Inference of Common Genetic Network Using Fuzzy Adaptive Resonance Theory Associated Matrix Method

Abstract

Inferring genetic networks from gene expression data is the most challenging work in the post-genomic era. However, most studies tend to show their genetic network inference ability by using artificial data. Here, we developed the fuzzy adaptive resonance theory associated matrix (F-ART matrix) method to infer genetic networks and applied it to experimental time series data, which are gene expression profiles of Saccharomyces cerevisiae responding under oxidative stresses such as diamide, heat shock and H2O2. We preprocessed them using the fuzzy adaptive resonance theory and successfully identified genetic interactions by drawing a 2-dimensional matrix. The identified interactions between diamide and heat shock stress were confirmed to be the common interactions for two stresses, compared with the KEGG metabolic map, BRITE protein interaction map, and gene interaction data of other papers. In the predicted common genetic network, the hit ratio was 60% for the KEGG map. Several gene interactions were also drawn, which have been reported to be important in oxidative stress. This result suggests that F-ART matrix has the potential to function as a new method to extract the common genetic networks of two different stresses using experimental time series microarray data.

Computational models reconstruct gene regulatory networks

“…Over the last few years, various approaches for GRN reconstruction have been proposed in the literature. [4][5][6][7] These approaches range from linear models [8][9][10] and differential equations [11][12][13] to Boolean 14,15 and Bayesian models 16,17 and also, neural networks, 18 fuzzy methods 19 and neuro fuzzy networks. 20,21 The current study is not intended to be a thorough review on the subject [22][23][24] but rather to present some of the most promising computational models that have been adapted to solve this complex problem.…”

A set of new amino acid descriptors applied in prediction of MHC class I binding peptides

“…Besides, identification of various affinity motifs [10], thousands of specific allelic genes, promiscuous MHC binders and T-cell epitopes [11] provides a great deal of abundant information for utilizing computers to predict the MHC-affinity peptides by computational means. Several approaches such as simple/linear motifs [12], quantitative matrices [13], artificial neural networks [14], fuzzy neural networks [15], support vector machines [16], fuzzy classifier with the SWEEP operator method [17], and Hidden Markov Models [18,19] have already been used to predict peptidee MHC binding. Molecular dynamics simulations [20] and homology modeling [21] have also been applied to investigate MHCepeptide interactions.…”