My recent column -- “Harvard Tackles Systems Biology” (see July 2005 Bio-IT World, page 50) -- provoked a couple of intense reader responses to efforts by Harvard Medical School researchers Jeremy Gunawardena and Aneil Mallavarapu to create a new modeling language for systems biology and to the manner in which systems biology is reported in the press. One reader labeled the current practice of systems biology as “a foiled enterprise” and press coverage as too superficial. On the latter point, he may be right.
| ||Jeremy Gunawardena |
Presented here, with minimal editing, is essentially the entire interview with Gunawardena, director of The Virtual Cell Program at Harvard Medical School’s Department of Systems Biology, and Mallavarapu, a research scientist in the program. They graciously gave me an hour and a half during which we discussed: systems biology as a whole; why Harvard thinks it’s important enough to start a new department; their efforts to create “little b” (b
; modeling language); the biopharma industry’s tentative reception of systems biology; and the pair’s enthusiasm for synthetic biology’s promise.
Presenting such a long piece is something of an experiment for Bio-IT World. It would never fit in print, but is easily accommodated online. We are attempting to learn how to maximize our use of the online experience to provide readers with a richer experience, greater depth, and in cases like this, something akin to a conversation. Occasionally we will run longer articles if readers deem them useful. Please send your comments on length and content to firstname.lastname@example.org.
First, a little background: Gunawardena is a self-professed mathematician (Ph.D. in algebraic topology) who fell from grace. He’s had stints in academia and industry, including a long stay at Hewlett-Packard doing industrial research. At HP, he caught the biology bug and moved to Harvard’s Bauer Center for Genomic Research where, “Pretty much everything I thought before I came [to Bauer] turned out to be…modified substantially,” he says.
Mallavarapu is a cell biologist and biochemist by training. He took his Ph.D. with Tim Mitchison at the University of California, San Francisco, working on photomarking technologies to visualize cytoskeletal dynamics. During the past few years, Mallavarapu worked at Millennium Pharmaceuticals developing technology, writing software, and thinking about what is now b, before joining the Virtual Cell program.
JR: Let’s start with Harvard’s systems bio department. What’s that all about?
JG: So far as I know, this is the only department of systems biology that exists. There are many [systems biology] centers or institutes across other organizations. MIT has CSBI (Computational and Systems Biology Initiative), which is sort of an umbrella. The interesting thing here is that it’s a new department with all the implications that means, faculty, students, and endowments and money and all the rest of it. It is quite unusual in my view. For one thing it actually encourages the presence of nonbiologists like myself in the middle of the medical school. I think there’s a real commitment to creating, not just sort of talking about, a discipline. It’s something to achieve, [to] actually try to live that and create it and extend to the new graduate program, which is a very important part of the landscape here.
JR: What so different about systems biology that it requires a new department?
JG: I have this kind analogy. If you ever bring two disparate languages together, like communities together, which happened occasionally in various islands in the Pacific during the British Empire, and they were basically indentured laborers from China and India. The first generation of people constructs a broken kind of language beginning with the everyday necessities such as for bartering goods and stuff. That’s called a pidgin language -- it’s not a proper language. What happens is the next generation of children somehow spontaneously creates, in place of the pidgin, a fully-fledged language and it’s called a Creole. A Creole is based on pidgin but it’s completely as good as any other language. The analogy here is [that] bringing together these different scientists is a bit like that. We’re the first generation. We speak pidgin here, but I really hope the graduate students that come through the program will grow up and construct the real Creole, the real systems biology language. We’ve just admitted the first nine students this year, so quite a lot of focus is going toward building up this graduate program and trying to understand ourselves, how we teach students systems biology.
It’s a small department [that’s] only been running for about a year, or a little over. There were three founding faculty, Marc Kirschner, Lewis Cantley, and Timothy Mitchison. They were the gang of three who believed sufficiently that something new was happening that they really wanted to come together and make this happen. A number of other people joined initially. Pamela Silver was one of the very early people who came along. I happened to be around at the right time. Walter Fontana was another. He’s a theoretician by background and was at the Santa Fe Institute before coming here. We’ve had a wonderful group of junior faculty.
JR: Did you shape the direction of the department?
