August 8, 2007 | At CHI’s 2007 Beyond Genome conference*, held in June in San Francisco, several keynoters with expertise ranging from proteomics and systems biology to RNA and molecular modeling, participated in an opening panel discussion moderated by Bio•IT World Editor-in-Chief Kevin Davies. Here are some of the edited highlights of that far-reaching interactive discussion, which highlighted several outstanding challenges still confronting the post-genomic era.
Q: What are some of the recent successes of systems biology?
David Galas (Institute for Systems Biology): It’s probably easier to describe the challenges! ... Systems biology is really not new. The thing that’s new — modern systems biology — has been transformed by the technology. Physiologists have been doing systems biology for 100 years or so... We can do global measurements, measurements of global quantities with quantitation that’s unparalleled. We’re actually able to look at dynamics in biological processes in ways we’ve never been able to do before. That really changes things fundamentally — the kinds of questions that you can ask are changed. When you perturb a system or cell, for example, you want to see how an enormous number of variables are changed... Just understanding the complexity of what we now think of as phenotype, or doing genetics, is completely different. The key [in genetics] is figuring out what the phenotype is very precisely... Defining that phenotype and joining systems biology with the power of genetics, we haven’t really learned how to do that yet.
Michael Liebman (Windber Institute): Systems biology, which is really old, has benefited quite a bit from the technologies in different areas, but I think also that there’s a lack of understanding of the complexity of the system... We’re applying systems biology to deal with clinical problems. Complexity in the clinic is very different than complexity in the laboratory. Stratifying the concepts of disease, the concepts of what a patient actually is, are critical aspects of where systems biology needs to head. What it’s done successfully is to lay the groundwork for building from a bottom up approach...but we need to focus on starting with the intact system, and taking apart the information that’s available — clinical, imaging, information that’s measured at a much more macro level, information that’s not always as quantitative or as clean as we’d like it to be when measured in the lab, but information that relates to the treatment of patients on a day-to-day basis.
What are some of the successes in the field of proteomics?
Josh LaBaer (Harvard Institute of Proteomics): Proteomics is the most important element — I’m a bit of chauvanist there! There’s nothing that captures the complexity of an organism more than the protein complement. There are many more proteins than there are transcripts or metabolites. That doesn’t even account for post-translational modifications... We’re still very early in that process. We’ve seen lots of technical progress, where most of the energy’s been placed, in terms of detecting rare species, identifying polypeptides. Where I think we need to work is in protein function. There are 20-something thousand genes in the human, [but] countless more proteins. Many have been assigned functions bioinformatically [but] we don’t really know what they do... People look at protein interactions in a very ball-and-stick, A binds to B, model. We don’t just need to know that. How fast does A bind to B, does A come off B? What is the affinity? How does that get changed when A gets modified by a phosphate group? Getting depth of knowledge is important.
What are the current trends in the world of RNA?
Adrian Krainer (Cold Spring Harbor Lab): Just by looking at genomes that are being sequenced, it’s apparent that there are a lot of non-coding and coding transcripts — a lot more than we anticipated a few years ago. We’ve certainly come a long way from the view of “one gene, one enzyme.” It’s become apparent that each gene can produce multiple transcripts, several producing proteins with different properties, others play a role in regulation. One of the challenges is understanding, What is a gene?... It’s a real challenge. You can look at genome browsers, but people have historically this tendency to imagine that there’s a standard isoform for a gene. A lot of that is historic from 20 years ago when we were cloning cDNAs... We need to understand that at the genome level much better.
Christopher Lee (UCLA): RNA has historically been the dark matter of the genome... The first step is... high-throughput analysis. The initial EST and genome project gave some grasp of the sheer magnitude of alternative splicing in the human and other genomes. This is uncomfortably but steadily progressing into the microarray field, where we’re hoping that soon, everything you can do with gene expression you’ll be able to do at the level of individual isoforms. Even the question of what level of resolution you should be looking at... We have no idea if the regulation of these events holds true for alternative splicing, RNAi, we don’t know if the level of regulation is at the level of the whole organ or the activation state of an individual developmental state of a cell... We have this very long pathway ahead of us.
How would you assess the significance of the Interactome Project?
LaBaer: It’s an important first step, they are trying to identify protein interactions, but it’s very, very crude. There’s a lot of false positives and negatives, they only measure end-state based on technologies that are used — pull down experiments, yeast 2-hybrid system, it has to reside in a yeast if it’s going to work. So they’re a start, they’re forcing us to think about the mathematics of it, and how to organize the information. But ultimately we’re going to want to understand the kinetics of interactions, and that’s going to be a big challenge.
