Episode Transcript
Announcer: Examining the latest research and telling you about the latest breakthroughs, the Science and Research Show is on The Scope.
Interviewer: It can be difficult to help a patient if you don't know what's wrong with them. My guest, Dr. Mark Yandell, is a computational biologist who specializes in finding the genetic causes of human disease. Recently published in The American Journal of Human Genetics, he reports on an exciting new approach to solve the most difficult medical cases. Dr. Yandell, what problem were you trying to solve?
Dr. Yandell: There's a lot of work underway both here at the U and worldwide right now, that involves trying to diagnose, for example, a sick child in the hospital by sequencing their genome to ask whether or not they have a genetic disease. The problem is that when you only have that single child's genome sequence and even if you have, say, a parent or sometimes both parents, there's simply not enough information in those genome sequences to come to a conclusive diagnoses. Now, every physician knows, they already know a lot about the cause of the child's illness often just looking at them, or looking at their test results, lab test results, etc. And so what Phevor tries to do is combine that phenotype information - information about the symptoms of the illness - together with the genome sequence for more accurate diagnosis.
Interviewer: What is Phevor?
Dr. Yandell: Phevor is a tool that, in a lot of ways, resembles this new class of computer programs that do things, for example, like play Jeopardy like IBM's Watson or what have you. And so what these tools do is to bring previous information to bear on a problem by doing very rapid searches of databases and things called ontologies that model human knowledge to give human-like responses to questions. So Phevor takes an approach that's similar both in spirit and in detail to this problem, and it tries to use existing biomedical knowledge about disease, symptoms associated with particular disease, and genetic lesions that produce those diseases. And so that gets combined in ways that it hasn't been previously been possible and has never been done before to solve that problem.
Interviewer: OK. So you're kind of compiling together these many layers of information to find the most likely candidate.
Dr. Yandell: Yes. Or different domains of knowledge.
Interviewer: OK. Phevor is spelled P-H-E-V-O-R. What does that stand for?
Dr. Yandell: The name, which is a pretty good one I think, came out of a contest in the lab with the students trying to think of a good name for the program. Unfortunately, the acronym is a bit crazed. It actually is an acronym for Phenotype-Driven-Variant-Ontological- Reranking-Tool.
Interviewer: OK. Yeah, I guess you could find those letters in there somewhere.
Dr. Yandell: Yes. They all had a really good time with it. Phevor was the one I chose - I thought it was a pretty good one. It had the health aspect to it and the idea that it was hot, but then it stands for something that's a little bit unobvious.
Interviewer: So what were you able to accomplish? In this study, you look at some real life cases.
Dr. Yandell: One of the things presented in the publication is two cases here in the hospitals at the U. This was a collaboration with my colleague, Steve Guthery, and Karl Voelkerding. First, a child who had been born from two very healthy parents with really quite a grisly intestinal syndrome - basically very, very life-threatening. We were able to identify there, even though we had the sequence of that child and the healthy sibling and the parents, the diagnosis had not been clear, and so Phevor was able, together with those genome sequences to come to what we believe is a very convincing diagnosis as to the nature of the actual disease. There caused by a spontaneous mutation common to the child not inherited by the rest of the family members. Another case was a child born with what appeared to be a relatively straight-forward case of a liver disease and that fit very well with a classical, inherited form of that disease, but standard diagnostic tests came back negative for no mutations in the gene that is commonly mutated to cause that disease. And so Phevor suggested very strongly, given the data, that it was that gene.
Interviewer: Armed with that knowledge, what can you do for these patients?
Dr. Yandell: Well, a lot of times it's just knowledge in the case of the child with the liver disease. If the conventional tests would have said "no, this child doesn't have this disease" then the physicians would have embarked upon a whole series of suggested treatments and tests, all of which were just inherently misguided and wrong from the get-go because they were proceeding from a false conclusion. So it's a whole realm. And there are also issues with certainty. I know, from talking to my clinical colleagues, that the families are greatly relieved in many cases - at least there's some knowledge as to what the disease is. I mean, as a parent myself, what's scarier than your child having a life-threatening disease and your physicians are just shaking their heads and throwing up their hands with "we don't know"?
Interviewer: So what're the limitations of Phevor at this point?
Dr. Yandell: Well, I think, like anything else, one of the biggest limitations comes from the inherent limitations of computers, which is that the world of computers inherently is one of 90 degree angles, straight lines, and perfect circles. And, of course, the world that we live in isn't that. So we shouldn't just believe what a computer tells us. Pretty obvious fact, but it's forgotten sometimes, I think. Secondly, I think there are limitations with regards to, simply, the amount of knowledge available - our knowledge is imperfect - and also our ability to employ different kinds of knowledge within Phevor. We're working right now, very hard on that. I mentioned earlier about providing information about which genes are expressed in the tissue. We're also getting information about gene interactions, etc., and so we're trying to extend it - to take into account - as much information as we can with the idea that it'll get better and better at the diagnosis game as we get more information.
Interviewer: So you've been working on this problem of finding the genetic causes of disease for quite some time, enveloping computational tools to be able to do that. What fascinates you about that subject?
Dr. Yandell: Well, there's a lot that fascinates me about it, but I think the most interesting thing about it is the idea that we can compute on a DNA sequence - the fact that, in a certain sense, you can think of us as being "made of words". Three billion base pairs - A's, T's, G's, and C's - and you start grouping those four letters in to longer sequences and all of sudden it really has something to say about who we are, why we're healthy, why we're not, etc. And the idea that, instead of being restricted to actual physical experimentation in the lab, we can directly compute on that and get real answers has fascinated me for 25 years and continues to fascinate me.
Interviewer: So what do you see in the future of this work? Do you see it being applied in the clinic?
Dr. Yandell: Absolutely. We see, increasingly, immediate - what they call, translational applications which is where you take a lot of this esoteric - what would seem at first glance basic science that would have no immediate applicability - and applying it immediately in the clinic. There are a lot of projects underway to extend its use. I think we're going to see more and more these kinds of tools including Phevor playing sort of a first-round defense against, basically, human suffering and mis-diagnosis in the hospitals going forward.
Announcer: Interesting, informative, and all in the name of better health. This is the Scope ǿմý Sciences Radio.