JG: It was in discussion when I came. I find myself here as the result of a series of accidents, really. In many ways I wasn’t anticipating coming back to academic life from my previous job [Hewlett-Packard R&D]. I came to Harvard on the other side of the river to the Bauer Center for Genomics Research, as a visitor for two years, and having gotten very interested in biology, it was an opportunity to actually live among biologists. Pretty much everything I thought before I came turned out to be modified substantially.
Discussions about what the medical school was going to do were taking place roughly at the same time. I got sort of entrained in a lot of those discussions, and when the new department opened, they made me an offer to come and start up a group, really centering around this idea that Aneil and I had been working on -- the idea of the virtual cell which emerged in the form of little b in its current iteration. In time, the work I’m doing has broadened. We have an experimental group as well as theoretical work. Little b is the thing I guess that has been [worked on] longest. I’m particularly interested in signaling in complex cells, but little b is the aspect of that that deals with creating a computational infrastructure. It has a much broader scope than the specific biological application that we’re [pursuing].
AM: Originally I began thinking about this exchange language in the context of relations between academia and industry. This was at the heart of how Jeremy and I came together. Jeremy was a great person to talk to. We shared, at least in some ways, at a superficial level, a similar trajectory of being in academia and going to industry and thinking about coming back. Jeremy actually was at CGR and we started talking about some really interesting stuff that came out of MIT.
In particular, one was a language which enabled one to describe complex structure that was based on small parts. This was inspired by biology, and yet we also need something like that for biology so we can describe it formally. [That] was a very immature idea at the time. Jeremy got me so excited, I said, “I’ve got to continue working with him somehow.” I managed to convince some of the people at Millennium [Pharmaceuticals] that this was a good idea.
JG: And Mark Levin [then Millennium CEO] was very supportive.
AM: [Many] technology managers were supportive. They were very broadminded in that way. The initial agreement was to think about what was the role of systems biology. We became convinced it was really important to think about it in the context of drug discovery, but in particular with respect to modeling, which I thought was one practical application in industry.
We began to think about how modeling could play a role in industry. At Millennium, companies like Entelos and others come by, and there wasn’t really a quick uptake of that technology. I know with Pfizer, they [Entelos] had some great success with asthma models and great successes since then. I was just wondering if there was another way to think about how to make modeling happen.
JR: So why did you leave Millennium?
AM: Millennium went through a shift in its focus from a technology to a drug discovery company with a couple of drugs. All of those biotech companies that are successful get to make that kind of transition, and depending on who you are, that’s either plus or minus. It meant a real diminishment of the technology focus of the company on later things in the pipeline. It was hard to imagine how something like this [systems biology and little b] would go forward, which I felt needed not just a single full-time employee but probably a big group.
JG: Well, a tool business is hard to do even outside the biotech business.
AM: It is when you have such a small number of companies [to sell to and] you have to charge so much and the skills required for producing the software are so high. At the same time it’s clearly an activity that needs to happen. So it’s hard figure out what is the right model for this. I felt for the language, it was really important for the models to be trusted, to be tested widely. One way to test models if you buy the notion of some type of predictive biology -- which I think Jeremy has antibodies to -- is you need for people to have looked and tested all of the parts of it.
[Another] way to do that as a single company is to spend a lot of money to do it all yourself and have somebody trust you, and another way is to take an open-source approach and have many eyes on the components. When you do that, I have felt, and I think others in the modeling communities who are interested in sharing models have felt, that the more we have it in the public [domain] and part of the academic and open-source modeling world, then we can start to trust our formalized knowledge in biology and we can reuse it in many ways and test it in many ways.
I guess that’s kind of one of the distinctions between platforms like Entelos and Ingenuity [Systems], which are coding lots of knowledge, using similar techniques, actually, but as proprietary tools. I felt the [open-source] way was how a drug discovery company could actually take advantage of modeling that it could trust. In the early days of seeing some modeling companies come try to sell you something, there was a point of mistrust of how do we know whether these models are actually trustworthy. We’re being sold something but is it [trustworthy]? A part of that trust comes in the same way that we have trust in the scientific literature; that many people have tested the results. It’s strikes me that modeling also requires that same thing.
JR: Let’s go into more of what little b is.
