Medical equipment giant Philips is taking the initial steps to tackle the emerging molecular medicine market via an early-stage bioinformatics project at its US research facility in Briarcliff, NJ.
Philips researchers will present the first fruits of the effort â€" a genetic algorithm for classifying microarray and proteomic data sets â€" at this week's IEEE Symposium on Computational Intelligence in Bioinformatics and Computational Biology in San Diego.
"We're on the verge of an explosion of new knowledge in molecular medicine, and that's likely to make a large impact on clinical practice, and being a major medical equipment company, we certainly have to be aware of this and be following the latest trends," Dave Schaffer, a Philips Research fellow who heads the bioinformatics effort, told BioInform last week.
Schaffer stressed that the effort is a "long-term dance" that has only just gotten underway at the company, but it's certainly a growing initiative. Philips is currently seeking two more staffers to join the bioinformatics team (job descriptions are no longer available), and is also seeking clinical partners to collaborate on improved bioinformatics methods for high-throughput experimental data sets.
"There are simply too many measurements on too few patients to constrain the number of patterns that fit. We can, in some cases, find literally tens of thousands of patterns that appear to fit perfectly. We don't trust them."
The bioinformatics initiative is one example of a broader effort underway at the company to gain a foothold in the molecular diagnostics market. In February, Philips Research announced an agreement with Eindhoven University of Technology, Maasricht University, and Maasricht Academic Hospital to create the Center for Molecular Medicine in Eindhoven, the Netherlands. Work at that center will focus on molecular imaging. While this is an area beyond the scope of the project in Briarcliff, the goals of the two efforts are essentially the same, Schaffer said.
Philips is pursuing these initiatives "with the intent of developing new diagnostic tools that we could sell into the clinic that are based on new technologies for measuring biomolecules," he said. "Bioinformatics, of course, has to serve that goal by helping us do things like discover more reliable statistical patterns in, say, microarray data or mass-spectrometry data, or whatever the measurement technology is being used."
Nevenka Dimitrova, another Philips Research fellow working on the bioinformatics effort, added that the company's traditional medical portfolio is in diagnostic imaging, "and now we have projects in molecular imaging, but we see that in the future, in order to improve the clinical outcomes, we have to go at the cellular level, and that's the end goal."
This strategy is in line with that of other large medical systems firms, which set up in-house bioinformatics development efforts before jumping headway into molecular diagnostics. GE, for example, set up a bioinformatics group more than a year before it purchased Amersham in a bid to flesh out its molecular diagnostics effort [BioInform 07-04-03], and Siemens launched an exploratory bioinformatics project of its own about a year and a half ago [BioInform 06-21-04], and has since signed research collaborations with Biomax and Sequenom and has purchased the biochip technology division of Infineon.
An Embarrassment of Riches?
For now, the Briarcliff team is cutting its teeth on publicly available data from the microarray and proteomics communities â€" specifically the Golub leukemia data set, the Veer breast cancer data set, and the NCI ovarian cancer data set.
Philips is pursuing these initiatives "with the intent of developing new diagnostic tools that we could sell into the clinic that are based on new technologies for measuring biomolecules. Bioinformatics â€¦ has to serve that goal by helping us do things like discover more reliable statistical patterns in, say, microarray data or mass spectrometry data."
The Philips Research bioinformatics team has a strong background in pattern recognition, while Schaffer's own area of expertise is in evolutionary computation, so these well-studied data sets served as a convenient proof of concept for the team's approach, which has been adapted from previous work in the area of genetic algorithms. So far, Schaffer said, Philips' methods have proven to be "embarrassingly powerful â€" so powerful, in fact, that they show off the weakness of the small clinical datasets."
This realization has led to a new challenge for the team, Schaffer said. "There are simply too many measurements on too few patients to constrain the number of patterns that fit. We can, in some cases, find literally tens of thousands of patterns that appear to fit perfectly," he said. "We don't trust them."
The next stage, Schaffer said, will be gaining access to more biological data through partnerships with clinical research groups. So far, he said, Philips has some "tentative collaborations" with experimental partners, "but nothing official that we can discuss."
As part of that goal, the Philips team is trying to determine the optimal experimental sample size that will produce a statistically significant result, but "the kind of knowledge you need, unfortunately, to answer that question is missing," Schaffer said. "You really don't know what the variability in a large population is on these molecular measurements. That's the kind of question you get into when you start doing hypothesis testing. I think the field, it's fair to say at this point, is really in the phase of hypothesis discovery."
Dimitrova said that the Philips researchers also plan to expand the scope of their work into new algorithm development. "What we are seeing is that applying statistical methods just in a straightforward manner â€" and especially well-known methods in machine learning â€" is not giving results that can translate into immediate understanding of the biology." This means, she said, "that we need to go a little deeper into finding new algorithms."
Dimitrova said that the Philips team is already expanding into new areas of bioinformatics analysis through an "informal" collaboration with Michael Zhang's group at Cold Spring Harbor Laboratory in the area of epigenomic analysis. She stressed that this is "work in progress," however.
Philips Research has not yet identified a clear commercialization path for any of the technologies under development in the bioinformatics group. The work is still very early-stage, and any commercialization plans would be made by Philips' product division rather than the research staff.
Betsy McIlvaine, a Philips Research spokeswoman, told BioInform that "the job of research inside Philips is to provide options to Philips' product division. So clearly our first line is working with our medical systems division." That division, she added, "is becoming one of the largest of Philips' divisions" as the company is "increasingly becoming a healthcare lifestyle and technology company."
â€" Bernadette Toner ([email protected])