YouTube Facebook LinkedIn Google+ Twitter Xingrss  

Breaking Down Modeling Technology


Feb 15, 2006

Colin:  There’s a key question here that we haven’t really touched upon, which is that the technology is still evolving. I see the systems biology really being divided into two major approaches, and it affects the answers to all these questions. One is the mechanistic simulation of known pathways. That represents a certain kind of approach, which is ODE modeling of the well-studied pathways and representing a few percent of the total circuitry of cells. Then you have sort of the inference modeling of the rest of the universe, which is driven a lot by the low-throughput and the high-throughput data. That’s a different approach and goes more to discovering new biology, discovering MOA and biomarkers of compounds.

John:  Is it possible to map approaches to their best use?

Stan:  I would put one question for clarification to Colin because the distinction he just laid out between ODE-type modeling and more descriptive referential model omits a class of models with a long history in pharma -- pharmacodynamic models (PK/PD). I don’t know how to classify them. It’s curve fitting. We do a lot of curve fitting; whether it’s with an ODE model, it’s still curve fitting. We’re trying to estimate the parameters. Yet if I choose an equation that has a particular mechanistic interpretation, am I now doing mechanistic modeling, or am I doing inference because I’m letting the parameters float? There’s a huge history in literature and set of techniques that kind of straddles the distinction that you just made.

John:  Does Pfizer or AstraZeneca or Millennium distinguish between these model types?

David:  Well, to some degree. If you talk to people, the initial adopters who are [advocates] in the company, they may not be there in terms of understanding the fine distinctions between the types of models that they can use. I think understanding what tool is appropriate for what question is exactly what we need to bring. So what the business is trying to understand, what Pfizer is trying to understand, is what we get for our money. So the way to do that is by using the appropriate tools for the appropriate questions.

The spectrum is quite wide. I would make it wider than what Colin has made it; go all the way to clustering as an approach that is essentially, in essence, generating a model. We’ve become very good in industry at these very statistical approaches. If I look at where most of our strength is internally, a lot of the knowledge and expertise in the company is really in the area of statistical modeling. There’s a huge gap in between where we have little bits and pieces that are relatively spotty. I think those are the places where we need help, at least for the next couple of years, and where there are opportunities for providers and for collaborations with academia.

Keith:  The crux of systems biology is coming from looking at a lamppost, reductionist view of the world, to a much more integrated view. If you look at the approach that we’ve [Genstruct] taken to modeling this in space, it’s not trying to model an individual subsystem, but in fact trying to build a first-level detailed map of the park. How do you do that? Well, the genome project gave us the vocabulary for this, the genes, the proteins, and we have the metabolites from a huge history of biochemistry. So can we build a single integrated map that integrates all these lamppost views into one single view? That was one of the challenges we took on early. I think we’ve accomplished that.

Herbert:  That’s a static model.

Keith:  Well, in our case it’s a static qualitative map. We think this map is a representation of all the potential states of all the potential players in that system. In fact, there are many, many, many possible solutions. So what you have to do is evaluate all of those solutions to find the most likely. So there’s a lot of math in our modeling, but it’s not the kind of quantitative math of looking at the dynamics of the path. It’s evaluating all of the different paths in the context of all the data. I think this is a unique kind of modeling approach, in that it can take very large amounts of data, thousands of biological state changes, and put them onto a map that is very large, 250,000 or more individual cause and effect relations, and define the 1,500, the 2,000, the 3,000 that are representative of the biology that explains the data. Then by evaluating that model you can find the mechanism, and the biomarkers.

Colin:  So Keith, is that limited to the literature, to let’s say 5 percent of human biology that we do know as far as the maps?

Keith:  Well, I’ll argue we know far more than 5 percent of biology. I think the reductionist viewpoint has given us a sense that we have a small and limited knowledge of biology. But when you look at what we know from model organisms, when we look at the conservative nature of biology, it turns out that we know quite a bit. For example, if I learn the cause-and-effect relationship between two enzymes in a pathway, and I’ve done that in neural stem cells, and I see those same players in liver cells, do I assume that I have to re-determine their relationship? Or is biology conservative? Well, the fact is that it is conservative. Those functions are conserved across different cell types. If I look at the same kind of reaction that I determine in a mouse cell and I try to ask will those enzymes have the same activity in human cells? Do I have to re-determine their relationship? The fact is no. Biology is conservative across species.

Alex:  Some of the technologies are very good at letting you see sort of the whole map and the different networks that you should explore if you’re trying to understand the mechanism or whatnot. There are other technologies that are better at narrowing down what is an acceptable dose range or what is going to take us over the effective efficacy threshold. I think both have roles at different places in the pipeline. That’s what is great about the groups that, for example, Jack and David have is they have that kind of mission of we’re going to do modeling, but coupled with wet biology so we can do that kind of loop and narrow those things down.

Return to main article.

Click here to login and leave a comment.  

0 Comments

Add Comment

Text Only 2000 character limit

Page 1 of 1

For reprints and/or copyright permission, please contact  Jay Mulhern, (781) 972-1359, jmulhern@healthtech.com.