Structural Bioinformatics was founded five years ago with the goal of speeding drug discovery via computational proteomics. The company was recently awarded a two-year, $754,259 SBIR grant from the NIH to develop a small-molecule nonpeptide inhibitor of anthrax lethal factor it discovered using its approach. BioInform recently spoke to Kal Ramnarayan, SBI’s co-founder, vice president, and chief scientific officer, about the company’s technology and the role of computational proteomics in drug discovery.
SBI’s activities in structural determination are half experimental and half computational. Could you provide a bit of information about the computational side and how it may differ from other approaches?
What we’re doing that’s unique on the computational side of things is our ability to discover micromolar and sometimes nanomolar inhibitors starting from the protein structural information, which can come either from homology modeling of the type that we do or from external crystallographic or NMR structures.
We have a methodology called augmented homology modeling, where we take sequences from proteins and then we calculate three-dimensional structures. Then we have what is called DynaPharm technology, which takes these protein structures and uses certain molecular dynamics-based applications to come up with pharmacophore elements that are used for screening chemical libraries. We have worked on at least 10 different targets so far and these targets range from enzymes to protein-protein interaction targets, and in each one of these cases we are able to generate lead molecules very quickly.
Were the structures of the 10 targets that you’re working with identified using your computational tools?
The structural information for eight of the targets out of 10 was derived from augmented homology computer modeling.
What is augmented in that approach? How does your method differ from other homology-based methods?
A lot of structural homology-based approaches use a black box-type approach to find a match between a sequence that is to be modeled and known structures. Then, if they find a homology, they build a structure and fill in gaps or build loops in the structure by searching through databases that exist for other crystal structures. Then they latch those loops onto the protein to complete the protein model. We, on the other hand, use very intensive and human intervention-based homology modeling, where at every stage the model is checked for quality in terms of alignment and secondary structure prediction, for which we have a proprietary methodology based on neural networks. We also have additional fold-prediction algorithms that we use to go into very low-sequence-homology proteins, and algorithms to re-build loops on the protein surfaces, if needed, by ab initio simulations. Eventually we conduct extensive molecular dynamics on our proteins followed by quality-control procedures.
How many staff members do you have working on this?
We have 22 computational people. There are a total of 77 people in the company, with 26 wet lab people. Our staff includes MDs and PhDs in physics, chemistry, biology, biophysics, and computer science. We also have software engineers, Oracle engineers, and even people with PhDs in rocket science! It’s very interdisciplinary.
Would you say that bioinformatics for protein science needs to be even more interdisciplinary than other types of bioinformatics?
Yes, because there are a lot of applications that are very intense such as pattern recognition, sequence matching, string matching, and structural matching, and this is brought up by different disciplines. Bioinformatics as we knew it is no longer true. It isn’t just a question of sorting through data, finding information, and building databases. It’s beyond that — it is extraction of knowledge from all the data. It’s easy to put the data together, but you have to extract knowledge, which means you have to bring new technologies and new algorithms into play.
So this same technology is used both for your in-house drug discovery activities and to populate SBI’s databases?
All the software and database engineering that we do for the bioinformatics platform is made available to our customers on a subscription basis and is used extensively to find leads for our customers.
IBM has made an investment in SBI and the company is using some IBM equipment. Are you collaborating with them on a research level at all?
We’ve been working together for quite some time now and I advise IBM on their Blue Gene project.
What are you doing on the Blue Gene project?
IBM is designing the hardware as well as certain algorithms for protein prediction and protein folding. I meet with them and the rest of their Blue Gene advisory board every year to discuss the state of the art of protein folding and what other new grand challenge problems they should be considering in life sciences for potential Blue Gene applications.
Are they going in the right direction?
They want to take an outside-the-box approach. They want to do things that have never been done before. Time will tell whether they can really deliver or not.
IBM is working on a smaller version of Blue Gene. Is that something that would be of use to you?
Absolutely. If the structures generated are good, then definitely it would be a valuable addition to what we’re doing in house. Otherwise, the Blue Gene computer — even if it is a smaller version — will be quite powerful enough for other problems in life sciences.
Do you use other computer vendors besides IBM?
Most of the high-performance today is handled 80 percent by IBM machines. We have a 128-processor Linux cluster that we’re very happy with. We also have a 16-processor Silicon Graphics system wherein we do a lot of the docking studies. Most of the daily computing activities are supported by Sun Microsystems, a lot of Windows NT machines, and Silicon Graphics machines.
I’ve heard that SBI might change its name to something more pharmaceutically oriented. Is that true?
It’s a possibility.
Is keeping “bioinformatics” in your name really that bad of an idea these days?
Well, not exactly! We want to have a name for the company which truly reflects our capabilities and technologies. The term “bioinformatics” is over-used and it means different things to different people. If you look at Wall Street, most of the people are saying if you''re a tool and platform company, your valuation is not very high. Look at Incyte and Celera. People thought those databases had enormous value, but the value that one can obtain today is down significantly from prior years. They have changed their business models to generate a revenue stream and the bioinformatics tools companies are also struggling to maintain a revenue stream. What we want to do is build on our platform technology and our structural proteomics capability, and also enhance our drug discovery capabilities.
Has SBI always had this hybrid business model?
Yes. From the beginning we had interest in short-term revenue generation using our platform technology, which is databases and our service business, and long-term revenues by means of drug discovery collaborations.
Do you see any particular advantages in the way you’ve approached this hybrid model as opposed to the way some of your competitors have?
A significant difference from our competitors is that we have already demonstrated our technology by working on 10 different targets, several of which are being advanced by our partners in clinical development. Many of our competitors are just starting up and some of them have changed their business models. If I were a potential partner and I had to make the build vs. buy decision, I''d want to know how much I can save in the overall R&D budget and whether the technology has been proven.
So how do you convince potential customers that you’re able to save them money?
We are able to discover leads in 60 to 120 days. Many pharmaceutical companies have a need for more leads because there are so many new targets and the more targets you can get in your pipeline, the better off you are down the line. Resources are always in short supply and we offer a cost-effective approach with a proven track of success.
How effective is your technology for target validation?
Target validation has to come from the biological side of things. We can help with targets in terms of their functional characterization by looking at structural features of the protein using some of our pattern-recognition algorithms, and can also provide some insight. Unfortunately, the technology today to [computationally] validate the function of the protein is not available.