Beyond Graphs: A New Synthesis

Artiﬁcial neural networks, electronic circuits, and gene networks are some examples of systems that can be modeled as networks, that is, as collections of interconnected nodes. In this paper we introduce the concept of terminal graph (t-graph for short), which improves on the concept of graph as a unifying principle for the representation and computational synthesis and inference of technological and biological networks. We begin by showing how to use the t-graph concept to better understand the working of existing methods for the computational synthesis of networks. Then, we discuss the issue of the “missing methods”, that is, of new computational methods of network synthesis whose existence can be inferred using the perspective provided by the con- cept of t-graph. Finally, we comment on the application of the t-graph perspective to problems of network inference, to the ﬁeld of complex networks, social networks, and to the understanding of biological networks and developmental processes.

Evolutionary reverse engineering of gene networks

Daniel Marbach

The expression of genes is controlled by regulatory networks, which perform fundamental information processing and control mechanisms in a cell. Unraveling and modelling these networks will be indispensable to gain a systems-level understanding of biological organisms and genetically related diseases. In this thesis, we present an evolutionary reverse engineering method, which allows to simultaneously infer both the wirings and nonlinear dynamical models of gene regulatory networks from gene expression data. The proposed method reconstructs gene networks by mimicking the natural evolutionary process that constructed them. This is achieved by modelling both the way in which gene networks are encoded in the biological genome, and the different types of mutations and recombinations that drive their evolution, using an artificial genome called Analog Genetic Encoding (AGE). Since AGE mimics the evolutionary forces and constraints that shape biological gene networks, the reconstruction is naturally guided towards biologically plausible solutions. Consequently, the search space is explored more efficiently, and the networks are recovered more reliably, than with alternative methods. We have confirmed the state-of-the-art performance of AGE both in vivo (on real gene networks) and in silico (on simulated networks). In particular, AGE achieved winning performance in the in vivo gene network inference inference challenge of the 2nd DREAM (Dialogue on Reverse Engineering Assessment and Methods) conference, which consisted in predicting the structure of a synthetic-biology gene network in Saccharomyces cerevisiae from time-series data. In vivo performance assessment of network-inference methods is problematic because it is in general not possible to systematically validate predictions, except for few well-characterized gene networks. Consequently, in silico benchmarks are essential to understand the performance of network-inference methods. We have developed tools to generate biologically plausible in silico gene networks, which allow realistic performance assessment of network-inference methods. In contrast to previous in silico benchmarks, we generate network structures by extracting modules from known gene networks of model organisms, instead of using random graphs. Furthermore, we simulate network dynamics using more realistic kinetic models, which include both mRNA and proteins. We have implemented this framework in an open-source Java tool called GeneNetWeaver (GNW). Using GNW we have generated benchmarks for community-wide challenges of the 3rd and 4th DREAM conference (the DREAM in silico network challenges). Here, we assess the performance of 29 network-inference methods, which have been applied independently by participating teams of the DREAM3 challenge. Performance profiling on individual network motifs reveals that current inference methods are affected, to various degrees, by three types of systematic prediction errors. We find that these errors are induced by inaccurate prior assumptions of prevalent gene-network models. The evolutionary reverse engineering approach, which would have ranked 3rd in this challenge, can be used with a wide range of nonlinear models. It could thus provide the necessary framework for the development of models that better approximate different types of gene regulation, thereby enabling ever more accurate reconstruction of gene networks.

EPFL2009

GeneNetWeaver 3.0: realistic benchmark generation and performance profiling of network inference methods

Dario Floreano, Daniel Marbach, Thomas Schaffter

Numerous methods have been developed for inference of gene regulatory networks from expression data, however, their strengths and weaknesses remain poorly understood. Accurate and systematic evaluation of these methods is hampered by the difficulty of constructing adequate benchmarks and the lack of tools for a differentiated analysis of network predictions on such benchmarks. Here, we present the new version (3.0) of GeneNetWeaver (GNW), an open-source tool for in silico benchmark generation and performance profiling of net-work inference methods. GNW can be launched directly from any web browser and it has an intuitive graphical user interface. Using GNW it is possible to generate biologically plausible in silico gene networks and simulated expression data, which can be used as benchmarks for network inference methods. Realistic network structures are generated by extracting modules from known biological interaction networks. These networks are then endowed with dynamics using a kinetic model of transcription and translation, where transcriptional regulation is modeled using a thermodynamic approach allowing for both independent and synergistic interactions. Finally, these models are used to produce synthetic gene expression data by simulating different biological experiments. Simulations can be done either deterministically or stochastically to model internal noise in the dynamics of the networks, and experimental noise can be added using a model of noise observed in microarrays. Another important feature of GNW is systematic evaluation of the predictions from different inference methods on in silico networks in the benchmark. For a set of network predictions from one or several inference methods, GNW automatically generates a comprehensive report in PDF format. These reports include standard metrics used to assess the accuracy of network inference methods such as precision-recall and receiver operating characteristic (ROC) curves. Furthermore, the reports include network motif analysis, where the performance of inference methods is profiled on local connectivity patterns. The network motif analysis often reveals systematic prediction errors, thereby indicating potential ways of network reconstruction improvements. We are using GNW to provide an annual network inference challenge for the DREAM project. In the past three editions, a total of 91 teams submitted about 900 network predictions to evaluate the performance of their methods on GNW-generated benchmarks.

2010

Genetic Representation of Adaptive Neural Controllers

Peter Dürr

The manual design of adaptive controllers for robotic systems that face unpredictable environmental changes is often challenging. There is thus a growing interest in the development of automatic design tools to assist control engineers. One of the most common approaches in this domain is the evolutionary synthesis of Artificial Neural Networks (ANNs), which allows the automatic design of control systems with little or no human intervention. The performance of evolved neural controllers, however, critically depends on the choice of a suitable genetic representation, i.e., a description of the ANN on which the evolutionary search can operate effectively. In this thesis, I propose an alternative to existing representations for neural controllers based on an extension of Analog Genetic Encoding (AGE), an abstraction of the regulatory networks that control the expression of genes in biological organisms. The proposed approach unites three characteristics that have separately been considered advantageous, but have so far not been aggregated into a single representation: (1) it can be used to effectively synthesize the topology, weights, and numerical parameters of arbitrary networks; (2) it permits the use of heterogeneous ANN models that feature multiple types of signals; and (3) it facilitates the evolution of modular controllers. The results of a series of experiments demonstrate the advantages of these characteristics: (i) neural architectures synthesized with the proposed method display a performance competitive to the best hand-designed networks and networks synthesized with other representations; (ii) in tasks that feature an unpredictable changing environment, controllers with a heterogeneous ANN model outperform controllers with conventional, homogeneous ANN models; and (iii) in tasks that feature multiple modular sub-problems, the proposed representation allows the automatic decomposition of the problem, leading to a substantial improvement in the performance of the evolved controllers. In summary, the proposed representation in combination with a heterogeneous ANN model and an evolutionary algorithm represents a promising method for the automatic synthesis of adaptive control systems for robots.

EPFL2011