AM: Right now we’re focusing on biochemical modeling, and we’re using it to build fairly simple models called ODE (ordinary differential equations). You represent species as concentration or the amount of species as variables. We want to expand and use the language to describe other types of models like partial differential equations, and we’re working on what it will take to write stochastic models. Right now we’re just doing biochemistry, but there’s nothing in the language that limits it to that.
The thing that makes little b different from some of the other platforms -- I can’t say this completely authoritatively -- [is] we really focus on the issue of modularity. It’s meant for modular model building. When we think about modularity, we think about it on many levels but most importantly that users will appreciate is each model is composed from predescribed parts. Currently when people are writing models, such as in something like Matlab, you have many equations and each of those equations has contributions from many different parts…[and] also the focus is on coding many different mechanistic functions about how a reactions happens, how many steps it has, and so forth. So removing something or adding in something requires changes in many places; it’s complicated in deleting several terms or adding several terms or modifying terms or doing some other more complex operations.
JR: What would be a reusable concept?
AM: A species type is one type of reusable concept. A reaction type is another. These things aren’t part of the core language. This is another kind of modularity the language has, which is the ability for somebody we’ll call a theorist to then extend the language using the language and provide higher-level constructs. For instance, we can write a notion of something called an aggregate, which describes species types and reactions types, and that might describe a complex, or we might describe something else like rigid filaments. In that way you can also have theorists to extend the terminological set that a biologist can then use.
JG: Perhaps a different cut on this is to think who’s going to be using this. The situation at the moment is systems biology is something not everyone understands. Many people have different ideas about it, and the role that mathematical models play is also something [there are] many different opinions about. If we leave that aside, let’s assume that we use it for a biologist to look at the behavior of a system by representing that system in a mathematical description and interrogate that mathematical description to see what happens and elaborate that just a little bit.
One way to think of it is constructing virtual descriptions of biology on which we can perform virtual experiments in a way that would be very hard to do if we had to actually do the experiment in real life. Let’s just assume that you’re an average working biologist who’s studying, say, MAP kinase cascade and would like put together a model of this pathway, maybe in a particular cell type, and with all sorts of assumptions and context. At the moment, how would I do that? Basically I would hire somebody who is competent at the mathematics such as a Ph.D. student or a postdoc, and that person would go away and sit in some cubicle and come back later and show me his work.
You know a few people do that, but it is not something which the average biologists comes to sit at their computer [and does]. [It’s not like] summoning up blast or something to interrogate gene sequences, which has become part and parcel of everyday work life for the average biologist.
So if what we’re doing is going to become anything more than a kind of niche activity for a small group of computational biologists and become part of working practice then I think we have to solve a number of problems. One, of course, is we have to explain to people what they get out of this that’s different. That’s a very important thing, but let’s leave it for a second. One of the things we have to move away from is this idea that every time I want to study a particular system, I have to build from scratch a kind of monolithic model of it, and not take advantage of the fact that lots of other people will have actually bitten off various different parts of it.
JR: Can you give a more concrete example?
JG: So the kind of world that we think is in fact essential to progress. [It] is one where I can look at the EGF pathway and I can say: Doug Lauffenburger (MIT) and his people have been studying this receptor family for years. They have a huge amount of experience and knowledge of all the intricacies of working with it, and in fact the guys in their lab have written models for what the receptor does. And my MAP kinase cascade people have been working on this. And Deborah Morrison’s group (NCI) have been studying MAP kinase scaffold and things, and lo and behold they’ve written a little model that describes what they think happens. Then we’ve also got a whole set of transcription factors, and there are models of how those work.
I haven’t written those [models]; other people have. What I don’t want to do is to read the literature and put all this together myself. Why can’t I take at least a part of Doug’s model and a part of Deborah’s model and part of somebody else’s model and drop them into a kind of virtual space, and lo and behold the computational infrastructure that little b provides knits this all together with the minimum effort required to rewrite things. You can’t do that today. Those models might exist, but you would have to sit and rewrite them completely in order to put this together. The idea here is if we had a computational structure that would allow these things to be encapsulated in such a way that the system would wire them up for you.
JR: For that to be effective, doesn’t the language have to be widely adopted?
