Plectix Blog

Why Not Use Rules?

By Ty Thomson

 

Although the field of rule-based biological modeling has been around for at least ten years, it has yet to really take off.  The reasons why I think that it has yet to catch on should perhaps be the topic of another blog post.  Today I want to discuss how rule-based modeling compares to more traditional modeling approaches.

 

An often-highlighted aspect of rule-based modelingis that it allows you to build models with more species and reactions than you can build using traditional reaction-based approaches.  Often, researchers will wonder whether all those additional reactions means that they will need even more rate constants – and all you modelers out there know that the dearth of rate constants is a huge challenge to modeling. 

 

Although the researchers are right, there are indeed more reactions – which could mean more rate constants – the point of rule-based modeling is to take similar reactions that we suspect to have the same (or similar enough) rate constants and describe them using a single rule.  So the point of rule-based modeling is not to enable the modeler to build more complex models with additional rate constants, but rather to concisely build models where many reactions have the same rate constant.  This is not as uncommon as one might initially think, as, for example, the binding of a ligand to a receptor is unlikely to depend on the phosphorylation state of the intracellular domain of the receptor.  So the binding of the ligand to the different phospho-forms of the receptor can be described with a single rule instead of many reactions. 

 

Even more importantly, if you have reactions with distinct rate constants that therefore cannot be combined with other reactions into a more general rule, it doesn’t mean that you can’t use rule-based modeling tools!  In this case, each rule essentially represents a single reaction, so there’s no real cost to using a rule-based approach.  But even with only a few reactions that can be represented by a single rule, rule-based modeling can help by keeping your knowledge and assumptions organized (by keeping these reactions grouped into a single rule), and preventing you from making mistakes (by ensuring that these reactions actually have the same rate constant as you intended).

 

So I argue, why not use a rule-based approach?  Rule-based modeling is a strict super-set of reaction-based modeling.  You can’t lose.  I believe that in another ten years, the most widely used modeling tools will enable rule-based models, and rule-based modeling will be common.  Now we just need to develop the right set of tools to make it happen.  I am officially announcing a challenge not only to us here at Plectix BioSystems, but to all modeling tool makers out there.  Let’s make it happen.

Lessons from the Cutting Edge of Technology

By Joshua Havumaki

My over two years at Plectix come to a close this week. Below are my experiences working in an interdisciplinary start-up company.

I first joined as an intern in summer 2007. Plectix was a rapidly developing company yet to be fully established. During the summer, I was the only biologist amongst mostly computer science academics. This presented an exciting challenge for me. Since I had little quantitative experience and absolutely no exposure to modeling or computer science, it was an unparalleled opportunity to learn.  When I returned to Plectix full-time in early 2008, the team expanded as we began designing our flagship product, Cellucidate.

Aside from designing Cellucidate, the bulk of my time here has involved collating and modeling the large and complex EGFR pathway. Due to the fact that this was a new space for me, a significant portion of my time was spent learning and understanding the methods and nuances of modeling. This has also been a learn-as-you-go type of experience filled with many things that could not have been anticipated. Here are a few key lessons I’ve learned from working on the project.

First, the limitations of experimental and modeling methods should always be thoroughly discussed to ensure that experimental data can effectively complement the modeling work. Quantitative data can be deceptive, contextual and not appropriate for modeling. Second, when experts in different fields communicate it is quite possible to interpret terminology differently. This can result in a completely incorrect understanding of the technology or content.  Finally, being fully aware of the feasibility of large projects is essential. For instance, manually tweaking rates on my large model at first seemed quite effective for fitting the model to data, but eventually turned out to be nearly impossible.

Overall, this invaluable experience has exposed me to systems biology from both the academic and commercial sectors. I won't soon forget many of the lessons I've learned here and I know that they will prime me well for a future career in systems biology.

Biophysics in Boston

By Isha Antani

 

Biophysics is an interdisciplinary field that looks at biology from the lens of physic fundamentals. Relying heavily on quantitative means and mathematical theorems, the field adds an an extra dimension to the study of biological principles. Due to its quantitative nature, biophysics primarily focuses on networks and communications found in both intra- and extra-cellular space.

