Standfirst
By combining sequence motifs with biological information about kinases and phosphoproteins, researchers develop an algorithm, NetworKIN, to predict in vivo phosphorylation networks.
The problem in cell biology is not so much a lack of information, but how to combine the pools of knowledge that exist. A perfect example of this is the problem of assigning specific kinases to phosphorylated proteins, which—thanks to recent developments in mass spectrometry—are being discovered with improved quality and increasing speed.
When Rune Linding, with a doctorate degree in hand for computational prediction of protein interaction motifs, joined the laboratories of Tony Pawson (Samuel Lunenfeld Research Institute) and Michael Yaffe (Massachusetts Institute of Technology), veterans in the field of phosphorylation-mediated cell signaling, the stage was set for a fruitful collaboration, resulting in NetworKIN, an algorithm that predicts networks around phosphoproteins and their kinases.
Every kinase recognizes a specific linear motif around a serine, threonine or tyrosine. The presence of this motif, however, is not enough to unequivocally link a kinase and a protein substrate. The scientists realized that to accomplish their goal of matching kinase and substrate they would have to look beyond sequence motifs and take cellular context—for example, expression level or cellular localization—of all players into account.
Because developing a computational approach that integrates these factors from scratch would be unfeasible in terms of time and computational power, Linding sought out collaborators who could provide protein network information. His search led him to the lab of Peer Bork at the EMBL and the STRING database, which contains information on over 1.5 million interactions in almost 400 sequenced genomes. He says, "We had to do this with EMBL; no other community resource would allow you to do this, both in terms of hardware and software."
Now that they had all their collaborations lined up, the researchers developed NetworKIN; they matched phosphorylation sites with known kinase motifs and used the context network derived from STRING to arrive at the final prediction of the kinase that phosphorylates each site.
They compared the performance of their algorithm to a purely motif-based search and saw that NetworKIN 'won' hands down. Linding calculated that as much as 60–80% of substrate specificity comes from context rather than sequence motifs.
Not prone to bask in their success, Linding and Pawson already plan to expand NetworKIN. They want to include common interaction motifs and make the algorithm available as a webserver. Linding predicts, "people will be able to upload their proteins with the phosphorylation sites and then get predictions."
The scientists emphasize that the algorithm yields most biological insight if applied in a data-driven approach. Rather than look at a whole phosphoproteome, Linding suggests focusing on sites that are dynamically regulated in response to a certain treatment. He forsees that "NetworKIN will be very powerful for modeling when applied to a specific system."
This algorithm has the potential to be of great use to the scientific community, plans for its expansion are in place, but what is not in place yet is the funding to guarantee NetworKIN's long-term survival.
Linding and Pawson comment wistfully that it is relatively easy to start a new database, but that its maintenance is more difficult. "It is also a question of political will," Linding says; "there is very little support to maintaining things, but when you want to do really good science, you need to maintain the databases and keep them current all the time."
Let us hope that those exercising the political will agree.