IRIS: a method for reverse engineering of regulatory relations in gene networks

Discrete Logic Modelling Optimization to Contextualize Prior Knowledge Networks Using PRUNET

High-throughput technologies have led to the generation of an increasing amount of data in different areas of biology. Datasets capturing the cell's response to its intra- and extra-cellular microenvironment allows such data to be incorporated as signed and directed graphs or influence networks. These prior knowledge networks (PKNs) represent our current knowledge of the causality of cellular signal transduction. New signalling data is often examined and interpreted in conjunction with PKNs. However, different biological contexts, such as cell type or disease states, may have distinct variants of signalling pathways, resulting in the misinterpretation of new data. The identification of inconsistencies between measured data and signalling topologies, as well as the training of PKNs using context specific datasets (PKN contextualization), are necessary conditions to construct reliable, predictive models, which are current challenges in the systems biology of cell signalling. Here we present PRUNET, a user-friendly software tool designed to address the contextualization of a PKNs to specific experimental conditions. As the input, the algorithm takes a PKN and the expression profile of two given stable steady states or cellular phenotypes. The PKN is iteratively pruned using an evolutionary algorithm to perform an optimization process. This optimization rests in a match between predicted attractors in a discrete logic model (Boolean) and a Booleanized representation of the phenotypes, within a population of alternative subnetworks that evolves iteratively. We validated the algorithm applying PRUNET to four biological examples and using the resulting contextualized networks to predict missing expression values and to simulate well-characterized perturbations. PRUNET constitutes a tool for the automatic curation of a PKN to make it suitable for describing biological processes under particular experimental conditions. The general applicability of the implemented algorithm makes PRUNET suitable for a variety of biological processes, for instance cellular reprogramming or transitions between healthy and disease states.

De novo reconstruction of gene regulatory networks from time series data, an approach based on formal methods

Reverse engineering of gene regulatory relationships from genomics data is a crucial task to dissect the complex underlying regulatory mechanism occurring in a cell. From a computational point of view the reconstruction of gene regulatory networks is an undetermined problem as the large number of possible solutions is typically high in contrast to the number of available independent data points. Many possible solutions can fit the available data, explaining the data equally well, but only one of them can be the biologically true solution. Several strategies have been proposed in literature to reduce the search space and/or extend the amount of independent information. In this paper we propose a novel algorithm based on formal methods, mathematically rigorous techniques widely adopted in engineering to specify and verify complex software and hardware systems. Starting with a formal specification of gene regulatory hypotheses we are able to mathematically prove whether a time course experiment belongs or not to the formal specification, determining in fact whether a gene regulation exists or not. The method is able to detect both direction and sign (inhibition/activation) of regulations whereas most of literature methods are limited to undirected and/or unsigned relationships. We empirically evaluated the approach on experimental and synthetic datasets in terms of precision and recall. In most cases we observed high levels of accuracy outperforming the current state of art, despite the computational cost increases exponentially with the size of the network. We made available the tool implementing the algorithm at the following url: http://www.bioinformatics.unisannio.it.

Discretization provides a conceptually simple tool to build expression networks

Biomarker identification, using network methods, depends on finding regular co-expression patterns; the overall connectivity is of greater importance than any single relationship. A second requirement is a simple algorithm for ranking patients on how relevant a gene-set is. For both of these requirements discretized data helps to first identify gene cliques, and then to stratify patients.

We explore a biologically intuitive discretization technique which codes genes as up- or down-regulated, with values close to the mean set as unchanged; this allows a richer description of relationships between genes than can be achieved by positive and negative correlation. We find a close agreement between our results and the template gene-interactions used to build synthetic microarray-like data by SynTReN, which synthesizes “microarray” data using known relationships which are successfully identified by our method.

We are able to split positive co-regulation into up-together and down-together and negative co-regulation is considered as directed up-down relationships. In some cases these exist in only one direction, with real data, but not with the synthetic data. We illustrate our approach using two studies on white blood cells and derived immortalized cell lines and compare the approach with standard correlation-based computations. No attempt is made to distinguish possible causal links as the search for biomarkers would be crippled by losing highly significant co-expression relationships. This contrasts with approaches like ARACNE and IRIS.

The method is illustrated with an analysis of gene-expression for energy metabolism pathways. For each discovered relationship we are able to identify the samples on which this is based in the discretized sample-gene matrix, along with a simplified view of the patterns of gene expression; this helps to dissect the gene-sample relevant to a research topic - identifying sets of co-regulated and anti-regulated genes and the samples or patients in which this relationship occurs.

Learning gene regulatory networks from only positive and unlabeled data

Background

Recently, supervised learning methods have been exploited to reconstruct gene regulatory networks from gene expression data. The reconstruction of a network is modeled as a binary classification problem for each pair of genes. A statistical classifier is trained to recognize the relationships between the activation profiles of gene pairs. This approach has been proven to outperform previous unsupervised methods. However, the supervised approach raises open questions. In particular, although known regulatory connections can safely be assumed to be positive training examples, obtaining negative examples is not straightforward, because definite knowledge is typically not available that a given pair of genes do not interact.

Results

A recent advance in research on data mining is a method capable of learning a classifier from only positive and unlabeled examples, that does not need labeled negative examples. Applied to the reconstruction of gene regulatory networks, we show that this method significantly outperforms the current state of the art of machine learning methods. We assess the new method using both simulated and experimental data, and obtain major performance improvement.

Conclusions

Compared to unsupervised methods for gene network inference, supervised methods are potentially more accurate, but for training they need a complete set of known regulatory connections. A supervised method that can be trained using only positive and unlabeled data, as presented in this paper, is especially beneficial for the task of inferring gene regulatory networks, because only an incomplete set of known regulatory connections is available in public databases such as RegulonDB, TRRD, KEGG, Transfac, and IPA.