The microarray analysis market is relatively mature, compared to other areas of the bioinformatics sector, but that isn't stopping startups from staking new claims in the space.
Informed Bioscience, a six-month old firm based in Bozeman, Mont., is probably one of the field's more recent arrivals, but the four-person company is targeting a niche area of microarray analysis subgroup classification that it hopes will set it apart from the field's more established players.
Last week, the company announced that it had begun beta-testing its GenePart software, which is based on a classification algorithm developed at Montana State University. The beta program involves several undisclosed academic and industry researchers, who are trying out the program for 60 days, at which point the company will gather feedback and prepare the product for an anticipated late fall launch date.
"The marketplace [for microarray analysis software] is more and more competitive," Weston Fricke, operations manager for Informed Bioscience, conceded. However, he told BioInform, "the software that we've created is focused on one specific functionality, and that is to obtain better information from those microarray data sets where class discovery has been difficult, or where there is still ambiguity with respect to both gene expression and the appropriate partitioning of the respective data into classes."
Brendan Mumey, an associate professor of Computer Science at Montana State University and founder of the company, said that the GenePart algorithm should be particularly useful in cases where researchers are looking to uncover novel subgroups within very large data sets, such as cancer subtypes or subclasses of drug response. The method generates better results than other methods for unsupervised classification, such as hierarchical clustering, which rely on heuristics and can provide only a "locally optimal solution," Mumey said. Those methods "don't know anything about the global properties of the data set."
Mumey said that the GenePart algorithm considers clustering "as a combinatorial optimization problem," so it is able to simultaneously partition a large data set into the correct classes and subclasses, while also identifying those genes that are most characteristic of each class. Using a "color-change statistic" that evaluates how well a gene characterizes the potential partitioning of tissue samples, the software performs a combinatorial search to identify the optimal clustering that is supported by the best set of genes, Mumey said.
"There are basically two problems: You want to find the correct clustering, and you also want to find a set of genes that best characterize the clustering," Mumey said. "So in our approach, what's new is that we try to solve both those problems at the same time."
The software addresses the computational complexity of the search problem with a "branch-and-bound" technique that "prune[s] off portions of the search space that we can guarantee won't have the optimal solution," Mumey said.
In internal validation studies, Informed used GenePart to analyze a 30-sample data set for cutaneous T-cell lymphoma. According to the company, the software was able to discriminate short-term survivors from long-term survivors and normal controls, with only one error, without any prior knowledge about the clinical outcomes of the patient samples. The company has not published the results of the study, however.
Nor has it published any peer-reviewed papers on the method itself. "We have some stuff in submission, but nothing formal yet," Mumey said.
Nevertheless, the company has the support of MSU and its surrounding economic development community. Fricke said that Informed has "close ties" with the university as well as its tech-transfer office, through which the company negotiated the licensing agreement for GenePart. The firm is housed at TechRanch, a Bozeman-based technology incubator.
Nick Zelver, technology manager at MSU's tech-transfer office, told BioInform that bioinformatics software makes up a "very small subset" of the office's broader technology portfolio, but the GenePart algorithm "popped out as being attractive" for commercialization because of its "unique aspect" of simultaneously classifying samples and identifying specific genes.
Zelver said that it is difficult to estimate the potential market for GenePart, and that assessing the software's commercial viability will be one of the goals of the beta project. "From a revenue standpoint, that's what we leave up to the companies to give us feedback [about], and it's hard to tell truly what kind of sales there are in a particular product. That's always challenging."
However, he added, "because of this differentiation by doing both classification and gene identification, as we move forward and it shows that it will provide what's needed, it could quickly move into a niche and fairly quickly have a market position."
The tiny company views its product as complementary to those of the big microarray analysis firms such as Rosetta and Silicon Genetics, and Fricke said there is a chance that the firm may consider licensing the software "to a bigger player," but added that "the mission of this company is probably to do it in house because there's some secondary technology that we're going to start working on in the fall."
Bernadette Toner ([email protected])