CONICET | Buscador de Institutos y Recursos Humanos

Background: Gene regulatory networks (GRNs) play an important role in the progression of life phenomena such as cell cycling, developmental biology, aging, and the progressive and recurrent pathogenesis of complex diseases, among others. The amount of gene expression time series data is becoming increasingly available, providing the opportunity to reverse engineer the time-delayed gene regulatory networks that govern the majority of these molecular processes. In this context, data mining methods constitute suitable approaches for performing the inference of the relational structures of a GRN [1].Method: The aim of the research presented here consists on the enhance of GRN based on association rules from multiple microarray time series datasets given as input. In this regard, a rule-based inference algorithm (GRNCOP2) [2] was combined with a biclustering technique (BiHEA) [3] in order to increase the useful information extracted from the datasets. The association rules establish causal links between two genes, where the semantics and the interpretation depend of the input data and on the rule type inferred. This provides a global view of the relation between each pair of genes since it considers all the data available on the expression profiles. On the other hand, the biclustering algorithm can be used to extract co-expression (similar or opposed) relations between genes that may only occur in a subset of the experimental conditions, extracting additional associations with a local view of the data that may not be captured by the main inference algorithm. In order to combine both methods, a pair-wise analysis is performed to extract association rules from the biclusters obtained from all the datasets, adding the best rules to the GRN inferred by the ruled based method. The proposed approach was applied to time series datasets [4, 5] composed of twenty yeast genes that are highly relevant for the cell-cycle study, and the results were analyzed in terms of the novelty and soundness of the rules provided by the biclustering algorithm. In order to assess the soundness of the rules, the average accuracy for the rules was measured regarding a freely available database of associations between yeast genes known as Yeastnet [6]. The Figure 1 shows the average accuracy for the rules obtained by the ruled based approach, the biclustering algorithm and the combination of both methods. It also shows the expected accuracy if the rules were picked randomly. The Figure 2 shows the network obtained by the rule-based approach alone and the same network enhanced by the rules obtained through the biclustering algorithm. As it can be observed, the set of rules inferred by the two algorithms and the combined results achieve high accuracy values regarding the Yeastnet benchmark database, performing above the random selection as expected. Although the rules inferred by the biclustering algorithm are less accurate than those extracted by the rule-based approach, these rules represent new potential relations that were not discovered by the main inference algorithm, thus enhancing the overall inference capabilities.Conclusions: In this work, we have introduced an approach to integrate the results of a rule-based method with a biclustering algorithm for the inference of gene regulatory networks. The method was validated with well known publicly available gene expression datasets. The results have shown that the combined approach infers a gene regulatory network with high average accuracy regarding the Yeasnet database, providing new relations that were not present in the GRN inferred by the rule-based method alone. This shows the importance of combining different approaches in the inference of gene regulatory network, since it provides alternative views of the data and allows the discovery of significant relations that may no be detectable by an specific approach. Further analysis is required in order to confirm these promissory results.

enviar mensaje