JG: Absolutely. So there are many issues. One is they would have to be writing in the language, so we have the chicken-egg situation. How do you get everyone to be speaking the language? But there are more subtle issues. For instance, let’s suppose we’re describing a particular protein pathway. Let’s suppose it’s any old protein. I might choose to describe this by saying that it’s activated or not activated. It may turn out that activation state is determined by phosphorylation and there are eight phosphorylation sites. Somebody else who studied this may actually want to know what exactly is going on in each of those sites, so they want to know which of the 2^8 possible states it’s in, not just the two. So now which of these is the right description? Well, they’re both right, and we could have much more complicated examples.
So there are many possibilities. There’s no unique way of describing any particular component, really. It depends on the question you want to ask. When we come to studying and building models of biology, we confront a situation which is really quite distinct from what we have when we describe, say, engineering structures and airplanes or physics, which typically have unique descriptions. In biology, your choice of how to describe things may depend on the question you ask of the systems, which introduces another level of complexity and it’s one that ultimately can’t be solved. If you choose to describe your protein in this way, and I choose to describe the protein this way, the system can’t work out what to do. The best it can do is to try to fit everything together when it can, and when it can’t and encounters some kind of inconsistency, it should fail gracefully.
Now because we’re working within a language we have the possibility that the user could explain how to fix this. Or it may be possible, if there are two sorts of levels of description, that people can, in the language, write a way to translate from one model to the other mathematically. All these things remain to be understood as the language becomes more widely used. But it’s certainly the case that unless there’s a certain degree of consistency about the way people are describing things, there’s nothing much the infrastructure can do about it. But I think that’s fine. That’s life.
JR: Why couldn’t I accomplish much the same thing using the existing systems biology markup language that’s in use now?
JG: So what SBML is, is an XML format. It takes the view that models are data, and it specifies the components, the various data structure. That’s a very reasonable thing to do if what you’re trying to do is to wrap the model up, and having constructed this model in one tool, I can ship and actually reexamine it in another tool. So it’s a sort of data exchange format, and it allows many different tools to access in the same model. And it was a very important development in bringing the systems biology community together because it allowed people from control analysis, from development, from control theory -- all of whom had different assumptions and different tools -- it allowed this community to come together and exchange models. And for that it’s fine. But what we’re trying to do is something quite different. Which is not merely to exchange models but to construct [them] in a way where I can build on all the work that other people have done.
AM: So I think the unit of exchange is different. It’s a finer grain, so where exchanging the objects and the mathematical assumptions and the language actually combines those things together to produce a model. So little b would produce something that looks like SBML or Matlab or mathematics.
JG: We’re coming back to the point Aneil made here about how we should formalize biological knowledge. Eric Raymond wrote a little book called The Cathedral and the Bazaar to highlight the distinction about the two ways of thinking about software. The cathedral was a reference to Microsoft basically building a monolithic form. The bazaar is the open source [where] anyone could contribute. One can add to it, and the way you have confidence in it is because everyone is using it. Therefore, this becomes tested in a way that’s not like Microsoft. I think [that] when we come to think about how we formalize biological knowledge, much the same analogy applies. We can take the point of view that some companies -- like Ingenuity and Entelos and the like -- that they’re going to sit there and collect all the information and knowledge that they can find and they’re going to be assembling this model entity. But there’s another way of doing it, which is to say let’s distribute the resources for building knowledge to everyone and allow everyone in the course of their normal working life to contribute. It’s very much the open-source view. That’s kind of where we’re going with little b because we feel what little be allows is if I write the model and I put it into little b and then put it back into the community then anyone else out there can take advantage of it.
JR: If somebody uses little b to make models, are they bound to make those models public?
JG: No, not the models themselves. We would like to release little b under some sort of GNU public license. As far as the language itself is concerned, if people make additions to the language, they have to contribute that [to the public], but they’re at liberty to use the language to write their own models and lock them up and never show anyone. I suspect that if this kind of vision of building formal biological knowledge in a distributed way works, then many companies will begin to take advantage in the way. It’s very similar to the way the Internet evolved, essentially being a very distributed activity. Now that it exists, many people can start making money on the back of it and are doing so very successfully.
