YouTube Facebook LinkedIn Google+ Twitter Xinginstagram rss  

Feb 15, 2006 | EXCLUSIVE ROUNDTABLE | Wouldn’t it be fantastic to interrogate virtual living systems, informed by the wet lab and translated in silico, to accurately identify targets, promising compounds, and decisive biomarkers? Computational modeling of biological systems holds this tantalizing promise.

We’re hardly there yet. We may never be “there” in the way semiconductor engineers stitch together microprocessor circuits using models because modern chips are too massive and complex to be created any other way. But steady progress has been made by a small community of model providers working with biopharmaceutical companies eager for predictive tools that will cut development costs. The technology works, though challenges remain.

Last October Bio•IT World and Health Industry Insights (an IDG-owned sister company) held a two-hour roundtable on the state of models. Researchers from Pfizer, AstraZeneca, and Millennium met with executives from modeling technology providers Entelos, Genstruct, and Gene Network Sciences, and an academic researcher from Keck Graduate Institute. It was a rich, wide-ranging discussion.

Presented here are highlights from this fascinating roundtable, moderated by executive editor John Russell and Health Industry Insights research director Alan Louie. A longer version of this article covering more topics is available online at Bio•IT World and Health Industry Insights thank the participants for their time, candor, and good humor.

Roundtable Participants

David de Graaf Jack Beusmans Stan Letovsky Alex Bangs

David de Graaf
Director of Systems

Jack Beusmans
Principal Scientist, Pathways

Stan Letovsky
Sr. Director of
Computational Biology,

Alex Bangs
Co-Founder and CTO, Entelos

Keith Elliston Colin Hill Herbert Sauro John Russel

Keith Elliston
Co-Founder and CEO, Genstruct

Colin Hill
Co-Founder and CEO, Gene Network Sciences

Herbert Sauro
Asst. Professor,
Keck Graduate Institute

John Russell
Executive Editor, Bio-IT World

Brock Reeve
COO, Life Science Insights

Alan Louie
Research Director,
Health Industry Insights

Kevin Davies
Editor-in-Chief, Bio-IT World

Is Pharma Ready to Bet On Modeling?

David: I think we’re still in the tire kicking stage. Modeling approaches have made an inroad in the sense that the concept is not frowned upon anymore. You don’t see large groups of scientists freak out when people talk about a mathematical representation of a biological process. I think of that as being the big gain. But for every company that I’m aware of, people are still working on understanding [modeling’s] impact.

If you think about the different aspects of systems biology in the sense that you have people who have a computational view of biological problems, as well as people who need to inform them through doing benchwork. Those bits and pieces are represented throughout a company like Pfizer. One of the challenges for Pfizer is to really get the view of how all these people can work together and conquer some of the organizational barriers.

FDA Plants a Flag
What is interesting is the white paper that [the FDA] put out on Innovation or Stagnation -- The Critical Path Initiative. They talk about trying [to] encourage use of technology and specifically mentioned modeling and simulation. I think the fact they even stuck a flag in the ground is encouraging. The industry reaction has been maybe we need to take a little closer look at this, which is exactly what the FDA wanted.
  — Alex Bangs, Entelos

Colin: It’s important to keep in mind that the pharma companies, even now, do have systems biology units. Certainly David’s group and Jack’s group, and it’s beginning to happen throughout most of the top half of Big Pharma. I think the questions are getting a lot more specific and as far as the particular technologies that can be aligned to particular problems. It’s not just systems biology as a big catchall phrase. Now people are focusing on Bayesian inference or ODE modeling and so things are becoming differentiated and the attitudes of people have evolved.

Alex: What we found is the people who work with us come initially with the idea that it’s a general capability that would be useful to them. I think what keeps them working with us is that they see that we have a unique scientific expertise in addition to the technology that really makes it valuable. Back to your question about where we are in terms of adoption, I think for the majority of the industry, we’re in that, “OK and we’re ready to do something with this” kind of phase. For some people though, we’re in multiyear agreements where they’re directing efforts in their pipeline based on our work.