 

Recently the Council for Systems Biology held an all-day session on Biophysics as part of its annual Cell, Circuits, and Computation meeting. Many of the talks held my interest, but one was a pure standout - that of Dr. Jennifer Ross from University of Massachusetts on the complexity in cytoskeletal networks. Dr. Ross gave a dynamic, vivid presentation on the "walking" mechanisms of the kinesin and dynein motor proteins and the reason for their specific localizations in the cell. It wasn't just the nature of the talk but the manner in which the talk was delivered that captivated the audience. Other talks in the conference covered areas from nuclear pores to cutting-edge microarray technology. Overall, the conference presented a fair insight into the interesting work being done in biophysics.

Shift to Vaccines

By Vy Hoang

Due to high failure rates in clinical drug trials, generic drug competition and loss of patent protection for blockbuster drugs in the upcoming years, big pharmaceutical companies are shifting their research attention towards creating new vaccines to prevent a range of illnesses and diseases. During the booming years of drug discovery, vaccines were overlooked by the pharmaceutical industry, but now vaccines are back in the spotlight because of a growing global market.

Various factors have renewed this interest in vaccines, such as rising prices and an increased demand for vaccines. In general, vaccines are a safe investment because they work. Since 1900, mass vaccination along with improved sanitation has increased the life expectancy from 47 to 77 years in wealthy countries. Just one or several doses of a vaccine can produce a lifelong protection against infections, which in the long run is very cost-effective to the consumer and insurer. This leads to the reason why vaccines are increasing in demand.

With the growing populations and economies, more money is being spent on healthcare. Developing countries such as China and India, which are emerging healthcare consumers, are increasingly spending money on vaccines in order to prevent diseases and save money in the long term. Nonprofit organizations, as well, are buying lots of vaccines to distribute among underdeveloped countries, thus escalating demand. Additionally, governments are willing to buy and stockpile large amounts of vaccines for naturally occurring diseases and against bioterrorism agents. For example, in the recent pandemic H1N1 influenza, governments worldwide spent several billions of dollars for the H1N1 vaccine.

Of course, another incentive for pharmas to invest in vaccines is the rising price of vaccines. In the 1980s, standard vaccines (measles, mumps, rubella) for children were around $7 a dose but now the price has shot up to $28. Even common vaccines (tetanus and pertussis) jumped ten-fold, from $1.60 to $16.64. Several pharmas, including Novartis, Sanofi-Aventis, and Merck, reported that many years ago only 3-5% of their revenue came from vaccines, compared to now, where it is about 11-15%.

All of these factors are good news for the big pharma companies who are struggling with drug discovery. These companies are currently working on new vaccines that will tackle diseases such as malaria, tuberculosis, and AIDS. Thus, the best news is for us, the consumers who are benefiting from their success.

Bringing Minds Together

By Isha Antani

Yes, many of us are aware of the issues that the current pharmaceutical industry is facing. It’s a well-known fact that more money being pumped into the drug discovery process is resulting in less drug discovery. This is due to high failure rates of potential drug candidates either in Phase I (safety) or Phase 2 (efficacy) of the pipeline. 

Academia, while known to have a good target discovery rate, is weak in the area of drug discovery and often has to spend a good amount of time seeking financial means to support its work. Additionally, while pharmaceuticals spend large sums on high-risk drugs with blockbuster potential, academia is more attuned to researching rare cases and is often funded by foundations set up for these specific cases. Both pharma and academia could use a bit of help from one another. And, slowly, both are.

Up until the 1980s, there was a huge barrier to get pharmaceuticals to invest in targets discovered in academic labs. Academic labs that had been funded by the government were not allowed to patent their efforts. Therefore, pharmaceuticals didn’t really have the incentive to go after a drug that any other company could also potentially develop. However, in the 1980s the Bayh-Dole act was passed which allowed patents to be placed on federally funded research projects, leading to a much better technology transfer between the two arms. However, true collaboration was still lacking. 

