What do you get when you bring computational and molecular biology, data science and engineering all together? That’s what Daniel Reker and his lab are investigating. Daniel Reker is an assistant professor of biomedical engineering at the Pratt School for Engineering who joined Duke University’s Department of Biomedical Engineering in 2020. In his research, Reker uses machine learning and modeling to explore how drugs, excipients and nanoparticles behave once they enter the body. Duke’s Office of Government Relations asked Reker five questions about his work and its implications for the future of big data-driven protocols for precision medicine and personalized drug delivery.
“It was clear to me that both [machine learning and pharmacology] were incredibly powerful and that a lot of impact can be achieved by combining them. To me, it’s the perfect combination of intellectually challenging questions that have a very clear societal impact.” – Daniel Reker
What influenced you to study machine learning and its applications in designing drug therapies?
I have always been interested in both computer science and chemistry. Computer science enables us to design smart algorithms to solve problems and chemistry can explain complex, real-life phenomena from molecular interactions.
My determination to combine these two fields productively led me on a winding educational path, weaving different programs that emphasized the computational or chemical sciences. This allowed me to learn a lot about these two disciplines and collaborate with and learn from amazing peers and mentors that helped shape my path. I realized that machine learning and pharmacology are specialties that were particularly fascinating to me since they enable the autonomous development of predictive algorithms and the design of molecules to treat diseases. It was clear to me that both were incredibly powerful and that a lot of impact can be achieved by combining them. To me, it’s the perfect combination of intellectually challenging questions that have a very clear societal impact.
How does machine learning aid researchers’ ability to develop new drug therapies? What does your research and other research in the field mean for future developments in designing drug therapies?
Researchers have proposed impactful approaches to use machine learning to improve all stages of therapeutic development – from refining our understanding of the disease to analyzing clinical trial data. Our work is focused on drug discovery, development, and delivery, where we aim to identify new molecules that have useful therapeutic properties, optimize them, and create carrier materials to ensure that the medication can reach the desired organ.
Machine learning algorithms can learn from past experiments to predict the outcome of future experiments. This enables us to focus our experiments on the most promising candidates, and thereby save time and resources.
Can you provide an example where machine learning proved useful?
One particularly promising direction is the use of “active machine learning”, where the machine can ask a scientist to perform experiments that the algorithm is uncertain about – which are most informative and thereby can generate data that helps the algorithm improve future predictions.
More recently, we have started to use machine learning to predict a wide range of other properties of drug candidates, such as potential side effects or ease of synthesis. Such machine learning models enable us to anticipate potential “dead ends” in our drug development campaigns and thereby de-risk the process. 90% of all drug development campaigns fail, so reducing risk could help us to bring more life-saving medications to patients.
In the future, I believe that we will be able to combine thousands of machine learning models to predict all the positive and negative effects medication could have on a specific, individual patient. Such systems will enable pharmaceutical companies and clinicians to develop and prescribe the safest and most effective medication for every individual patient – thereby providing an important tool to personalize medicine and making pharmaceutical research and development more equitable.
What findings in your research have you found most significant?
The most significant part of our work to me is when we can see a “real world” effect of our predictions. For example, it is always exciting to see when our designed molecules change the behavior of cells and proteins in the laboratory. Other examples are predicted side effects of medications: many patients reached out to us to thank us for helping them better understand these issues and for raising awareness about complications that they or a loved one have been struggling with for decades. Similarly, I still vividly remember when we were treating mice with our computationally designed nanoparticles and saw them improving much faster compared to the mice that received the standard treatment. It is this real-world impact that has drawn me to study pharmacology, and it is what keeps me motivated every day.
What are you looking forward to most in your time at Duke?
Aspects that set Duke apart from other institutions are the sense of community and the collaborative climate. The work of my laboratory synergizes with many other scientists and, although we have only been at Duke for a bit over a year and started working here during a pandemic, we have already established seven collaborative projects with colleagues in Pharmacology, Biology, Chemistry, Biomedical Engineering, Environmental Engineering, and Immunology. Especially the proximity between the university and the hospital provides unique opportunities for translational research at Duke.
Another big advantage for Duke is the incredible students that are smart, driven, and creative. These students drive a lot of our research and are also excited to explore new research directions. I have recently established a new class “Machine Learning in Pharmacology” at Duke that is very popular. We are also setting up a “Biomedical Data Science Master’s Certificate” at Duke together with some of my colleagues. These are just some examples of how we are working to further enhance the training at Duke. Seeing the student’s excitement and passion for the field is hugely rewarding and gives me hope for a brighter future.
What would you like to tell students who are also interested in computer science and its biomedical applications?
It is a very exciting time to work in computational biomedical sciences. Not only has computational power and algorithms improved in the last years, but we now also have access to larger datasets thanks to increasing automation from biomedical experiments. Maybe even more importantly, the quality of biomedical data has been dramatically improving and technologies such as CRISPR, cell painting, single-cell biology, and broader access to sequencing technologies are improving our understanding of the biological systems we are trying to model and treat. All these improvements have created a big hype around AI-driven drug development, which is currently resulting in rapidly growing educational and job opportunities. It is very interdisciplinary work that profits from many different perspectives, and I have seen students with many different backgrounds be very successful in this field. My advice for students is to seek out learning opportunities not in one area alone but in both biomedical sciences and computation as well as at their interface. For example, our laboratory has access to a cluster computer, but we also run our biological experiments in our wet laboratory. I have specifically designed our laboratory in this way to increase our scientific impact but also to provide a training environment where students are engaged in both computation and experiments. The future of our field will require scientists that are “multilingual” to translate between fields, identify relevant medical challenges, and tailor appropriate computational algorithms to solve them.