Find a Champion
John: I’m curious how [modeling] engagements start?

Jack: The experience at AstraZeneca has been that it’s usually a — [and we have] a number of engagements with Entelos, Gene Network Sciences — a champion within the company who believes that this is a valuable approach that should be tried. [W]e have been in the tire-kicking stage actually a lot longer ago than just a few years ago. AstraZeneca had a trial of the obesity PhysioLab with Entelos almost five years ago. So, at that time, we had a number of people within the obesity research group at AstraZeneca that were very interested in the approach, but realized it was a novel approach, was certainly at that time somewhat risky approach, but were willing and able to sign off on taking that risk.

John: Stan, how about at a smaller company [Millennium]?

Combo Confusion
I think the area of combinations is an interesting one where system modeling might actually start to pay some dividends because the FDA has been saying we’re going to have to use combinations more aggressively. The industry is at a loss what to do because all of our notions on how to conduct clinical trials, how to do pre-clinical development are all based on a single compound. There’s all sorts of obstacles when we start thinking about what would happen if we do a combination trial with a competitor’s drug or with a second, unapproved drug. It’s a nightmare and yet it seems like there’s a lot of potential.
— Stan Letovsky, Millennium

Stan: Oh, so we pinch pennies to a scale that these guys can’t conceive. We would have to be very clear about what the benefit is going to be up front, and this goes for internal modeling efforts, as well. So, at this point, we’re not shopping for deals. We’ve had some discussions in the past with several people around the table. I think at those points there wasn’t a tangible enough benefit to a particular program to warrant an investment on our side.

One interesting question to ask is, Where is the impact of these technologies on the pipeline? Is it target discovery that systems biology is helping? Well, probably not, because you’re usually, if you’re looking for new targets, you’re looking in the bushes where you don’t understand what’s going on well enough to build a model. Biomarkers, the other end, clinical biomarkers, [are] very often based on correlative evidence, machine learning techniques, and statistical models.

So, in between, mechanism of action studies, translation, I think that’s where the biggest hurdles are right now, figuring out oncology, which cancers your drug is going to be effective against and why. If you’ve got a model that can shine a light and say we believe that 90 percent of the efficacy of this drug is coming from this little sub-network here and if you track these elements through the various stages of biological model development, you’ll be able to guarantee that you’re getting efficacy. That’s where I think the sweet spot might be for this technology.

John: Were there a lot of mistakes made when systems biology was getting started?

Jack: Well, I can give one perspective. They (pharma researchers) have heard about how this beautiful systems biology is going to solve their problems and they’re very naïve. They had their math phobia way back in school and oh my God, I will never understand it, but I think that there’s something there. So, they’re very naïve and very optimistic, and then they get burned a little bit. I think that’s happened a little bit in our experience. In the beginning people were expecting too much of an easy answer by just cranking some machine.

Climbing the Learning Curve
John: What went wrong with early models?

Jack: In some areas we just don’t know all of the details to sensibly model, and it wasn’t appreciated.

Desperately Seeking Wow
We’re all basically hoping for a really big hit, a wow. We’re looking for a result in which there’s no question that systems biology contributed to our understanding of why some patients respond and others don’t, questions of that magnitude. If we get those results, then the rest will follow. We’re not there yet.
— Jack Beusmans, AstraZeneca

Alex: I think early on we viewed models as a product that we could offer to people, and what we came to realize pretty rapidly with [the] people that we were working with was that while [they] had the enthusiasm for modeling, [they] didn’t necessarily have the expertise — and we didn’t necessarily have the right approach to be able to deliver it along with everything that they really needed to know.

So we shifted into a collaboration mode where they’re welcome to use the technology, we’re going to use it with them, and I eventually think we’ll come back around to people being able to have the expertise to take on a model and use it. But in the early days, people were like, oh, this must go in the bioinformatics group or something like that, but it’s a very different technology.

