A hybrid approach for identifying drug repurposing candidates and their mechanisms: An interview with Vanessa Lage-Rupprecht and two co-authors.

Senior researcher Vanessa Lage-Rupprecht and two collaborators talk about what data science means to them and illustrate how they managed to create a data and lab coexistence in their drug-repurposing project, which was recently published in Patterns. In this article, they have developed a drug-target-mechanism-oriented data model, Human Brain PHARMACOME, and have presented it as a resource to the community.

Main text

Vanessa, Bruce, and Marcin, please tell us a little about yourselves and your recent paper published in Patterns

Vanessa Lage-Rupprecht: At Fraunhofer SCAI Bioinformatics, I use my neuroscientific expertise in the generation of formalized knowledge graphs and their application in translational approaches. My main focus is on the discovery of knowledge in the field of neurodegenerative diseases in order to advance the development of new therapeutic approaches. I became a scientist because I have always been interested in researching and understanding complex biological interactions. I moved to data science because I was attracted to the idea of putting findings from experimental research studies into a big picture and looking at them holistically. Professor Hofmann-Apitius and the philosophy of his department played a big part in making me fall in love with data science, a multifaceted field that, at its core, focuses on the extraction of data and the knowledge derived from it. I see myself as the one who thinks about concepts to bring data and domain knowledge together to establish dedicated data-mining workflows. We chose Patterns for our submission because it is an excellent platform for sharing novel interdisciplinary research ideas, which is exactly what our work is about (Figure 1).

Figure 1

Left: Vanessa Lage-Rupprecht, middle: Bruce Schultz, right: Marcin Namysl

Bruce Schultz: I currently work as a research scientist at Fraunhofer SCAI with a strong focus on the development and application of formalized knowledge graphs. My background was initially in biochemistry and immunology, having worked for several years in HIV vaccine research. I then pivoted my career to a more data-oriented direction and obtained a MSc in bioinformatics. Though most of my work is now dedicated to writing new software packages and developing computational algorithms, my love of biology is still the primary focus. Understanding every facet and mechanism associated with a given pathology is incredibly difficult for an individual, which is why I was excited to join the team at Fraunhofer SCAI and begin working on methods for creating a comprehensive resource encompassing everything we know about Alzheimer disease and other neurodegenerative disorders. Ultimately, we seek to make biological research much more focused and efficient by providing evidence-based insights on the targets and pathways which are likely to provide results.

Marcin Namysl: I am a research scientist with a software engineering background and a PhD candidate majoring in computer science. I am doing research that involves natural language processing and pattern recognition. In particular, I am focused on methods for improving the robustness of information extraction performed on input documents from diverse domains. Although many information-extraction tasks are regarded as largely solved problems, this assumption only holds true if the input text is free of errors and the correct reading order of a document is preserved. This motivated me to investigate the approaches for improving the robustness of the entity and table extraction systems in real-world scenarios, where the input may contain digitization errors and complex layout elements. Such documents cannot be simply serialized as a sequence of natural language tokens by ignoring the inherent structure of the data, e.g., contained in tables. In our current work, we complemented the enrichment of our computational model with quantitative information that was automatically extracted from tables in published biomedical manuscripts.

Vanessa, how did this project come to be?

VLR: Clinical trials in the context of Alzheimer disease continue to fail. An incomplete understanding of pathophysiological mechanisms and erroneous selection of target and drug candidates are partial reasons. In our project, a collaboration between two computer science and one life science institute, we aimed to establish new ways to repurpose drugs to overcome these shortcomings. In our holistic approach that was recently published in Patterns, we combined knowledge mined from a variety of sources with focused experimental screening data of selected drug candidates into our drug-target-mechanism-oriented data model, the Human Brain PHARMACOME (HBP). Mutual analysis led to the identification of interesting repurposing candidates in the Alzheimer disease context.

Vanessa and Bruce, what’s next for the project?

VLR: Our interplay of in-silico and in-vitro analysis has yielded far-reaching findings that are being further explored experimentally on the one hand and taken up in new studies on the other. The generalizability of our approach allows us to investigate a wide range of biological questions and to continue the successful symbiosis of computational and wet lab work.

BS: Though our approach published here will provide a lot of utility in generating new insights and hypotheses, it is only the beginning. We wish to improve our integration of quantitative data within our knowledge graph and develop new methods capable not only of pinpointing new modes of action for drugs, but also predicting dosages, likely side effects, and even unknown drug interactions.

Vanessa, how important was the collaboration to the success of the paper? How important do you think collaboration is to research in general?

VLR: The collaboration was enormously important for the success of the paper. The mutual exchange between data and laboratory scientists helped a lot to identify the needs of both sides and to achieve optimal results. We believe that cross-domain collaboration is a groundbreaking concept to overcome limitations, e.g., in therapeutic research.

Vanessa, what advice would you give other scientists/data scientists looking to establish collaborations?

VLR: It is important to have a common goal in mind and to continuously consult with each other. Communication and mutual feedback are the most important and limiting factors. As a data scientist, take the time to explain what modern data science is already capable of so the collaborators do not limit their ideas from the start. Vice versa, try to get the best possible understanding what types of data are relevant and what the key areas of possible new insights are that could drive experimental progress.

Marcin, how do you stay updated with both advances in data-science techniques and the advances in the field/domain that you work in?

MN: I keep updated with advances in research areas of interest by reading conference proceedings and journal issues as well as the latest preprints published online. Moreover, I enjoy going through the data and the code that were made publicly available by the authors. Last but not least, discussion of the latest research advances with peers gives me further insights into the up-to-date solutions to the most pressing problems in my fields of interest.

Marcin, which of the current trends in data science seem the most interesting to you? In your opinion, what are the most pressing questions for the data science community?

MN: From the perspective of my research interests, the development in the adversarial robustness field excites me the most. Moreover, there are many other interesting trends in data science such as fostering interpretability and reproducibility of machine learning research or promoting sustainable and efficient solutions that reduce the carbon footprint and financial cost of running computationally expensive experiments. In my opinion, advocating responsible and ethical AI research is one of the most pressing questions for the data science community.

Bruce, what is the fun part of being a data scientist?

BS: The rush a biological scientist gets when their experiment that they have been troubleshooting for months finally produces meaningful results is the same elation that a data scientist feels when their code runs to completion and returns what was expected. Being able to take the same collection of data and manipulate it in a creative, logical way to identify new, useful patterns is invigorating.

Bruce and Marcin, have you ever used your data science skills in your personal life? If yes, how?

MN: Actually, I use them quite often, especially to manage my finances, i.e., to analyze incomes and expenses, or to track my fitness activities.

BS: Data science skills have a wide variety of day-to-day applications such as being able to identify troublesome applications and bottlenecks in my home network as well as being able to automate repetitive tasks, especially those involving Excel spreadsheets. Currently, I am working on developing a machine learning model to find recipes I would enjoy online.

Bruce, what attributes make a data scientist successful?

BS: Data science requires two major skills: a strong foundation in an applicable programming language (Python, R, Julia, Java, etc.) as well as a decent understanding of both mathematics and statistics. Depending on the work environment, there are several other factors that can help one to succeed, but being able to find the faults and issues within a dataset, fix them, and create meaningful algorithms to apply to it are highly dependent on the skills mentioned above.