What is the role of model systems in systems biology?
Liebman: We see very significant differences in various [tumor] cell lines, without even going into an animal model... We look at pathways frequently, homology at the molecular level, but what we need to understand is, Do they function the same way? Just because [enzymes] are structured in the same topology doesn’t mean the pathway behaves in the same way. We need to couple a lot more of the simulation. Coagulation is frequently evaluated in dogs, but we need to understand that the pathway doesn’t behave the same way, even though the enzymes are similar. That kind of modeling needs to be added to everything that we’re doing at the genomic level. That kind of detail needs to be incorporated into the analytics to enable us to see what the best model of a system is.
Galas: We’re just really beginning. There’s so much we don’t know... Even though this complexity keeps unfolding, e.g. the RNA area, it’s clear that the processes, the technology, the tremendous pace of development, in some sense this is a technical problem, which will then lead us to an enormous mathematical and computational problem. Nonetheless, it’s completely doable, but we’re not even close. Taking model organisms that are actually used to reflect some sort of surrogate for human response is going to take an enormous amount of work. At the Institute of Systems Biology, we generally think of model organisms as a way to look at something simpler to work out methods.
What are some of the biological questions you want to address using molecular dynamics simulation?
David Shaw (Columbia University): The part I’m more confident about is the short term. We’re going to spend some time looking at kinases, some work on ion channels,... When you go further down the road, I start worrying about a progression of increasingly serious nightmares! If we can find the structure of a couple of proteins, and figure out how they interact, that’s within range. If we don’t, we’d have no idea if it interacts with what. We’re going to have a very closed relationship with systems biology... Further down the line, we start dealing with what kinds of undesirable interactions might we have? It’s one thing that two proteins might bind, but toxicity results from unfavorable protein interactions [e.g. HERG and cardiotoxicity]... Cells are not bags of liquid -- not only are they viscous jelly, they have little compartments, things that move along... Just figuring out what interacts in the abstract without knowing what’s near what else is a small part of the puzzle.
What are the biggest challenges for your field in the post-genomic era?
Liebman: Those things we’re trying to do, being translatable from laboratory to clinic, unfortunately don’t always understand what the clinical problem is. They start with good scientific foundation, but lack the ability to address the true clinical problem. For those of us interested in translational research, we need to form a closer bond to the clinic, with clinicians and patients, to understand where there are gaps of knowledge on a day-to-day basis, and bring those back to the laboratory — not just some of the interesting scientific problems.
LaBaer: There’s a vast dynamic range of concentrations of proteins in various systems, e.g. serum, as well as the complexity of the proteome. There’s a sociological issue — most proteomics folks are gearheads, they want to solve technical problems, but they need to get together with biologists, physicians, and statisticians, to tie proteomics technology into something more translational. I think there’s a resource problem — if we want to study proteins at a large scale, we’re going to ultimately need antibodies for all proteins to look at protein function. And lastly, a technical challenge: it’s very easy to gather a huge amount of data very quickly, but harder to look at data and come up with a useful hypothesis. Figuring out how to do that is not trivial.
Galas: Just as in the last decade, the value of the interface between chemistry, physics, computing, and biology became evident, that convergence is continuing. We need to continue the convergence with clinical science. The challenges are enormous, but it’s also going to be a lot of fun.
Lee: We have two real challenges [in alternative splicing]. On the technical side, the basic question of how we’re going to get the complete set of variants. EST data kind of petered out, and the question is will microarrays completely take their place and take us down to the level of every cell and its differentiation and activation states?... A cultural problem is that bioinformatics and comparative genomics together can provide a huge base of candidates for biologists to look at that are clearly functional alternative splicing events, with the obvious problem being that to really understand their full significance to disease and medicine, I think, will require a discipline-wide attack on an almost 3X or 4X multiplication of the human genome in terms of number of isoforms relative to number of genes. There’s some intriguing data from Spain that suggests that splicing mutations may be the main cause of human disease. This is something that’s been lurking as dark matter.
Shaw: It would be useful to have more colleagues doing [computational biophysics]... One problem is...to deal with multiscale models, from detailed atomic simulation, up through a level of large molecular machines, inaccessible even with the machines we’re building, up to whole organelles, and ultimately a whole cell, tissues, and organisms. My hope would be that we can find some way or technique to abstract those models, build up a hierarchy of models from the bottom up.
*Beyond Genome, CHI, San Francisco, June 20-22, 2007.
Subscribe to Bio-IT World magazine.