Tsutomu Matsunaga

Efficient Algorithms for Finding Maximum and Maximal Cliques: Effective Tools for Bioinformatics

Many problems can be formulated as graphs where a graph consists of a set of vertices and a set of edges, in which the vertices stand for objects in question and the edges stand for some relations among the objects. A clique is a subgraph in which all pairs of vertices are mutually adjacent. Thus, a maximum clique stands for a maximum collection of objects which are mutually related in some specified criterion. The so called maximum clique problem is one of the original 21 problems shown to be NP-complete by R. Karp (19). Therefore, it is strongly believed that the maximum clique problem is not solvable easily, i.e., it is not solvable in polynomial-time. Nevertheless, much work has been done on this problem, experimentally and theoretically. It attracts much attention especially recently since it has found many practical applications to bioinformatics (see, e.g., (2; 15; 27; 28; 37; 3; 9; 4; 8; 14; 55; 23; 25; 22; 13)) and many others (see, e.g., excellent surveys (34; 5), and (17; 20; 31; 49; 54; 51)). This chapter presents efficient algorithms for finding a maximum clique and maximal cliques as effective tools for bioinformatics, and shows our successful applications of these algorithms to bioinformatics.

An experimental study of learning curves for statistical pattern classifiers

Statistical pattern classifiers are designed by population parameters of pattern distributions estimated by a set of training samples. Therefore, classification performance depends considerably on training sample size. Learning curves exhibit asymptotic behaviors where a probability of misclassification decreases as a number of training samples increases. This paper presents asymptotic behaviors of effects of training sample size and shows that learning curves for practical purpose can be obtained using available samples.

A method for designing dictionary using simulated annealing

A dictionary design problem is treated as an optimization problem and a method for designing a dictionary by applying simulated annealing, which was introduced as a technique for solving combinatorial optimization problems, is presented. This method is applicable to various fields in pattern recognition which use template matching. It requires no assumption of pattern distributions. A high recognition rate can be achieved with a single-template dictionary, i.e. a single reference pattern per class. The authors consider handprinted Japanese kanji character recognition and show the effectiveness of the method through experimental results.<<ETX>>

DISEASE-RELATED CONCEPT MINING BY KNOWLEDGE-BASED TWO-DIMENSIONAL GENE MAPPING

There is a strong need to systematically organize and comprehend the rapidly expanding stores of biomedical knowledge to formulate hypotheses on disease mechanisms. However, no method is available that automatically structuralizes fragmentary knowledge along with domain-specific expressions for a large-scale integration. A method presented here, cross-subspace analysis (CSA), produces a holistic view of over 3,000 human genes with a two-dimensional (2D) arrangement. The genes are plotted in relation to functions determined by machine learning from the occurrence patterns of various biomedical terms in MEDLINE abstracts. By focusing on the 2D distributions of gene plots that share the same biomedical concepts, as defined by databases such as Gene Ontology, relevant biomedical concepts can be computationally extracted. In an analysis where myocardial infarction and ischemic stroke were taken as examples, we found valid relations with lifestyle, diet-related metabolism, and host immune responses, all of which are known risk factors for the diseases. These results demonstrate that systematizing accumulated gene knowledge can lead to hypothesis generation and knowledge discovery, regardless of the area of inquiry or discipline.

A study of document format identification based on table structure

A method to identify formatted documents through table structure is described. This method extracts connected components of white pixels in documents for identification, using a subspace classification method. The effectiveness of this method is discussed. The authors focus on the discriminant functions of this classification and consider the applicability of the method for practical use.<<ETX>>