration. We will use machine learning to identify fine-scale, regional movements of flocks of birds.  

Purpose and intended result: 

Through repeated radar measurements, we will track fine-scale movements across space and time, enabling studies of the roles of habitat, topography and environmental stressors on movements. Tracking of large aggregations of birds will facilitate strategic planning to avoid conflicts between birds, airplanes and wind turbines. 

Dr. E. Oppersma, Universiteit Twente 

The diaphragm is the most important respiratory muscle. When it fails and breathing needs mechanical support, it is essential to monitor diaphragmatic activity, both to prevent further failure and optimize treatment. This diaphragmatic monitoring is traditionally done by inserting a catheter with electrodes through the nose into the stomach. However, it is invasive, difficult to position and signal analysis is cumbersome. A novel, alternative approach is to measure the electromyogram (EMG) via electrodes attached to the skin. Surface EMG is easy-to-use, non-invasive and applicable in both intensive care as in home mechanical ventilation. Yet, analysis of these inherently complex data remains so far largely limited to research.  

Purpose and intended result: 

We hypothesize that additional advanced signal analysis techniques will simplify diagnostics and help to extract more information than the fixed parameters currently used. Specifically, we propose to develop a software tool for automatic analysis of surface EMG data to improve mechanical ventilation. 

Dr. J.J.J. van der Hooft, Wageningen University & Research 

The soil and human gut are both thriving with many sorts of microorganisms that each possess unique metabolic capacities. Different omics technologies are used nowadays to measure and map microbial chemistry. Microbial genomes are characterized with genomics and contain their chemical blueprint, whereas their actual metabolic output is determined by many factors and measured by metabolomics. Some of the produced molecules can kill or disarm specific microbial species, i.e., they could be developed into antibiotics to suppress pathogens.   

Purpose and intended result: 

To accelerate the finding of these special bioactive molecules, this project will build a unique framework that can handle large-scale omics profiles, analyze them in concert, and visualize multiple sources of information in one place. This will allow us to develop novel algorithms to solve molecular structures and prioritize novel chemistry. Our proposed community-driven framework will boost integrative omics analysis and will pave the way toward accelerated finding of novel potential antibiotics. 

Dr. A.L. Lamprecht, Universiteit Utrecht 

A “Bake Off” is a contest in which preselected candidates have to perform in a series of baking challenges, at the end of which the best baker is determined. A similar approach can be beneficial in science, where researchers regularly face the challenge of selecting the optimal software workflow for analyzing their data: first explore the possible workflows and select a set of promising candidates for further consideration, then let the candidate workflows run on well-known reference datasets to determine the best-performing one. The new data is then analyzed with the “winning” workflow.  

Purpose and intended result: 

In this project, we will develop a software system that facilitates a “Great Bake Off” for workflows in computational proteomics, a thriving bioinformatics discipline. The new system will enable the researchers to create better data analysis workflows and thus obtain better scientific results with less effort. 

Dr. L.X. Xue, Radboud UMC

Conventional vaccines (flu, corona) present a part of the pathogen to the immune system that learns how to deal with the same pathogen in case of future infections. Cancer vaccine uses fragments (peptides) of a patient’s mutated tumor proteins to fire up the patient’s immune system to attack tumor cells. Cancer vaccines have successfully eliminated some advanced tumors, but general applicability requires that we predict which tumor peptides can protect people from tumor growth. To be a good candidate, a tumor peptide must bind a protein called MHC, which then ships the peptide to the cell surface to be visible to immune T-cells.

Purpose and intended result:

We will apply a recently developed deep learning method for 3D point clouds (called geometric deep learning) to 3D structure models of peptide-MHC complexes to predict the best vaccine peptides. If successful, the project will spur the safety, efficacy, and development speed of personalized cancer vaccines.

Dr. H. Cheng, Universiteit Twente 

How do we keep dikes safe with rising sea levels? Why are ripples formed in sand? What can we prepare for landing on Mars? At the center of these questions is the understanding of how the grains, as a self-organizing material, collide, flow, or get jammed and compressed. State-of-the-art algorithms allow for simulating millions of grains individually within a computer. However, these simulations can take very long, and the big data about particle motion is very difficult to interpret and generalize, say from a simulation of avalanches to free-standing sandcastles.  