Now, due to the current dismal state of drug pipelines, the two sectors have been coming together for various initiatives with hopes that better targets generated by academia will lead to better (and cheaper) drug development. AstraZeneca, for example, has collaborated with Columbia University to understand the area of metabolic diseases while Pfizer, a big believer in academia-industry collaboration, has multiple projects with academia. The collaboration makes sense and in a way, there are incentives for both parties involved—the academic labs, who carry the focused expertise, retain the intellectual property rights on their discovery while the pharmas, who are profit-making companies, save money in the discovery process and get the right into developing the process and marketing the discovered target in their preferred manner. It’s a win-win situation for both parties and for the greater public, who gets to see effective, cheaper drugs on the market at a faster rate.

Second Annual Molecular Programming Project Workshop

By Traci Freed

The Second Annual Molecular Programming Project Workshop is being held this weekend in Oxnard, CA.  Walter Fontana will be presenting a talk on January 10th at 10 am.  If you are attending this conference, be sure to go to Walter's talk to learn more about Cellucidate, our web-based, collaborative workspace for cell biology.  We hope to see you there!

New Cellucidate Features

By Traci Freed

In recent months we have received valuable feedback from our users and wanted to share information on some exciting, new features that have been introduced to make Cellucidate not only easier to use, but a more efficient platform for doing work.  If you are not a current Cellucidate user, register for an account at www.cellucidate.com.

Save Multiple Simulations
The redesigned model page lets you run a simulation multiple times using different parameters each time and then save these simulations to easily compare the results of each change of parameters.  Manage your simulations to keep those that are important in your research and delete the ones you do not need.

Copy cBooks Faster
The copy feature has been updated to significantly improve the speed at which cBooks are copied.  Remember when you were unable to copy those large, complex cBooks?  Well, now you can copy them quickly no matter the size.  If you've also wanted to share a cBook with a colleague to copy but haven't been able to because it takes too long, now is the time to share it.

Improved Search
Cellucidate users have been asking for a better search function and based on that feedback we've improved search so you can better find the cBooks you are looking for.  Most users want to find cBooks about a specific pathway and with the old search it was difficult to find the particular content you needed.  We've updated this feature so that when searching you will only be given the results for cBooks and users so you can better find exactly what you are looking for.  Why not check it out for yourself?  We've got some great content related to the EGFR pathway and you can see the new search in action.

New Desktop Application
We are excited to announce a new feature many of you have asked for - a desktop application for building Cellucidate models.  We are almost done building this functionality and the new application will provide an environment for writing Kappa text files directly with the added benefits of syntax highlighting, error checking, rule visualization, simulations and static analysis tools such as the contact map.  The new application also integrates Cellucidate to facilitate easy upload of models to Cellucidate.  Those of you who have used our old desktop application, Kappa Factory, will love this new one!  If you would like to be notified when an advance beta copy is available, email us at support@cellucidate.com.

Cellucidate Help Resources
We recently introduced comprehensive help and support resources to assist you as you learn to navigate Cellucidate, use its many features and incorporate it into your daily workflow. The detailed help content includes topics such as "Getting Around Cellucidate" and "Building a Model".  Get stuck while using Cellucidate?  Feel free to ask for help with your issues and questions, request new features you'd like to see and interact in discussions with other Cellucidate users.

As we continue to evolve the Cellucidate experience, we welcome your feedback.  Be sure to email your feedback to support@cellucidate.com.

Universal Analysis of Complex Networks

By Joshua Havumaki

We live in an age of complexity. Large data sets and intricate networks are ubiquitous and underlie many daily events from flight patterns to the spread of H1N1. Biological networks (like those explored by proteomics) are effectively mapped using statistical and other computational approaches.  Although these networks can be represented relatively easily, the technology used to generate the data is startlingly inaccurate. For example, yeast two-hybrid, a commonly used method for discovering protein-protein interactions, is considered to have "medium" accuracy - on the order of 50% false positives.

A recent article in PNAS titled, "Missing and Spurious Interactions and the Reconstruction of Complex Networks" by Guimera et al. outlines a method to accurately analyze different types of complex networks. It strongly relies on the fact that the nodes in networks can be organized into groups. Group membership governs which nodes can interact. 