I believe it’s the distributed model, which will ultimately work because it’s the only one that scales. The other thing I think [relates to] the nature of knowledge. Here again, this comes back to the issue of prediction in biology -- my own sense, as an outsider coming into biology from the physical sciences, is that biological knowledge is very context dependent. It’s often very ambiguous and prone to refinement, if not corrections. You can see the distinction between why something works in this cell type and something doesn’t work in that cell type, and it’s a very organic kind of knowledge. It’s not like physics where after levering away for 50 years, Maxwell wrote down Maxwell’s equations and from that point on if you want to know anything about electrical magnetic waves, you go to Maxwell’s equations and work it out. It ain’t like that in biology. The knowledge kind of grows, gets more and more nuanced distinctions between things, and it expands and doesn’t contract.
JR: That contradicts what most commercial modelers say. They cite the Boeing 777, which as I understand it was entirely modeled by computer. They suggest drugs and clinical trials will be modeled the same way eventually?
JG: I don’t take that view. I think that I can understand exactly why they come up with this analogy, but to my mind there’s a real difficulty trapped underneath. A simple way to say it is in the engineering world, we’re accustomed to the idea we can actually predict things. If I want to put up a building in the [Harvard Medical School] quad, I could -- well, I couldn’t, but people could -- work out the design exactly and what the stress will be on this particular girder. You’re very confident it’s going to be correct. Sometimes people build bridges that fall down, but most of the time, they don’t. The reason we have that confidence is at the end of the day we have it all boiled down to Newton’s laws, electromagnetic waves, quantum mechanics; these are natural laws [of how] basically things work.
Now biology runs on the basis of all of these [laws], but unfortunately, we cannot use those theories to tell us what happens in biology. If I want to work out how a cell is working, I can’t go back to quantum mechanics and work it out.
JR: Well, some companies are trying to do just that. BioNumerik Pharmaceuticals has modeled enzymes and therapeutics using first principles and Cray supercomputers.
JG: Yes, that’s a great point, and for the structure people who are really at the angstrom level, I can really see there are still legs in that game. For us who are still working at a molecular level, there’s a different problem, which is not computation. Let’s just suppose I was interested in EGF pathways and I could write down all the quantum mechanics and there was a computer out there [I could compute them on]. All I’m interested in is whether this transcription factor is going to get phosphorylated. I think even if we had the computational power, we’re still going to be in the process because we want to understand things in a way that we can understand ourselves. We’re going to have to deal with models, and because of that, because it’s phenomenological, it means that basically I make some pretty coarse assumptions.
AM: [But] every model, including quantum mechanics, is phenomenological.
JG: No. I think phenomenological means something different. The caution I have about this issue of models is simply that if you’re dealing with phenomenology, all your model is really doing is a complex piece of mathematics that is basically telling you whether your conclusions are rigorously justified based on the assumptions. But that’s all it’s doing. So really I feel much safer when I make a model of saying what I’m doing here is putting my assumptions to a really, really hard stress test.
JR: What’s the best way to validate these models?
JG: All of this validation has to come back through experiments. I think the difficulty -- and this is something we haven’t to faced up to in systems biology yet -- is the kind of experiments we might want to do to really understand phenotype, which is what we’re doing here, is actually a different sort of experiments from the ones molecular biology has been mostly living with for the last 50 years. Typically people study systems, they knock out a gene, they overexpress a gene, they do this on lots of genes. Well, that’s OK, but the kind of thing we would like to do is [different].
So James Ferrell studied the mathematics of pathways in MAPK and showed that it behaves like a switch, and he claimed very reasonably that this switch is necessary for it to cause differentiation. How would we really believe that? Not by knocking out genes here and there. What we would really like to do is to go back into that system and reengineer that switch so it was either not a switch or it was a much sharper switch. Then I might be able to understand whether it was a “switchness” of [the pathway that] was really contributing to what Ferrell thinks it was.
We have no way of doing that experiment because we don’t understand how many different ways we have to tinker with lots of different proteins in the system in order to change the shape of this response. We’re beginning to get a handle on some of those things. But I think what’s interesting is [that] for systems biology to become a mature discipline, and [for] people [to] really come to grips with it, we’re going to have to do a different kind of experiment to the one that we’ve previously done.
JR: On a commercial note, what biopharma companies want to use systems biology for is to prioritize leads, simulate clinical trials, etc. They want to do things that get compounds in the clinic.
AM: There’s also another possibility, and that’s what we’re going to learn from systems biology. Right now we’re focused on single targets and single molecules. There are certainly people who are starting to think about multiple targets and multiple drugs -- like CombinatoRx. Maybe the way we end up going isn’t the way we go now. It might not even involve modeling, but something empirical.