Keith: One of the things I think — and certainly in your statement Alex — is to look at the output of what you do technologically as being a scientific outcome. When you’re working with a therapeutic area at a pharma company, you’re trying to have an impact on a discovery program or development program. What they’re most interested in is whether you can, in fact, provide insights into the biology they need to understand.

Stan, you [said] if you can shine a light on the mechanism of action and tell me that this is the mechanism you’re going to see clinically and define some biomarkers for it, that’s something you are quite interested in. I think that’s what modeling can do, but not if you give a naïve end user a very complex model. I think biological systems modeling is going to be the space where computation finally becomes a scientific discipline within the pharmaceutical industry.

Building the Case for Success

It’s the DATA...
I think there’s a danger lurking in some of what people want models to be able to do, which is something disconnected from the experimental data. What’s very important in modeling is that models be able to identify the parameters you need to measure in the clinic to be able to predict the biomarkers to follow the mechanism of action or the mechanism of resistance. We can’t disconnect the models from the data.
  — Keith Elliston, Genstruct

Keith: Pfizer’s been one of our customers since early 2003, and we’ve gone through three different relationships exploring utility across not just discovery but through development and through safety and tox. Through that we’ve had a number of key programs that overall if we looked at the success of our results, to date, none of our hypotheses have been refuted. I think there’ll be some details presented on some of our work at the Society of Toxicology meeting next year. Also with Berlex and Schering AG, we’ve done some very good work in breast cancer. We’ve been able to identify sets of targets that they’ve pursued and as well worked to predict expression-based biomarkers clinically.

Alex: In terms of business success, we’ve worked now with a number of large pharma. We have multiyear agreements that have grown out of earlier agreements and they keep continuing with companies like J&J and Organon. Pfizer invested in Entelos. So that’s part of the proof  they’ve been happy with the early results. In terms of the science we’ve been able to, as I said, [we’ve] predicted negative outcomes for people. They’ve chosen to stop that work and then had those confirmed by work that was later published. J&J, for example, has talked about how we’ve helped them stop work in certain programs, how we’ve helped them with in licensing, how we help them predict clinical trials that they then run and confirmed the predictions.

I know one of the questions you had was ROI and metrics. Some of our partners have done that internally. Trying to get them to share that externally has been more of a challenge, but we’re very at least happy that they’re able to talk about some of the scientific benefits that they’ve gotten.

Colin: Three GNS/pharma collaborations have been made public. There was an early project with Merck around target validation and colorectal cancer cells and the models proved to be much more accurate than any of the other methods they used internally. That was based on mechanistic simulations of known pathways, and we were definitely proud of the result. But we also felt like that’s a very tough business to scale because it’s very labor-intensive and the models are always somewhat subjective. [Another] example, which is published, is a collaboration with Novartis, and that was more of a retrospective study looking for mechanisms of toxicity for a compound in clinical trials. The process of doing the inference modeling generated some just truly new predictions of biology surrounding the compound’s mechanisms.

Where I’m most excited in terms of the results is a more recent oncology collaboration with Johnson & Johnson. There was some work presented three weeks ago at the Biomarkers Conference in Philadelphia by one of our partners at J&J which showed preliminary results. So we’ve been able to actually distinguish the mechanisms of efficacy of a couple of different J&J compounds and other compounds in that class. But that project is still evolving. We expect that by the end of the year we’ll be able to share some of those results with the rest of the community. 

Productive Machinations
I think inference modeling gives the chance to start having biological discovery at machine speed, and it’s a much more data-driven approach. So we still develop mechanistic technology and mechanistic models based on the literature, but where we see some of the big impacts that can be rapidly generated and then rapidly experimentally validated will come somewhat from the marriage of the two things.
— Colin Hill, Gene Network Sciences

John: Have Pfizer or AstraZeneca run systems biology projects that made you think, “This helped us do something we could not do prior”?

David: Yes.

