- 著者
-
TUANNAMTRAN
TUBAOHO
- 出版者
- 一般社団法人情報処理学会
- 雑誌
- 情報処理学会研究報告知能と複雑系(ICS) (ISSN:09196072)
- 巻号頁・発行日
- vol.2004, no.125, pp.213-218, 2004-12-07
- 参考文献数
- 30
Inductive Logic Programming (ILP) is differentiated from most supervised learning methods both by its use of an expressive representation language and its ability to make use of background knowledge. This has led to successful applications of ILP in molecular biology such as predicting the mutagenicity of chemical compounds predicting protein secondary structures and discovering protein fold descriptions. In this paper we attempt to apply ILP to the problem of predicting protein-protein interactions which plays an essential role in bioinformatics since many major biological processes are controlled by protein interaction networks. We have used the Yeast Interacting Proteins Database provided by Ito Tokyo University as training examples. Various kinds of background knowledge have been constructed by either extracting from protein databases or using computational approaches. Early results indicate that ILP is useful for obtaining comprehensible rules to differentiate those protein-protein interactions that are highly reliable. The predictive accuracy obtained using ten-fold cross-validation is nearly 80% demonstrating a promising result of using ILP for predicting protein-protein interactions.Inductive Logic Programming (ILP) is differentiated from most supervised learning methods both by its use of an expressive representation language and its ability to make use of background knowledge. This has led to successful applications of ILP in molecular biology, such as predicting the mutagenicity of chemical compounds, predicting protein secondary structures, and discovering protein fold descriptions. In this paper, we attempt to apply ILP to the problem of predicting protein-protein interactions, which plays an essential role in bioinformatics since many major biological processes are controlled by protein interaction networks. We have used the Yeast Interacting Proteins Database provided by Ito, Tokyo University as training examples. Various kinds of background knowledge have been constructed by either extracting from protein databases or using computational approaches. Early results indicate that ILP is useful for obtaining comprehensible rules to differentiate those protein-protein interactions that are highly reliable. The predictive accuracy obtained using ten-fold cross-validation is nearly 80%, demonstrating a promising result of using ILP for predicting protein-protein interactions.