The algorithm predicts missing or incorrect interactions and reconstructs networks from noisy data. It was successfully applied to five different types of networks ranging from connections within a neural network to interactions within a pod of dolphins, and in all cases it effectively separated signal from noise. This was confirmed by introducing errors into the network that were culled by the algorithm.

Although it has only been tested on relatively small networks, the researchers suggest applying this method to larger protein-protein interaction networks. If successful, this could begin to remove the plethora of false positives created by our somewhat crude technologies.

Uncertainty in Modeling

By Ty Thomson

Earlier this week I was talking with a co-worker’s husband, and he was asking about exactly what we do here at Plectix.  I was trying to explain the idea of biological modeling to him, and I realized that I wasn’t entirely sure if this type of modeling differs from any other.

Like all modeling, you start off with a bunch of knowledge and assumptions.  The model is just an expression of the knowledge and assumptions in a formal (generally mathematical) way that allows you to use computational power to reason about the system (often directly though simulation).

From my understanding, the mathematical descriptions of chemical kinetics are actually fairly accurate, so assuming that the same principles apply to protein-mediated reactions, it would seem that the overwhelming source of inaccuracy in models of protein networks is whether the reactions themselves are right.  Thus, for modeling in systems biology, lack of knowledge is the problem, not representation of the knowledge in a formal/mathematical form.  Is this the case with other forms of modeling?  Or, for example in financial modeling, do we understand the system being modeled but have poor or inaccurate mathematical descriptions of the pieces of the system?

The inaccuracies in biological models appear to me to be firmly rooted in the overwhelming uncertainty that we have in our understanding of what’s actually happening in living systems.  It’s actually pretty valuable that our mathematical descriptions are reasonably good, so that changes in our hypotheses or “knowledge” can be evaluated directly via models without worry that we’re just evaluating our mathematical descriptions of events.  After all, we are much more interested in improving our knowledge about the system itself than in understanding the mathematics required to describe it.  So while biological modeling is pretty lucky in this respect, is this also the case for modeling in other fields?  I’d be interested to find out.

Moving to the Cloud

By Vy Hoang

With the recent interest in cloud computing, it isn’t surprising that big pharma companies, such as Pfizer, Eli Lilly and Johnson & Johnson, are also experimenting with the cloud. Cloud computing offers many advantages for a low price. These companies can easily predict the time spent and the cost of a job on the cloud if they know what job is pushed into the cloud. Thus, the pharmas can manage their money and time in a more cost-efficient way. In addition, cloud computing is fast because of the access to a vast number of parallel CPUs. Pfizer reported that running their Rosetta macromolecular modeling took only 3.5 hours using 500 of Amazon’s Elastic Compute Cloud (EC2) instances instead of 48 hours on their internal computers. The New Jersey based Johnson & Johnson is hosting analytical software such as NONMEM (non-linear mixed effects modeling) in the cloud. Taking advantage of the on-demand cloud infrastructure, Johnson & Johnson is able to scale up or down to run large simulation jobs at will. Eli Lilly is running both non-research and discovery applications in the cloud, a few of which could not run internally. Therefore, the cloud can provide more efficient, scalable servers to run jobs that couldn’t be run in house.

Although the results look promising, cloud computing poses some challenges for pharma companies. The main issues are security and accessibility. Some people view the cloud as an unsecure space and mistakenly feel that cloud resources are just extra servers that Amazon is not using. However, as they learn more about how the model works with its on-disk encryption, in-memory encryption and wire encryption, the people are beginning to come around. Yet, the complexity of cloud computing is an additional factor that is making cloud computing difficult to use. Without the help of an IT team, the pharmas would have considerably difficulty in understanding how to use the cloud. Pfizer stated that it was hard to get metrics out of the cloud such as what jobs are running or how long they are taking without any IT help. Whatever’s going on in the cluster is not as simple as you think. The pharmas simply want the cloud to be easier to use with less technical complexity so the users can do what they need to do. Even with these challenges, pharma companies are not deterred because the vendors are giving in to their demands, making it more manageable to use the cloud.

These early cloud adopters believe that cloud computing is the next step in improving the quality of research. Researchers will be able to solve problems in the cloud that were considered impossible before. As long as you know what you’re getting into, cloud computing can be the right way to go.