Jack: We’ve had those experiences. I’ve always said systems biology is a very large area. You have very different scales of modeling that you can engage in, from the very small sort of very quick hit, on the order of a couple of weeks or months, where you can actually with a very focused question in the right area of biology, you can put pieces together and make some inferences that were not obvious to the biologist in question. We’ve had some success in that sort of focused effort. When you start talking about PhysioLabs, of course, the scale is orders of magnitude larger, because of the kind of problems you are trying to deal with, and also, they have much larger time scales.

Alex: The way we’ve been able to get short-term impact for people, for example, has been in cases where we’ve predicted that a target was not going to be effective, and the partners have — and this has happened with us multiple times — the partner has chosen not to pursue it. Then results have been published later by someone else that showed it to not be effective. Then everybody’s been doing a lot of high-fives about how much money they saved by not pursuing that. So that’s why I joke that I wish we could get anti-patents. We’ve had people say before, “Well gee, we keep giving people negative news about things.” And the answer is “Well, a lot of stuff fails. So it’s not surprising that there are a lot of negative answers.”

We want to try to enable better use of these models than what we’ve had up until now. Pfizer’s had contacts or some program with virtually every company around this table and a good couple more. What we’ve learned is that PowerPoint slides aren’t the best way to get the information out of the company and that understanding of what and how we need to transfer things is something we’re still developing.
  — David de Graaf, Pfizer

David: Hank McKinnell [Pfizer CEO] said, “We’re an industry with a 98 percent failure rate. The only thing we have to do to double our success rate is to drop our failure rate by two percent.” So these decisions not to progress are the most common decisions and enabling those as early on as possible saves us lots of money.

The fact that the three of us (Pharma) are sitting around this table is the first proof that there’s some acceptance, right? I mean, companies are starting to hire in this space; big pharma, medium-size pharma are starting to hire in this space. So, I think ask us again in two years and if we’re still sitting around this table, then we will have shown adoption.

John: That all sounds promising. So what’s the problem?

Colin: There’s always such a hard time to get to clear-cut results that the biologist could say, “Yep, that definitely came from new insight from the model.” I remember we had a meeting with Bob Weinberg at the Whitehead. This is a few years ago, and after we did some modeling of apoptosis, we ended up with a pretty interesting result of some synergistic results for these compounds. And he said, “Yeah, that’s a pretty novel thing. I probably wouldn’t have come up with that. But you never know.” It feels like there’s always this — it’s hard to get at “Boom! The model came up with that.”

GenBank of Models?
It takes much more to create a GenBank of computational models. At GenBank [submission], if you could just shove a sample through a gene sequencer these days, [you’ve got a submission]. But for a computational model, there’s an awful lot of intellectual value goes into it. They’re very difficult. To curate those models on the current takes a couple of weeks for each one.
— Herbert Sauro, Keck Graduate Institute

David: So one of the problems with novelty is what do you measure novelty against? And so to some degree I’d say the answer is actually a very satisfactory one. You actually didn’t know the area very well and you were [still] able to do this because, with all due respect,  you’re not Bob Weinberg, a Nobel Prize winner, right?

Where Should Systems Bio Reside?

There is a range of opinions about whether we are actually competing with companies in this space because they’re essentially doing our science for us, which should be our bread and butter, all the way to people are saying no, this is a provider area. We should not invest internally. We should go and work with outside partners and have them enable this type of work for us. Then, in a couple of years when we understand the value, we understand who comes out of the slugfest as a winner, that’s when we’re going to invest. The truth is probably somewhere in the middle.  – David de Graaf, director of systems biology, Pfizer

We absolutely needed and benefited from our deals with Novartis and J&J in helping to focus particular aspects of our technology in the sweet spot of their pharmaceutical drug development problems.     — Colin Hill, CEO, GNS             J.R.

Read More:
FDA Is Trailing Indicator

Breaking Down Modeling Technology

Are Systems Biology Standards Needed?

Where's the Money in Modeling?

For reprints and/or copyright permission, please contact Angela Parsons, 781.972.5467.