For translational medicine to succeed, biomedical informaticians need to be part of the team, Indra Neil Sarkar from the University of Vermont's College of Medicine wrote in a paper published in the Journal of Translational Medicine in late February.
Sarkar, who is director of biomedical informatics in the university's Center for Clinical and Translational Science, said biomedical informaticians should be involved wherever translational barriers need to be surmounted, including support for clinical decision-making, standards development, and developing methods in natural language processing, information retrieval, and electronic health records.
BioInform spoke with Sarkar last week at the American Medical Informatics Association Summit on Translational Bioinformatics in San Francisco. He will be chairing next year's summit. The following is an edited version of that conversation.
What was the meeting's highlight for you?
Perhaps the most amazing thing is that this is happening so fast. When we started this meeting four years ago, we had notions of gene chips and now it's whole-genome sequence. We knew we would get there but never this quick.
There was a time when all of GenBank fit on three floppy disks that people passed around and then all data transfer became too big for so we have invested in infrastructure to pass data around. I have tried to get a large dataset from a colleague in the Midwest, so I started the ftp job and burned it to a hard drive and FedExed it, which got there before the file finished transfer via ftp.
We have far exceeded Moore's law in terms of data output, especially with all the metagenomics projects out there in which people sequence buckets of samples. Traditional microbiology has said, 'You have to culture it before you sequence it.' Now the approach is, 'I don't care if I can't culture it, I am going to see if I can sequence it.'
Is the infrastructure being built for metagenomics, say for the Human Microbiome Project, going to be able to be transferred in some way to human genomics projects?
That idea is very close to my personal challenges in some of my earlier work in biodiversity research. Both communities can learn from each other, while researchers in biodiversity are focused on conservation and we are concerned with health.
Microbiomes across species will differ and the infrastructure will need to be robust enough for that. Both the amount of data and complexity matter. Looking at the mouse gut or human gut will yield much data. After all, bacteria are tiny.
We have found our calling. We've been developing ontologies for many years, particularly in the biomedical community and the microbiology community. We have built GenBank, [Stanford University researcher] Russ [Altman] has done work with [the Pharmacogenomic Knowledge Base] and we all knew it was going to be important someday. The day is now. It's really amazing. So it's hard to predict what next year's program for the [AMIA Translational Bioinformatics] Summit will look like.
Are you expecting many more attendees, given the interest in electronic medical records?
AMIA is making very good headway in this community. There are only about 2,000 to 5,000 of us who are formally certified informaticians. Someone at a medium-sized hospital might have the title informatician, but they are really IT and not informatics. And if there is one term I have an issue with, it's 'health-IT,' it's the misnomer that we're stuck with.
Informaticians need IT, but if you ask me to fix something on my computer, I am not a hardware guy. It's a way of thinking. Many informaticians here fell into the field by accident. I grew up with computers and had strong ideas about the role of computers in microbiology. I am not a physician but I have a lot of interest in medicine. I had the notion I would spend most of my time in the lab using a computer on the side, but I have a dry lab, I don't maintain a wet lab.
Creating data is not the problem; it's understanding the data, and that is where AMIA fits in with its history. Its main meeting is more oriented toward electronic health records. This meeting is, 'Let's take EHRs and basic bioinformatics for granted. Now what can we do with the data?'
What types of questions come up in that next step?
Right now people want to go from genotype to phenotype. We thought that was going to be really easy. How hard can it be? If an animal has furry legs, you knock out that gene and the furry legs go away. But in the context of diseases it is hard, because of the complexity.
This is where there is a constant challenge with animal models, because they are bad models for humans. Some of the keynotes here alluded to that. Lab animals are inbred, but no two people are the same in and of themselves. We are all wild types.
We don't just develop nice new databases to separate data. We are data mushers. We have lots of different types of data and we hope it tastes good. An informatician is like a good cook, who understands the ingredients, creating an amazing sauce.
How far along is the integration of genomic information in medicine?
It's hard to say, but for certain well-known Mendelian traits, it's a no-brainer. But even with slightly more complex disorders, not cancer, it's less clear. Why do some people get completely decimated by a cold and other people get the same cold and their immune system handles it fine? There is no good answer to the question.
Will informaticians will be drawn into helping to interpret genomics in the practice of medicine?
The spectrum of medicine is now from bench to bedside. And from the clinic it has to reach the community — for example community hospitals. Teaching hospitals are wonderful but can be isolated institutions. You have done something really important for society when you have taken something out of a teaching hospital to primary-care physicians. Then that can be used to influence policies.
Smoking might be one example. We have a lot of laboratory research, evidence from the bench, and it has made it to the hospitals and primary care physicians and policymakers. Underneath all of that, getting across that space, that's the informatics. It's happening behind the scenes, while the clinicians, biologists, and policymakers are in front.
Where do you best position yourself as a tool developer in that translational space?
The hardest thing about being a biomedical informatician is that you have to have a foot in at least one of the domains. I have a microbiology background, many attendees here are clinicians, and it's about solving problems in your domain. There are informaticians who make beautiful mathematical models for other informaticians. Then there are informaticians who solve problems for domain scientists, creating tools for their consumers.
Where do ontologies come in? It seems as if many speakers mentioned them as part of their projects.
I do think ontologies are one of the key universal glue elements that allow these systems to talk to each other. Developing ontologies is hard. You have to completely divorce yourself from what you think you know and understand how a computer works because they only do what you tell them. Ontologies have to be very well-designed, especially when you handle large volumes of data. But the perfect informatics solution is not one that is going to be 100 percent computation. A human has to be involved in data analysis.