[Metabolic] control analysis -- and Jeremy, correct me if I’m saying some things wrong -- has shown us the notion of the rate-limiting step is actually not a real one in many circumstances. So even with a simple metabolic network, you think if you increase some rate-limiting step you’ll get higher throughput, and if you just do the mathematics and some very complicate mathematics, [you find] it’s a steady state system. You have to change multiple steps in order to get the change in the behavior of the system.
[I]t strikes me even though the way the industry is working right now…asking for [systems biology] to tell which of these leads I can cross off my list...my sense [is that] the real gain is going to come from thinking about the target discovery, or even what it means to develop a therapy or what it means to understand disease, and therefore maybe target a few points along the pathway, for instance, identify what the most rational strategies might be as we start to get more and more information around pathways.
I mean already Millennium has made a big commitment to thinking about pathways and for them what that has meant, last time I was involved in the analytical process there, is accumulating information about those pathways. Largely they’ve been using that [information] to take a look at their transcript profiling results and [to] become more certain of it. But I think there’s a sense there’s some way to apply that pathway information to pick out the right target, and that’s what has consistently been seen as very attractive use of that technology. But the business demands of where that company is at, and most of the industry at this point, just needing to get some sort of products out, doesn’t really validate that. Nobody wants to be in the target discovery space right now.
JG: My sense is the drug industry is still living in the magic bullet kind of view of drugs that Paul Ehrlich originally formulated. [But] one of the things you find when you have even systems of moderate complexity [is] often their dynamical behavior is really quite counterintuitive. You look at what appears to be an obvious thing to do, which is to knock X out, and that should have an effect of reducing the output of Y, but actually there is a feedback loop which compensates for it so that it up-regulates something else and therefore Y doesn’t change. Actually, even for a mathematician, it’s not easy just to look at something and say, well, it’s obvious that you should do X. So I think that we’re in the infancy of developing a kind of intuition for how…
AM: Maybe we won’t have intuitions about what happens when there are two or three feedback loops and then another couple of negative feedback loops.
JG: Because it can depend on what they are doing and what their parameters are. There are simple notorious examples in other places. There’s a famous example from road traffic network design, which is called the Braes’ Paradox. It’s a simple example where if you build a new road in a certain place, everybody’s journey time gets worse.
JR: I think that must have happened here in Boston.
JG: Oh, it’s definitely happening here. And when you tell people this, they say, no, this is a conundrum, how can that be? Everyone’s travel time got worse, but that’s the point; our intuition for what happens in complex systems is poor. So I think one thing in the long run [that] we might see [happen] in systems biology is a reevaluations of how we approach the discovery thing.
JR: If we look at little b, what contributions do you expect the larger community to make?
AM: I imagine the first thing that people are going to do -- I’d like them to do -- is just start adding data to the store of things that we can talk about because that will be the first and most useful thing. Second, a very close runner-up will be theoreticians actually taking up the language to describe their mechanisms of abstracting bits of biology. So I’ve done a little bit of that at a very trivial level. The other thing that happens is people don’t just publish monolithic models; they also publish very elegant methods for describing small pieces of biology. Yes, there’s mass action kinetics, many people know how to do that, but then there are other complicated issues of how do you write a model for scaffold or some other complicated assembly. How does it behave, how do you do experiments such that you can then model the things -- and people have written papers that already describe the mathematic algorithms -- that’s another type thing that I would really love.
JG: I think it’s the classical Catch-22 problem. Until there’s a community of people speaking it, why would you want to learn it? So how do you create a community in the first place? There are probably a couple of early adopter communities in critical places where languages become used. One we’ll see in the medical school here simply because I’m very keen on little b in the context of some of the courses that we’ll be teaching in the graduate program. I think that’s going to be a very influential early adopter community. I think it’s likely that some of our colleagues at the medical school and also at MIT are certainly very interested in it. We haven’t yet been able to put the language out there in the way that they can use it, but at the end of the day, I think that it’s really a kind of push and pull that ultimately makes a language successful.
JR: When will little b become available, and what remains to be done?
JG: The end of the summer is what we’re aiming for, August or September. There are a couple of awkward issues; one of which is because little b is based in Lisp, we have found it difficult to find a free Lisp version that will allow us to get it to be out there.
