John Quackenbush, MD, is a professor of biostatistics and computational biology at Harvard’s Dana-Farber Cancer Institute. His research focuses on integrating multiple sources of information – genomic data, clinical information and knowledge gleaned from the published scientific literature – to map out the networks that drive processes within cells and to understand how these networks change as diseases such as cancer develop and progress.
What kind of research are you currently focused on? What most excites you about the work you’re doing?
I work in computational and systems biology. My goal is to develop a more complete understanding of the complex nature of human disease, and to do this I take advantage of the large-scale, multifaceted data that we can now generate. We are using this data to build models of cellular processes. These models are allowing us to explore the multifactorial nature of human disease, to identify new potential therapeutic interventions and to study the differences in disease progression we have learned exist between males and females. We have even extended our methods to create models for any individual in a population, which is opening up new opportunities for network-based precision medicine.
How would you describe the current state of healthcare innovation?
When I am asked to talk about the future of healthcare and biomedical research, I like to make the point that discovery and innovation in any field is driven by one and only one thing: data. Working in health and biomedical research today, I am excited by the unprecedented quantity of data now available. But enormous amounts of data are useless without the ability to make sense of it in a way that advances healthcare. And this is more difficult than it seems. We struggle with incomplete data — data derived from fragmented systems, data that don't give us a complete picture. Our biggest challenge is that we have a messy data problem.
When will we start to see practical applications arise from advanced computational techniques?
There is a lot of excitement about leveraging biomedical data by using techniques such as deep learning to address a host of issues in healthcare. But, so far, the failure of deep learning systems to deliver, reflects both the complexity of biological systems and the lack of proper training data with outcomes. We need to avoid the hype and carefully identify those problems that we are equipped to solve.
If we want to address the complexity of human disease, we also need to bring our understanding of biology to bear on our models. This requires supplementing deep learning with human understanding. In 1997, Wolpert and Macready published a paper describing what they called the “No Free Lunch” theorems, pointing out that black box algorithms generally fail to deliver optimal solutions to complex problems. One instead needs to build methods that start with a reasonable guess as to how the system under study operates in real biological systems.
Looking back at your own career ― what particulars shaped your own development? What promoted or enabled your own productivity?
I am driven by the desire to understand the fundamental principles of the world around me. This is what led me to pursue a degree in physics. The most important thing I learned in my studies was how to attack a problem, and it is an approach that I still take today: Reduce a big problem into smaller problems that are easier to solve, and then figure out how to integrate the solutions in a way that is consistent with our understanding of how systems work. This approach leads you away from having to answer nearly impossible questions like, “Is this model right?” and instead puts the focus on answering the question, “Does this model inform our understanding of the system we are studying?” This second question is far more useful and far easier to answer.
My team has also played a deep role in my professional development: I am fortunate to be surrounded by incredibly bright, talented, driven postdocs and students who create an environment where they can work collaboratively and cooperatively. They are not my trainees, they are my colleagues, and we operate in an atmosphere of mutual respect. Many of our best ideas come from brainstorming solutions to difficult problems, but the key to success is you have empowered everyone on the team to be willing to tell you, and each other, that 90 percent of the ideas floated are stupid (and why), so that the 10 percent that we agree on have a pretty good chance of being successful.
[See how the right data platform can improve productivity and speed critical information to providers – Link to C26]
Lastly, my productivity, and that of my team, is driven by the question, “Does anything we do have the power to advance the field and utility in treating disease?” Nearly every student who walks through my door wants to find biomarkers to predict survival in disease. I immediately tell them that will not be of use. If we predict that someone has a good prognosis, he or she will be treated. If we predict a poor prognosis, well, he or she will almost certainly be treated. Rather, I urge them to look at the data and develop more useful, more informative questions to answer. The best discoveries typically begin with well-posed questions.