When it comes to boldly going where no man has gone before, we often think of frontier expeditions traversing the depths of outer space. But there’s also much to be explored on Earth—specifically in the compounds that play crucial, if often unnoticed, roles in our daily lives.
From paint pigments to promising drug candidates, there appears to be a compound for just about every need, but scientists say that the number of compounds known today is far eclipsed by theoretical calculations. Given the seemingly infinite combinations of elements, the chemical space—or the set of all possible organic compounds—has been estimated to contain 10180 compounds, more than twice the number of atoms in the universe.
Given the sheer scale of compounds that remains to be discovered. it’s no wonder researchers are looking to adopt quicker and more efficient methods to study and apply compounds and molecules. One such solution comes in the form of artificial intelligence (AI) and machine learning (ML)—a field that A*STAR National Science Scholar Jacqueline Tan hopes to harness to provide better ways for researchers to advance chemical research.
Supported by a strong belief that science is meant to improve the world for future generations, Tan’s postdoctoral studies at the Massachusetts Institute of Technology (MIT) currently focus on ML methods in theoretical chemistry, as well as developing greener catalytic reactions.
In this interview with A*STAR Research, Tan shares the inspiration behind her work as well as key findings setting the foundations for her future research.
1. What inspired you to be a scientist and why did you apply for the A*STAR National Science Scholarship?
In secondary school, my science teacher signed me up for a science and design competition. My team designed an anti-spill customizable hot water carrier that can prevent scalding and was crowned runner-up.
This experience made me realize how important educators are and convinced me that science was a way to improve the world. From then on, I believed that science and education go hand-in-hand—so applying for the A*STAR National Science Scholarship was not a difficult decision. Since then, the scholarship has helped me pursue my interest in science and equipped me with the necessary skills to propagate what I’ve learned in turn.
2. You studied quantum and computational chemistry for your PhD. Why did you later focus on machine learning and theoretical chemistry?
Once you step into the realm of computational chemistry, you’ll find that you’re dealing with data science, physics, engineering, math and of course chemistry. It truly becomes interdisciplinary so it’s not really a change—more of a transition into other areas of science that tie in closely with my current work. Science is really exciting, so you have to continuously learn and adapt to make the most of all these available tools.
3. Could you elaborate on your research findings in radical cation chemistry?
For one project, I worked with another experimental group at the University of Oxford to uncover how introducing intramolecular hydrogen bonding can significantly reduce the barrier to the rotation for non-biaryl atropisomers. These molecules are key in applications like medicinal chemistry and supramolecular chemistry.
In a project I worked on with Robert Paton, my supervisor at Oxford, we found that by taking out an electron for atropisomeric biaryls and creating radical cations, the resistance to racemization is reduced dramatically, so much so that some reactions can occur at room temperature—contrary to prior research. We also explored the mechanism of how these radical cation molecules operate and such a discovery will have implications in pharmaceutical applications.
4. Could you tell us more about the most exciting project you are working on now?
Currently, I’m working on electrocatalysis, the class of catalysts that increases the reaction rate of electrochemical reactions. We are focusing on oxygen reduction reaction (ORR), or the process of harvesting electricity from the conversion of O2 and H2 to form H2O. So far, this process only recovers 25 percent of the total energy of the reaction.
To achieve a future renewable hydrogen economy, it is necessary to decrease the cost, improve the stability and increase the efficiency of hydrogen fuel cells. It is therefore interesting to explore the use of other metals rather than just state-of-the-art platinum alloys, which is one key aspect of our project right now.
5. What key problem do you hope to solve through your current research work?
Broadly speaking, you can see many different fields of research trying to incorporate ML and AI into their projects. ML can help in handling large amounts of data and a good working ML code can perpetuate important information as the code ‘learns’ from what it is doing—growing more sophisticated in the process.
6. Why do you think science communication is important?
Science communication and science education are both important to me. I’ve always looked for opportunities to mentor others and share my research in a more palatable manner with the general public.
Before I left Singapore, I mentored fourth-year university students during their summer program. When I was at Oxford, I worked closely with a company that produces scientific toys for youths to inspire them about STEM research. While at MIT, I volunteered at the Cambridge Science Festival and at the MIT Museum to showcase amazing research to adults and children alike. Across the different universities and countries I’ve been to, the audiences I’ve engaged with have always had a strong interest in science communication.
I even incorporated science into my wedding! Over the years, the medium of communicating and promoting science has shifted online. But I believe that scientists will adapt to these changes as long as they are excited to share their findings with others. After all, excitement is contagious!
7. How has machine learning changed the way we understand chemistry?
The use of ML can accelerate chemical and drug discovery by exploring the combinatorial space of chemical motifs more effectively. Just to put it in perspective, there are an estimated 1060 possible small to medium-sized molecules to explore. Traditional methods could not possibly tackle this astronomical number. ML, together with high-throughput virtual screening and experimentation, can at least help us approach this difficult task.
Back when I was doing my research internship at A*STAR's Institute of High Performance Computing (IHPC), I worked in polymer research and made use of simulation tools to understand how the aromatic substitution of a surfactant molecule can help with encapsulating materials for drug discovery. We also investigated how the ring size of cyclic ketene acetals can help with polymer biodegradability, which in turn could help with reducing negative environmental impacts.
As our technical capabilities increase, can you imagine how much faster we can achieve these results, and how much deeper we can go in our scientific exploration?
8. How do you see your research evolving in the next decade?
Based on what we can achieve now with ML and AI, and at the speed that we are going today, the future is full of possibilities. But at the end of the day, I do think sometimes we need to take a step back and remember that we are trying to improve the world for the next generation—something I realized for myself when I became a mother. I truly hope that my child will be able to thrive in the future that we are building for them.