AM: We have to fix a few bugs in some of the compilers. I mean the language itself runs on LispWorks, which is a commercial proprietary Lisp system.
JR: There is an ANSI Lisp, isn’t there?
JG: There’s a standard Lisp and several free software implementations of that. They’re all slightly different, and slightly different in a way that we haven’t yet been able to get little b to compile.
AM: Though we haven’t really tried very hard.
JG: We’re working through those issues. We would like to get it into a form that is accessible to anyone in a platform-independent way and also easy to use; that’s very important. So that’s why we’ve been holding back. We don’t want to put out something which requires people to use a really gross tool to get something out of it.
JR: How are you planning to solve the ease-of-use problem?
JG: Well, I think [having] some really good tutorials and some nice libraries there out of the box so that people can take bits of code. Initially, yes, it is going to be code that people are going to have to modify. So these are going to be people that have some kind of bioinformatics background or modelers who want to try taking this for a spin, or students who are forced to. But later we’d like to write a GUI for it this summer, and that is going to be really important. I think for the working biologist that’s an essential feature. Without a user interface that allows access to it, they’re not going to write language.
AM: The success will be when people don’t even know about the language at all.
JG: HTML is a classic example. It runs every time you run a Web page, and who knows HTML? It is a sign of its success that it kind of disappears into the background and nobody knows that it’s there. I don’t think it’s going to be quite that straight forward with little b, but I think the principle is. Still, for the working biologist to use it, it needs to be something that they see as [easy to use].
AM: You asked one other question about when industry will start using little b. I’m actually really interested in that question. The question is how will industry start using it, for what purpose. Clearly they are not going to take little b and throw it into their lead optimization program tomorrow or even within a year. The question is whether anybody in industry is going to get interested in playing with the behavior of complex pathways and using that at an early stage to formulate their product development strategy. I’d be delighted to see that happen and make that happen. I mean when I first conceived of little b, it was really to write simulations, but it’s general enough that we can write any type of mathematical -- well, I shouldn’t say any -- but we can write a broad set of mathematical models. It is up to us to describe what kind of models we want to write and then analyze them some way.
The main thing to me, the heart of it, was to be able to throw a bunch of assumptions together and then very quickly formulate some simulations. Just playing with assumptions, and playing with many assumptions, and seeing what the [results] look like is going to give people a feel for how these things are behaving. [It will] actually give them intuition about what to expect.
JG: Right now when we plan our strategy, we look at a pathway diagram, which is usually linear, and has a number of elements in a row, a number of arrows, and we decide which one we’re going to target. I think that needs to change. I would really like to see it.
JR: Who do you think are likely collaborators -- industry, academia, or postdocs?
JG: I think it’s interesting that both Entelos and Gene Network Sciences, the two companies that are closest to us in systems biology in a modeling sense, both have internal languages, which in some sense do parts of what little b aspires to do. But they are not very interested in the issue of modularity per se. I think they will be interested if they could --instead of having to do a lot of that stuff in-house -- they could build on stuff that’s available from the community. Alex [Bangs, CTO of Entelos] and I talked about this.
JR: He doesn’t feel threatened?
JG: No, not at all. I think it would allow them to reduce their burden. So I wouldn’t be at all surprised if the language was to mature and maybe it would be something that was interesting to those lead companies because they are another possible early adopter community.
JR: After all is said and done, in your view, what is systems biology?
JG: I guess you could get ten different answers. My take on it, as somewhat of an outsider, is that it’s something that’s actually happening in biology for very natural reasons. That basically we’re moving out of the phase where the interesting thing for a biologist to understand was to identify what players are in whatever systems they are studying, and there are lots of players, proteins mostly. Ah, but there are a finite number of them in any system. So at some point it’s going to be an issue of diminishing returns to keep looking for new components. I think some biologists have actually begun to come to the conclusion that tracing out the components is no longer the interesting thing to do.
So if it’s not a question of finding components, then what’s the interesting issue? I think when you start to think about it, the interesting question has to be if we know the components well, how is it that they collectively produce the behaviors that we see. That’s the obvious thing to go do. I mean, if you know the actors, now what’s the play? Can we explain this? I think that is where system biology starts. It’s a natural progression from what molecular biology has been doing the last 40 years in providing all these wonderful tools. Now a lot of the time we know -- at least we think we know; sometimes there’s a surprise -- but nevertheless we think we know.
