A Trajectory to Biomedical Informatics

Isaac S. Kohane, 9,22.04

I've been often asked how and why I became interested in biomedical informatics. I always then try to steer away from the topic as fast as I can because I've not thought too much about it. So, it is probably a sign of advancing age that I've decided to spend an hour trying to introspect and articulate the trajectory that has brought me to this particular confluence of computation, medicine and biology. I also hope this provides me with a convenient way of answering the students who ask me what are the opportunities for them in biomedical informatics.

From early in high school I was drawn to the challenges of biology but simultaneously repelled by the seemingly ad hoc and anecdotal nature of scientific investigation in that discipline. To my naive eyes, Physics and Chemistry presented the canonical examples of what it was to have useful and testable scientific models. The collections of disparate factoids that characterized biology lacked the beauty and power of these older disciplines. And so, it was with surprise[1] in my first year in the U.S.A, that I was introduced to a mainframe IBM computer accessible to undergraduates at Brown University in the 1970's. The availability of this resource in a congenial, if nerdly, atmosphere revealed opportunities that no formal curriculum in biology ever would have. Under the mentorship of the late Professor James Kidwell, I realized that research in biology could be systematized and leveraged by the use computational methods to simulate and explore previously intractable hypotheses. The solution of previously intractable probabilistic models by computational techniques, particularly of evolutionary processes over time was tremendously appealing. It also was a sobering introduction to the ease with which computational models could be tweaked to fit any prevailing theory.

Subsequently, during my MD-PhD training at Boston University, I worked on my doctoral research with Peter Szolovits at the then Laboratory for Computer Science (now part of the Computer Science Artificial Intelligence Laboratory) at MIT on knowledge representation for automatic medical reasoning. In particular, I was interested in how to diagnose patients based on their history. At that time, and sadly to this day, so-called expert systems classify/diagnose patients based on the presence or absence of signs or symptoms rather than on the order and progression of these findings. I developed a novel temporal constraint propagation system, called the Temporal Utility Package that I then used to create a prototype that was capable of diagnosing diseases based on temporal patterns of presentation. This work formed the basis of subsequent research in trend detection in clinical data in the intensive care unit and in pediatric growth.

Admission to the Children's Hospital pediatric residency training program, in 1987, introduced me to an environment that remains dear to me to this day.  The combination of researchers who had defined diseases and who cared for children and their families is viscerally and intellectually compelling. Interestingly, one my two interviewers for the residency program was Alan Guttmacher, at the NHGRI, which represented a kind of foreshadowing that I did not appreciate until later. During the time that I was privileged (and frightened) by the responsibility for our precious pediatric patients and their stressed families, I had a chance to reflect on the relevance of my doctoral research to the practice of medicine. It was all too apparent that with so little medical information available in electronic form, any automated decision support would be working within a vacuum. Furthermore, my unmet need as a pediatric resident was not clinical knowledge, (cracking open the textbook before I saw each admitted patient was surprisingly effective in that regard) but relief from the time pressures caused by the inaccessibility of patient data, coupled with burdensome clinical documentation requirements. As a result during my subsequent fellowship in Pediatric Endocrinology at The Children's Hospital (currently named the far more measured and modest  "Children's Hospital, Boston") I started a multi-year exploration of the societal and technological challenges of electronic medical record systems, at the single hospital level (the Clinician's Workstation), at the  healthcare system level (the "W3-EMRS" system) , and at the patient-consumer level (the "PING" project). It was also around that time, 1992, that I founded the Children's Hospital Informatics Program (CHIP) energized and empowered by a significant grant from the National Library of Medicine.

One day, in 1998, I crossed paths with one of the former interns at Children's Hospital, when I was a Senior Resident: Todd Golub. Todd and I updated each other about our work and it was then that I learned about his research on the clinical applications of expression microarrays. The notion of almost fully automated and affordable measurements of a large fraction of the transcriptome was compelling and that same day I decided to create a major focus in the CHIP research program in functional genomics. In this I was blessed with the arrival of one of the new fellows of the NLM-funded training program, Atul Butte. Together and in collaboration with Todd Golub and the NCI, we developed some insights into finding drug targets using microarray and chemosensitivity data. My research in functional genomics also allowed me to return to the topic of by undergraduate and doctoral research: modeling the dynamics of gene expression over time. Most recently, this has led me to study the processes of development and aging in the genomic and temporal contexts.

Despite my focus on functional genomics and the study of dynamics at the cellular level, I have become increasingly concerned about the obstacles facing medicine's attempt to take advantage of the "insurmountable opportunities" presented by the advent of industrial scale biology.. If all the enthusiasm about the revolution in medical care is to bear fruit them somehow, we are going to have to fix a number of "broken" aspects of the medical system. These include: the relative ignorance of medical professionals in the clinical application of the fruit of the genomic revolution, the obstacles to clinical research in the genomic era. To this effect I have worked with my colleagues on developing educational programs (e.g. our new course on Genomic Medicine), and a pre-doctoral program (funded by the NHGRI) to train individuals with strong backgrounds in the quantitative sciences to become leaders in bioinformatics and genomics. Mot recently, the challenge of enabling clinical researchers to effectively leverage the "new biology" using computational tools and methodologies led me to work with John Glaser, CIO at Partners Healthcare Systems to propose a National Center for Biomedical Computing: Informatics for Integrating Biology to the Bedside.The recent funding of this proposal suggests that our vision about the role of biomedical computing in clinical research is widely shared.