NEW YORK, March 29 - Protein interaction networks are a lot like the “Six Degrees of Kevin Bacon” game, where any celebrity can be directly linked to the actor by six or fewer degrees of separation, according to Edward Marcotte, a scientific co-founder of Los Angeles-based bioinformatics startup Protein Pathways.
So, in order to determine protein pathways, Marcotte and company co-founder Matteo Pellegrini, have developed a computational method that “constructs a large network based on local relationships,” Marcotte said in presenting his data at the Genomics 2001 meeting at Harvard University on Wednesday.
“This is how proteins interact in the cell,” he added.
Marcotte and Pellegrini initially developed the method as postdoctoral fellows at the Lab of Structural Biology and Molecular Medicine, a lab jointly run by the University of California, Los Angeles and the Department of Energy, and bioinformaticists at Protein Pathways have refined it and used it to create Proteome Navigator, a database of protein functions and interactions across 80 microbial genomes.
The year-and-a-half old company just began marketing Proteome Navigator to pharma and biotech companies for anti-microbial drug discovery and is now integrating information from the human genome into it.
Pellegrini, who serves as president of the company, hopes to leverage the $6 million in first-round financing obtained last May, along with a second round of financing he is currently raising, to make this tool widely available for human drug discovery.
“The strategy of the company is now to move into human biology, to create a product that is the computational tool to do discovery of protein-protein and protein-small molecule interactions” in human, said Pellegrini.
To predict protein pathways, Marcotte and Pellegrini’s method layers multiple algorithms that compare protein coding sequence data across multiple genomes, and integrates them using the computational power of a Beowulf cluster on Linux machines.
One algorithm, the phylogenetic profile, groups proteins according to which genomes in which their coding sequence appears. Marcotte and Pellegrini found that proteins with similar inheritance patterns also have similar functions and can tend to participate together in biochemical pathways.
Another algorithm, the Gene-Neighbor program, clusters proteins based on similarities in their relative position on multiple genomes. For example, if proteins A and B are sufficiently close to each other on genomes A, B, C, and D, the program calculates that there is a statistically significant relationship between the two genes.
Similar algorithms look at whether proteins in one species are fused in certain genomes, perform cross-genome similarities in regulatory sequences surrounding them, and compare their mRNA expression patterns.
These algorithms are combined in an overall model, and cross-referenced with the publicly available Database of Interacting Proteins, which was developed at UCLA.
“What’s unique is that [protein pathways technology] delivers a number of computational tools that integrate functional and interaction information from a number of different sources,” said Marcotte.
Marcotte, Pellegrini, and others have published the algorithms in numerous scientific papers. In his academic research, Marcotte has used the methods to predict the function of half of the yeast genes whose functions were formerly unclassified, and Alex Vanderbleek, a researcher at UCLA, verified the mitochondrial proteins predicted by this program, Marcotte said.
Last year, the company also signed a deal with UroGenesys, a Santa Monica, Calif.-based prostate cancer drug discovery company, to predict protein interactions in Urogenesys' database of cancer-associated antigens. Protein Pathways also received $100,000 in funding from the National Institutes of Health to use its computational methods to predict functions for 1,500 genes from the Mycobacterium tuberculosis genome.
Now, the company’s 14 employees are working to refine the computational methods developed by Marcotte and Pellegrini. “In terms of predicting interactions, we have to integrate information about structural biology, and inter-molecular interactions,” said Pellegrini. “By the end of this year, we will probably have a far more sophisticated tool to prove interactions.”
Meanwhile, Marcotte has moved on to use the program in his academic research, in his newly appointed position as assistant professor at the Institute of Cell and Molecular biology at the University of Texas. Marcotte is currently utilizing the program in his academic research, and hopes to license out any commercial applications to Protein Pathways through the University of Texas- Austin’s technology transfer office.
Protein Pathways is planning to offer a demo of Proteome Navigator on its website, www.proteinpathways.com , starting in May.