Purpose and intended result: 

In this project, we will use machine learning to (1) extract hidden links between grain, microstructural and macroscopic properties, and (2) instantaneously generate microstructures that satisfy given macroscopic constraints, from an existing database. As proof of concept, the workflow will be deployed on the cloud and used to find optimal microstructure/grain properties that define a SMART granular material. 

Dr. M. Khodadadzadeh, Universiteit Twente 

Climate change is widespread and intensifying rapidly. Temperatures continue to rise; droughts, forest wildfires and floods caused by extreme weather are becoming more frequent and severe. Climate change is undeniably impacting our planet and, as such, altering plants’ distribution and growth. The timing of plant’s life cycle events (like leafing) is clearly changing. Phenology is the science that studies these changes, their causes and interrelations.  

Purpose and intended result: 

This project aims at supporting such studies by providing efficient tools that facilitate and improve the process of discovering patterns and knowledge from phenological and environmental spatio-temporal data. These data are voluminous and characterized by complex spatial, temporal, and spatio-temporal correlations. This project proposes a regression machine learning technique that allows dealing with such complex data and providing both accurate and interpretable predictions. It will help experts to better understand phenological changes and more accurately analyze the impact of environmental and climate drivers on plants. 

Dr. J. Citrin, DIFFER 

Fusion energy is a promising technology for high density, CO2-free energy production. The next generation of magnetic fusion experiments aims to demonstrate net energy gain for the first time. The increased complexity and scale of these experiments compared to present-day devices requires a leap in computational capabilities for predicting the behaviour of the fusion reactor plasma, magnets and wall materials. This is vital for experimental performance optimization, plasma control and safe reactor operation. Central to these efforts is how to efficiently design software that couples together different physics and engineering codes that separately describe parts of the interacting system into a single and consistent description.  

Purpose and intended result: 

Coupled with ongoing international efforts, this project implements methods and standards vital for a tokamak simulation suite which is modular, robust, easy to maintain, and with a user toolkit that enables efficient application for large numbers of simulations needed for designing experiments and assessing sensitivities. 

Dr. D. Coumou, Vrije Universiteit Amsterdam 

Society is vulnerable to weather extremes like heatwaves and droughts, and current operational forecast systems are typically only accurate up to 10 days ahead. New, massive climate datasets from satellites and climate models, combined with novel machine learning techniques provide great promise to push that forecast horizon further, potentially to several weeks or months ahead. To achieve this, we want to develop code that enables reproducible analyses according to best practices, which also provides insights into where the predictability is coming from.  

Purpose and intended result: 

AI4S2S will build an open-source python package that can efficiently run across different Big Climate Data platforms and that will include the latest advances in machine learning. We will actively involve experts worldwide to generate a sustainable, community-driven coding effort, via dedicated workshops and online outreach. AI4S2S has the potential to make a huge impact on research (enabling scientific breakthroughs), education (lowering technical barriers) and society (forecast-based risk reduction). 

Dr. D.V. Voskov, TU Delft 

Subsurface reservoirs are used for various applications that are part of the energy transition towards zero-carbon energy sources. They can serve as a direct source of energy (geothermal production), cyclic energy storage required by renewable energy production technologies (sun and wind) and sequestration of “energy waste” (carbon dioxide sequestration). Making optimal use of subsurface reservoirs is a great challenge for both academia and society.  

Purpose and intended result: 

In this research project, we are going to develop a numerical framework capable of modelling and optimizing energy transition applications relevant to subsurface reservoirs. The developed software will allow for accurate prediction of the effects of the subsurface use and assessment of techno-economical risks (e.g., induced seismicity) for applications relevant to the energy transition. The results of this project can be directly used for more efficient energy production, risk assessment in energy transition applications and various legislative initiatives. 

Dr. P.E. Bleeker, Huygens ING 

Literary works are dynamic entities: they go through different stages of development before publication, and often continue to change even after their first publication. The early versions of a work, such as notes, draft manuscripts and typescripts, still show the traces of this dynamic development in the form of deletions, additions or substitutions. Today, these documents are carefully transcribed, annotated and encoded in a machine-readable language. Using text comparison tools, scholars can automatically compare the encoded text versions and examine the different stages in the work’s development. So far, however, it is not possible to include the annotations in the comparison process. This means that relevant scholarly information is lost.  

