380
Genome Informatics 13: 380–381 (2002)
Comparison between Two Different Genetic Network Inferring Models on Expression Profiles in S. cerevisiae Yoriko Takahashi1
Yuji Arikawa1
Shoji Watanabe1
[email protected]
[email protected]
[email protected]
Yukihiro
Maki2
[email protected]
Sachiyo
Aburatani3
[email protected]
Yukihiro
Satoru Kuhara3
[email protected]
Eguchi1
[email protected] 1
2
3
Mitsui Knowledge Industry Co., Ltd, Harmony tower 21th Floor, 1-32-2 Honcho, Nakanoku, Tokyo 164-8721, Japan Laboratory for Applied Biological Regulation Technology, Graduate School of Bioresource and Bioenvironmental Sciences, Kyusyu University, Hakozaki 6-10-1, Higashiku, Fukuoka 812-8581, Japan Laboratory for Molecular Gene Technics, Graduate School of Genetic Resources Technology, Kyusyu University, Hakozaki 6-10-1, Higashi-ku, Fukuoka 812-8581, Japan
Keywords: AIGNET, genetic network, Bayesian model, Threshold-Test model, DNA microarray
1
Introduction
The development of DNA microarray technology makes it possible to measure the expression levels of thousands of genes simultaneously, and the constructions of the genetic networks from the large-scale data have been studied extensively. However, any inferring method of the genetic network, which is able to elucidate the relations among genes comprehensively, has never established yet, because of the experiment error in the measurement data of this technique. We have proposed the integrated comprehensive network inferring system, named AIGNET (Algorithms for Inference of Genetic Networks) [1, 2]. It has four completely different network models, those are Threshold-Test model [1], Bayesian model [1], S-system model [1, 2, 3], and multi-level digraph model [1, 2, 3]. That is, it is possible to construct the high accuracy genetic network which cannot be derived from one model by combining of models appropriately. Or according to the type of experiment data or the purpose of analysis, Researchers can choose the suitable model according to the experimental type or purpose of analysis. Here, we present the effectiveness of combination of the different models comparing two genetic networks derived by Threshold-Test model and Bayesian model respectively.
2
Method and Results
Profile Data In this study, we used the 175 gene expression profile data which 5871 yeast genes of 156 disruptants were measured (19 disruptants were measured twice). It is assumed that logarithmic value of expression ratios of almost 6000 genes in one experiment data follow a normal distribution. In order to analyze two or more experiment data simultaneously, all data were standardized so that they could fit a standard normal distribution.
Comparison between Two Different Genetic Network Inferring Models
381
Network Analysis For network analysis, we selected 605 genes that were identified on Yeast Proteome Database [4] and 156 disrupted genes. The data set was analyzed by Threshold-Test model and Bayesian model with various threshold values respectively. The number of common binary relations between two models is shown in Table 1. Its number was the maximum with threshold value = 2.5 on both models. Then, the comparing of two models was carried out in this condition. Here, the relationships controlled by LEU3 are shown in Table 2. The 51 relationships were inferred by Bayesian model. On the other hand, the 40 ones were inferred by Threshold-Test model. There were 22 common relationships between these models, including the 2 known-relationships on YPD. The genes controlled by LEU3 were classified into the following genetic functional groups; hydrocarbon metabolism, nucleotide metabolism and mRNA transcription, in addition to amino acid metabolism, to which all genes of known relationships belong. Moreover, these groups were found out of both models. Table 1: The number of common binary relations derived from Bayesian model and Threshold-Test model.
Threshold Value of Threshold-Test model
Threshold Value of Bayesian model
2
2.5
3
2
531
338
232
2.5
755
763
480
3
655
650
637
Table 2: The number of relationship controlled by LEU3 derived from Bayesian model and ThresholdTest model with threshold value = 2.5. Bayesian model Threshold-Test model Common Known-relationship on YPD (7 relations) Inferred relationship by each model
3
2 51
2 40
2 22
Discussion
The result of comparison of networks based on the Bayesian model and Threshold-Test model using the same data showed that strong candidates of new pathway could be inferred efficiently from data with experimental error. We are going to establish the way of combination of these models, and improve our system more user-friendly.
References [1] Arikawa, Y., Takahashi, Y., Watanabe, S., Maki, Y., Okamoto, M., and Eguchi, Y., Inference of a gene network from the experimentally observed expression data by using AIGNET, Genome Informatics, 12:274–275, 2001. [2] Arikawa, Y., Watanabe, S., Maki, Y., Tominaga, D., and Okamoto, M., AIGNET: Improvement of a system and application to the experimentally observed expression data, Genome Informatics, 11:291–292, 2000. [3] Maki, Y., Tominaga, D., Okamoto, M., Watanabe, S., and Eguchi, Y., Development of a system for the inference of large scale genetic networks, Proc. Pacific Symposium on Biocomputing, 446– 458, 2001. [4] http://www.incyte.com/