We’re past the point of diminishing returns, and now we have to start thinking about phenotype and looking at collective behavior. When you start doing that you begin to realize that it’s hard to even formulate the notion of phenotype linked to behavior without starting to think in mathematical terms to say, well, I’m interested in these concentrations and of these things in compartments over time. So I start writing down experiments to see what these concentrations are doing, and they have these crazy curves, and now I have to understand where these crazy curves are coming from. Now I need mathematicians who can tell me why these particular sets of molecular interactions are necessary for the results of the experimental curves I see. And that’s systems biology to me.
From a different standpoint, you know, you’re interested in looking at this transcription regulatory network that arises in development of this signal transcription and pathway that arises in a particular cell type, and it’s not high-throughput data you really want. You really want to know what are the concentrations of these things, which molecules are involved. It’s actually a situation, which is pretty data poor. A lot of those numbers are not there. It’s a different kind of systems biology from ’omics systems biology, which is about looking at a lot of smaller subsets the cell pathways, networks, and really understanding how the behavior emerges from a molecular integration/understanding.
AM: I almost wonder if we’re going to get to a point where we get to know certain types of familiar friends and [also] new types of objects that are systems objects, topologies of networks, or issues dealing with low copy number, or certain types of kinetics which will take us back to a certain extent to early biochemistry. But also we’re going to think of that in terms of larger-scale assembly. I feel like some of that is already happening. People in our theory lunch group, which Jeremy started, have a feeling about what a feedback loop does or what a couple of feedback loops do, or what the salient issues are there. There’s a language and an implicit understanding of what some of the issues with those assemblies are, and I feel like for me that is what systems biology is coming to mean.
JR: What are your thoughts about the emergence of synthetic biology?
JG: We’re very familiar with the synthetic biology crowd.
AM: I think it’s absolutely fantastic. I’m so excited about it. What’s so exciting is biology is breaking out to a whole different population of scientists. Right now, it [biology] is in its days of discovery science, and what’s so exciting is that the engineering mindset all of a sudden is part of what biology is all about. That really excites me because that’s what I love. The other thing is writing models for something you don’t know about is extremely challenging because you don’t know whether you’ve got it right. Whereas what they’re doing with synthetic biology is actually making things they think they actually know, and therefore they should be able to specify and formalize those and have a little fragment that represents that object.
JG: [It’s] like a model of a transistor then… We need to do a different kind of experiment. Part of doing those experiments [is] going to be about making much more complicated kinds of engineering devices that we put into cells in order to monitor these kinds of physiological behaviors. You know people are used doing the usual kind of plasmid engineering and introducing things we can induce. What we would want is to do much more complicate things; so building things which work and work quantitatively in a reproducible way.
I’m excited by the idea that the synthetic people are going to start putting bits and pieces together, and we can start really doing the kind of a study we are engineering.
AM: I feel like this is such a fundamentally exciting field. It’s like electronics. It’s a new materials platform. What I keep wondering about synthetic biology is what will be that killer application? It seems like it’s going to be so important, but I don’t think it’s going [to be] another computing [application]. They are building such things, and such things are going to be important, but I don’t think that’s the core of what synthetic biology is about. It’s going to be doing something new.
JG: Some of the people are thinking about gene therapy and things like that, and other people are thinking about actually engineering computers as an attempt to control disease. They’re far out at the moment, but I can see that beginning to really get some traction.
AM: Well, possibilities that are outside the FDA control and regulation are even more exciting.
JR: Ten years from now, what will we see and say about systems biology?
JG: Everyone will be doing systems biology, and it will have faded into the background, and we’ll say, well, what’s the big deal? To me that’s that mark of success. I think it will be like molecular biology. I think there is an issue, though. Perhaps it’s a problem we’ll run into if we get to that point. I think if you get back to the heyday of molecular biology, there was a kind of killer app. It was Crick and Watson’s discovery of DNA structure and its role [that] really precipitated the whole thing. I don’t think that’s how it’s going to happen [in systems biology]. What we’re talking about is really a change in the way people think, not a kind of scintillating new experiment that opens a whole set of things to do.