A systematic approach to biology
At the turn of the millennium, and as the Human Genome Project approached its last phase, the pharmaceutical industry was buzzing with optimism about the innovative new medicines that would derive from the coming avalanche of human genomic and proteomic data. The evidence since then, however, has shown that this hype has not been justified; in fact, fewer novel pharmaceuticals are entering the clinic each year now than during the 1990s. The industry is increasingly coming to realise that, just as biochemical processes do not work in isolation, a drug aimed at a single target is likely to affect others in the same pathway, with unexpected and potentially damaging consequences.
Many companies, therefore, are turning to the new discipline of systems or network biology to solve this dilemma. Adriano Henney, director of AstraZeneca’s Pathways Capability at Alderley Park, UK, defines systems biology in the context of the pharmaceutical industry as ‘understanding how a target will respond in the context of cellular networks’. Yet the same industry, strapped for cash, is hardly in a position to invest in complete new facilities. Sensibly, many companies are choosing collaboration: working closely with both academia and software companies to produce tailor-made solutions. One company that has found this approach productive is Merrimack Pharmaceuticals, based in Cambridge, Massachusetts, USA. The scientists at Merrimack collaborate closely with Massachusetts-based software company The MathWorks, and have been impressed with the performance of their novel systems biology platform, SimBiology.
Merrimack is a biopharmaceutical company, founded in 2000, that is developing a portfolio of biotherapeutics for the treatment of autoimmune disease and cancer. Its first product, MM-093, a recombinant version of human alpha-fetoprotein, has recently concluded a Phase 2 clinical trial for rheumatoid arthritis and a pilot study in psoriasis. It is now concentrating more on the oncology pipeline, developing monoclonal and bi-specific antibodies that will bind to, and so ‘turn off’ the growth factors that drive the growth of tumours. The Network Biology team at Merrimack uses the principles of chemical engineering, computational modelling and high-throughput biology to identify which of the many signalling proteins involved in a pathway is likely to be the best target for drug development, and what types of intervention will be the most effective. ‘The models and our Network Biology approach are used to better understand complex pathways in disease, to design targeted therapeutics, to predict synergistic drug combinations, and to identify patients most likely to respond to targeted therapies,’ says Birgit Schoeberl, Merrimack’s director of Network Biology.
Merrimack’s collaboration with The Mathworks’ goes back to the company’s beginnings. Initially Merrimack just used the latter’s complex Matlab environment to design and develop models of biochemical pathways. It has now been using SimBiology for about two years, initially as beta-testers. This product is essentially a graphical user interface (GUI) that sits on top of Matlab. Schoeberl explains the rationale behind Merrimack’s choice and use of this product: ‘SimBiology is a GUI that allows us to easily build and share mathematical models based on biochemical reaction schemes between modellers and experimentalists without losing any of the flexibility of writing our own code where we need to. Using SimBiology, it is very easy to change one or more input conditions and see how it affects the simulation results,’ adds Schoeberl. ‘The ease of use and the flexibility of basic Matlab in combination with all the toolboxes is why we implemented it. It makes it much easier for theoretical and experimental biologists to collaborate.’
Schoeberl and her colleagues examined a number of other modelling tools and GUIs, including open-source solutions, before settling on SimBiology. She is sure that one of the main factors influencing their choice was the fact that they were already familiar with Matlab for other applications. ‘In choosing SimBiology, we found a solution that enabled us to bolt the GUI seamlessly on to a comprehensive platform that we already knew well, and that includes bioinformatics and statistics tools as well as simulation software.’ They are also finding the facilities for searching for reactions and chemical species within models particularly useful, as some of the networks they work with can contain as many as 500 initial conditions and 200 kinetic parameters. They have found few disadvantages with the software. ‘Simbiology is still a little slow solving large systems of nonlinear ODE, and we would prefer it to be faster especially for sensitivity analysis or parameter estimation,’ admits Schoeberl.
The close working relationship between Merrimack and The MathWorks has also allowed Schoeberl and her colleagues to suggest novel functionality for SimBiology, and to see changes implemented promptly. One recent example has been the incorporation of parameter estimation and sensitivity analysis, techniques that Merrimack’s modellers have developed in-house, into the SimBiology platform. Schoeberl explains: ‘If you are modelling a complex protein interaction network you will want to know which parameters and conditions the output will be most sensitive to, as these are likely to be the most rational ones to adjust. Sensitivity analysis is a method of identifying these.’ As this issue went to press, Merrimack and the Mathworks were due to present a poster at the seventh annual international conference on systems biology (ICSB), held in Yokohama, Japan. The work presented there uses the example of the ErbB receptor network, which is important in cancer development, to explore differences between global and local sensitivity analysis.
Sensitivity analysis uses all the available experimental data about the pathway concerned and the parameters predicted to be most sensitive are then modified in order to fit the model better to that data. At present, the most important limitation of this method is that much of the data, particularly the kinetic parameters that describe the speed of reactions, is still not known.
A consortium of German academic groups is taking a similar approach to Merrimack in teaming up with a software company, GeneData from Basel, Switzerland, to simulate a cellular system of clinical importance. Jens Timmer from the Physics Institute in Freiburg, Germany, is the coordinator of the HepatoSys project, which was set up four or five years ago to model hepatocytes (liver cells). The consortium includes experimental and theoretical biologists and clinicians organised into local networks, some of which investigate specific biochemical processes while others develop generic methods. The German government recently agreed that HepatoSys’ first funding period had delivered ‘proof of principle’ and doubled its funding to 24 million Euros over three years.
HepatoSys scientists’ collaboration with GeneData came about because of the need to store quantities of data and models in standard formats and share them between all the groups in a disparate consortium. ‘When we set up the consortium we found that although all the groups were happy to share their data, it was stored in different formats and often poorly documented, so collaboration was very difficult in practice. Now we all use a single data format via an Oracle database built by GeneData. We hope that it will become a “gold standard” throughout Germany and beyond for the storage of data for systems biology, in parallel with the widely used Systems Biology Markup Language (SBML) standard for models,’ says Timmer. The choice of GeneData as a partner was largely based on that company’s years of experience in database design for the pharmaceutical industry. The database was written to a 100-page requirement specification produced by Timmer and his collaborators, who are also using GeneData’s Phylosopher software to integrate different types of molecular data and use it to reconstruct metabolic networks.
This collaboration has advantages for the industrial partner as well as the academic consortium. ‘The science funded through the HepatoSys project is cutting-edge, and it is very rewarding to work with them,’ says GeneData’s Hans-Peter Fischer. ‘There are also many technological challenges involved in setting up a large and complex database to store both experimental and simulated data.’ Fischer and colleagues are hoping that there will be further opportunities to market the database they have developed to the pharmaceutical and biotechnology industries when these industries are ready to put more investment into systems biology.
Nat Goodman from the Institute of Systems Biology in Seattle, Washington, USA agrees that The MathWorks and GeneData produce software that is useful, reliable and mathematically sound. However, he prefers to stress the limitations currently imposed on the modelling community by the dearth of accurate kinetic data. ‘So far, the most impressive network and pathway models have come from simple, isolated and well-studied systems where very detailed data is available, such as the effect of a single-gene “switch”: not so much “low-hanging fruit” as fruit that is already falling off the trees,’ he says. He believes that accurate simulations of more complex, interesting and clinically relevant networks will need data gathering ‘on a scale that most biologists cannot even imagine’. However, the first of Merrimack’s oncology products developed using network modelling are now in research and moving towards the clinic. A successful launch of one of these would prove the principle that even the imperfect models in current use are of significant value in modelling the complex process that is cancer development.