Purpose and intended result: 

The project proposes to employ machine learning technologies to develop a comparison tool that can take into account text as well as annotations. As a result, it will allow scholars to analyze the textual development at unprecedented levels of detail. 

Dr. ir. F. Hasibi, Radboud Universiteit 

Machine understanding of text is an extremely challenging task for intelligent systems; machines need to understand the meaning behind words and reason about the text and existing knowledge. A highly effective method to obtain this understanding is connecting unstructured text to semi-structured information in the knowledge graphs (e.g., Wikipedia and Wikidata). In this process, referred to as Entity Linking, entities such as people and locations are identified and mapped to their corresponding entries in a knowledge graph. Entity linking is a language-dependent, compute, and data hungry process. How can we make this technology usable for multilingual, formal, and informal texts, requiring only limited computational power?  

Purpose and intended result: 

In this project, we introduce REL 2.0, a publicly available entity linking toolkit that can operate on texts in a variety of languages and forms (e.g., long documents, queries, and conversations), in a reasonable time and using commonly available hardware. 

Dr. P. Huijnen, Universiteit Utrecht 

This project uses state-of-the-art machine learning techniques to study conceptual change over time. It builds on the seminal BERT infrastructure that has, in recent years, caused a breakthrough in the computational understanding of language. With the help of the Dutch National Library’s massive archive of historical newspapers, magazines and books, it is possible to show how Dutch words have changed their meaning and connotation in public discourse from the Second World War until the present day. The project aims to study the conceptual history of one of the most urgent issues of today: global sustainability. 

Purpose and intended result: 

In this project, we will re-train the base model to create multiple, chronologically ordered models based on historical Dutch textual data. With the help of this technique, we will be able to trace continuities and breaks in this discourse to, ultimately, gain insights into the forces at play when it comes to sustainability. 

Dr. G. Schumacher, Universiteit van Amsterdam 

When politicians speak, their choice of words, their facial expressions and even the pitch of their voice communicates specific emotions to us. These emotions may influence how we think about that politician or the message they seek to convey. Politicians differ in how they use their words, their voice and their face: they have a different emotion repertoire.  

Purpose and intended result: 

To capture this repertoire, we propose MEXCA: a system that integrates different existing software solutions to capture the emotion conveyed in words, voice and face in one efficient, reliable and accessible workflow. This allows us to address questions about how politicians use emotion, how intentional it is and how systematically it is employed. The system may also be used outside of the domain of politics to study emotion expression in general. 

Dr. S.R.L. Roijackers, Universiteit Utrecht 

Reusing research software, the software that advances society by supporting and contributing to academic research, is hard. It is hard to find the algorithm needed, it is hard to extract only the source code that is needed, and it is hard to give credits to the original research software engineers that developed the software. We have developed a software search engine that mines software repositories on the method level, instead of the project or source-code level. 

Purpose and intended result: 

We plan to develop a search portal that facilitates this process. Furthermore, we will use our database to efficiently detect reuse of methods across projects for the measurement of research software impact. In this way, we can monitor the impact of research software and thus give a fairer representation of its value for society. Research organizations, including the eScience Center itself, can use SearchSECO to measure the success of the software they developed themselves. 

Dr. ir. F. van Leijen, TU Delft 

Dike and infrastructure failures are events with a low-probability, but an extremely high-impact. Reliable warning systems can save lives and property. We propose an autonomous processing system based on near-continuous streams of satellite data, which enables the estimation of indicative surface motion at millimetre-precision. For a reliable estimate of these motions, additional spatial and temporal information on, for example, soil types, building age, and weather conditions should be incorporated, which enables a more reliable estimation by Artificial Intelligence (AI) techniques.  

Purpose and intended result: 

We will design, develop and demonstrate a generic toolbox for the homogenization of these additional data sources. Furthermore, an AI approach is designed and implemented which uses these homogenized datasets for the estimation of ground motion time series.