Recent biomedical research shows that many serious diseases – in particular cancer – are caused by persistent disruptions of cellular communication processes. These processes use a biochemical reaction called phosphorylation to control the information flow. This biochemical computer is faulty in cancer, and drugs that block rogue phosphorylation reactions have proven effective cancer treatments. However, despite half a century of intense research, discovering these critical phosphorylation reactions is still slow due to the extensive manual effort required.
Automated, highly confident prediction of previously unknown phosphorylation reactions is therefore much desired. This can lead to substantially increased rates of discoveries in cellular computing, which can in turn deliver the targets for more efficient cancer drugs.
Fujitsu Laboratories Ltd., Fujitsu Ireland and Insight (National University of Ireland Galway) have worked since 2015 on TOMOE, a novel discovery informatics platform based on knowledge graphs and statistical relational learning. The Irish part of the TOMOE team (led by Dr Pierre-Yves Vandenbussche and Dr Vit Novacek) has recently initiated a proof-of-concept collaboration with Systems Biology Ireland (University College Dublin) led by Professor Walter Kolch. The team has applied the automated discovery technology to the use case of phosphorylation prediction. The project focuses on phosphorylation reactions that could be cancer therapy targets, but the engine can also automatically generate predictions for types of phosphorylation reactions that have gone awry in other diseases. This will make it a powerful tool for discovering the next generation of drug targets by computation guided experiments that will allow us to systematically analyse and understand the role of phosphorylation in living organisms.
As of March, 2017, the team has finished working on the first functioning prototype of the prediction engine. The prototype is currently being tested on a comprehensive knowledge graph covering known phosphorylation reactions and related protein interactions in humans that are available in machine-readable format. Further implementation details and preliminary results will be released later this year.