Recently, Google researchers made a breakthrough machine learning algorithm, Alphafold, that predicts the 3D shapes of all known proteins.  Behind the scenes, Alphafold represents proteins as data networks, called graphs, with graph nodes of amino acids connected via edges of chemical bonds or biological links, and then finds correspondences among these amino acids. This fundamental problem, called graph alignment, is important not only in biology but also, e.g., for social network interactions. However, it has a severe efficiency bottleneck: graph alignment algorithms take days or even years if the graphs exceed as few as 100K nodes, which is tiny in many such applications.

In this project, Mottin and Assent aim to tackle the efficiency bottleneck of graph alignment by creating new algorithmic principles and solutions for next-generation computing hardware:  IPUs embed many CPUs in a single microchip but lack algorithms for their exploitation.  GAI proposes graph alignment algorithms that attain a considerable speedup for small graphs (<100Knodes) and scale to graphs with millions or billions of nodes in less than one day.

FACTS om AUFF NOVA

The aim of AUFF NOVA is to stimulate courageous and innovative research projects of high quality – projects, which may have difficulties obtaining alternative funding. The project must be pioneering in its field and show clear potential for scientific breakthroughs. The hypothesis or problem behind the project may require development of new methods and it may challenge existing paradigms.