Tokyo-based Daiichi Pharmaceutical, a subsidiary of Daiichi Sankyo, took a risk five years ago when it opted for a still-unproven in silico approach to predict protein-protein interactions from Celestar Lexico-Sciences.
Daiichi decided to use the approach, which Celestar calls PPIP (protein-protein interaction prediction), as an alternative to the yeast two-hybrid system, which is the most commonly used experimental method for identifying protein-protein interactions, but is notoriously problematic with regard to efficacy and predictability.
Before the collaboration with Celestar, Daiichi used yeast two-hybrid "because it was the most available and accepted method at that time," Yasuhide Hirota, manager of Daiichi's R&D planning group, told BioInform via e-mail. "However, the results of Y2H were very unstable. It was very troublesome to decipher the protein-protein interactions using the yeast system and the process would take between 6 to 12 months to obtain results."
In contrast, Hirota noted, "Celestar's prediction technology saves time, and therefore cost." The prediction technology identified candidate proteins that could interact with the bait protein "in only one or two weeks" with "very high" accuracy "about 40 percent, [which is] better than the Y2H method," he said.
In some cases, PPIP offers much higher accuracy than Y2H, while in other cases, it is "comparable." Nevertheless, "it's certainly a lot faster, and it's generally higher."
Daiichi's risk appears to have paid off. The company said in a statement in December that it identified more than 220 novel protein-protein interactions using the technology, and has filed patent applications for more than 44 of those interactions with "relevance to diseases."
Daiichi is pursuing one of the interactions between telomerase and mitogen-activated protein kinase-activated protein kinase 3 as a possible target for a cancer therapeutic.
Ian Welsford, manager of application science at Fujitsu BioSciences, Celestar Lexico-Sciences' US distributor, described PPIP as a "customized service" that combines text-mining with sequence modeling.
In the first step of the process, Welsford said, the approach uses natural language processing to extract "tuple" relationships from Medline that indicate likely associations between proteins, such as "protein A phosphorylates protein B."
Next, the method uses protein sequence motifs associated with particular signaling behavior "to match sequence information to annotations of those words," Welsford said. The result, he said, is "sort of a large and/or statement: This looks like it is an entity that is phosphorylated by protein B and is trafficked in the following way, and because it's part of a pathway that's trafficked the same way as protein C, then the hypothesis would be it's a member of the same pathway intermediate between A and C."
Welsford said that the accuracy of the method depends on the particular proteins that are involved. Protein families that are well described in the literature, such as kinases, tend to provide better results than less well-studied families, he said.
In some cases, Welsford said, PPIP offers much higher accuracy than Y2H, while in other cases, it is "comparable." Nevertheless, he said, "it's certainly a lot faster, and it's generally higher."
Fujitsu is still in the "early stages" of rolling out PPIP in the US, according to Michael McManus, vice president of the BioSciences group.
So far, he said, a large pharmaceutical firm that is already using Celestar's LisH (large-scale in situ hybridization) technology is "talking to us about combining PPIP with the LisH method," he said, "and we have another pharma who is starting to explore this with us as well."
McManus stressed that Celestar's business has mainly been focused in Japan to date, but as the company starts to grow, "we'll hopefully make more progress here in the US."
As for Daiichi, its collaboration with Celestar officially ended in December, but Hirota noted that Daiichi is still in the process of merging with Sankyo. "After this merger is complete we would plan to partner with Celestar Lexico-Sciences," he said.
Bernadette Toner ([email protected])