Abstracts of the Third School (21.10 2022)

Materials Informatics: The Marriage of Materials and Data Sciences

by Hanoch Senderowitz (10:45)

Materials informatics is rapidly developing as evidenced by the large increase in the number of publications in this field. This development is fueled by the continuous growth of available data, both experimental and computational, on one hand and by the ability to draw from the rich repertoire of methods in its sister field, chemoinformatics, on the other. Similar to chemoinformatics, materials informatics can help bridge the so-called data-knowledge gap by offering smart ways to navigate the enormous materials space in search of new materials with favorable properties.

A key computational technique in materials informatics is Machine Learning (ML). In this talk, we will therefore comparatively analyze the similarities and differences between materials informatics and cheminformatics, within the framework of what is required to derive reliable ML-based models.

Next we will focus on the application of ML methods to the study of solar cells. Such cells hold the potential to meet the growing worldwide demand for clean energy. Today most solar cells are based on silicon yet new alternatives are continuously emerging including organic photovoltaic cells, dye sensitized solar cells (DSSCs), and metal-oxide (MO)-based solar cells. However despite significant progress, all these cells could benefit from improvements in key components, e.g., the dye in DSSCs and the MO composition of MO-based solar cells. Thus, we will present how the concept of pharmacophore can help identify new dyes with favorable (predicted) electronic properties for DSSCs and how combining combinatorial materials sciences with ML can lead to predictive models for key solar cells parameters such as current, voltage and quantum efficiency.

Finally, we will shift our attention to the field of forensic informatics and in particular to the usage of ML in order to analyze physical evidence found in crime scenes. This work will highlight the usefulness of experimentally determined spectra, in particular such that reflect the elemental composition of such evidence. In particular, we will describe the development of a reliable of ML-based classifier from glass fragments retrieved from different types of car windshields using ion beam analysis.

As a final take home message, we emphasize the need to conduct research in the field of materials informatics in close collaboration with experimentalists in order to provide insight into the observed trends and to capitalize on the results.

Unlocking new training data sources for Drug Discovery Machine Learning

by  Hugo Ceulemans (12:00)

Advanced machine learning approaches enable the predictive models that support drug discovery to also benefit from unconventional data sources. A first example is the use of microscopy images for compound de-risking. A second is the leveraging of multiparty data through privacy-preserving and federated learning. In MELLODDY, ten pharma all realized aggregated model improvements by training on 2.6B+ confidential experimental activity datapoints, documenting 21M+ physical molecules and 40K+ discovery-relevant assays

Responsible Conduct of Research, how to do it

by Marcel van der Heyden    (14:00 and 15:30)

Being a scientist is complicated. You have to deal with responsibilities towards your colleagues, your profession, science in general and society. And on top of that, you have to perform state of the art research providing thrilling results and new insights for the life sciences. Everyone knows that one should not cheat in science, but still it happens. Why? This lecture and workshop will help you to remain a good citizen in science. The main aspects of responsible conduct of research will be discussed in short by examples taken from the life sciences. Tools will be provided to withstand the seductions and challenges put on you by supervisors, “the system” and your own ambitions. The afternoon will also be open to discuss daily life experiences in research practice which will help you to responsibly maneuver through the grey areas of science.