Séminaire externe : Hiroshi Mamitsuka
Séance du mardi 30 septembre 2014, 11:30 - 12:45,
Salle de conférences du CERLA
Mining from biological networks: preserving network localities
Hiroshi Mamitsuka
Bioinformatics Center, Institute for Chemical Research, Kyoto University
Biological networks are new data, being generated by recently developed high-throughput techniques or accumulated knowledge in biology. Typical information to build biological networks are protein-protein interaction, gene regulation, metabolic reaction, signal transaction and sequence similarity, etc., all resulting in gene networks, in which nodes are uniquely labeled by genes. In this talk, focusing on annotating functions of genes, a fundamental problem in modern molecular biology, I will raise two possible data mining approaches, clustering and label propagation, over multiple biological networks. In particular I will introduce the original techniques developed in my group, which can treat an intrinsic problem of multiple biological networks, i.e. the locality of substructures. That is, clusters can be obtained by only part of given multiple networks, and informative networks/subnetworks can be found only in part of given networks. They are local substructures to be captured correctly in mining from multiple biological networks. I will explain the key idea/concept behind these techniques and how significantly they can improve the predictive performance of annotating gene functions. Finally I will discuss the problems to be considered in these techniques to build new methods further.