We develop efficient and flexible algorithms for bioinformatics. We also study mathematical aspects of systems biology. The research topics include inference and analysis of various types of biological networks, prediction and analysis of protein/RNA structures, statistical models for sequence analysis, and scale-free networks.
Mathematical Analysis of Structures of Biological Networks
Scale-free properties of biological networks have been
extensively studied. We study how the topology of the scale-free networks is
changed by the line graph transformation, where the line graph transformation
relates two representations of a metabolic network: compound network and reaction
network. The main result is that the degree distribution follows a power law
P(
k)∝
k-γ+1 after the line graph transformation if the original network follows a power law
P(
k)∝
k-γ. We
also examined the degree distributions of compound networks and reaction networks
using the KEGG database. The results suggest that a similar property holds for
these networks.
Fig. 1. Relation between
Two Representations of a Metabolic Network
Support vector machines and kernel methods have been applied to various classification problems in bioinformatics. In order to
apply kernel methods to bioinformatics problems, it is usually required to develop a kernel function which provides a kind of
measure of similarity between two objects. Recently, we developed two kernels: the local alignment kernel for protein sequences,
and an extended marginalized graph kernel for chemical compounds. The results of computational experiments on chemical compounds
suggest that our method is much faster than the original marginalized graph kernel while keeping classification accuracy.
Fig. 2. Classification
of Chemical Compounds Using Support Vector Machines