Spencer Angus Thomas

Reconstructing biological gene regulatory networks: where optimization meets big data

The importance of ‘big data’ in biology is increasing as vast quantities of data are being produced from high-throughput experiments. Techniques such as DNA microarrays are providing a genome-wide picture of gene expression levels, allowing us to investigate the structure and interactions of gene networks in biological systems. Inference of gene regulatory network (GRN) is an underdetermined problem suited to Metaheuristic algorithms which can operate on limited information. Thus GRN inference offers a platform for investigations into data intensive sciences and large scale optimization problems. Here we examine the link between data intensive research and optimization problems for the reverse engineering of GRNs. Briefly, we detail the benefit of the data deluge and the study of ALife for modelling GRNs as well as their reconstruction. We discuss how metaheuristics can solve big data problems and the inference of GRNs offer real world problems for both areas of research. We overview some current reconstruction algorithms and investigate some modelling and computational limits of the inference processes and suggest some areas for development. Furthermore we identify links and synergies between optimization and big data, e.g., dynamic, uncertain and large scale optimization problems, and discuss the potential benefit of multi- and many-objective optimization. We stress the importance of data integration techniques in order to maximize the data available, particularly for the case of inferring GRNs from microarray data. Such multi-disciplinary research is vital as biology is rapidly becoming a quantitative, data intensive science.

Combining genetic oscillators and switches using evolutionary algorithms

It is hypothesised that complex biological gene regulatory networks can be evolved from simple networks through modularisation, duplication and specialisation processes. However, the biological mechanisms of this process remain elusive and little work has been done to verify this hypothesis in a computational environment. This paper aims to couple two simple regulatory motifs, one toggle switch and one self-sustained oscillator using an evolutionary algorithm, which can be seen as a computational simulation of natural evolution. We have successfully evolved several complex dynamics for two different connections arrangements between the oscillator and toggle switch networks in a master/slave set up, which confirms the previously reported results achieved manually. Our results indicate that generating complex dynamics by coupling of simple motifs using simulated evolutionary mechanisms is methodologically feasible and more efficient, which can be seen as an indirect and partial verification of the above hypothesis.

Evolving connectivity between genetic oscillators and switches using evolutionary algorithms.

Although hypothesised there has been little investigation into how complex gene regulatory networks can evolve from simple regulatory motifs through modularisation, duplication and specialisation processes. In order to simulate natural evolution in a computational environment we evolve the connection between a genetic oscillator and a toggle switch motif using an evolutionary algorithm. We observe a connectivity preference between the motifs that is dependent on the coupling arrangement rather than on objective set-up. In addition, our results indicate the existence of a threshold in the connection parameters for the resulting dynamics for a specific coupling arrangement and objective set-up. We demonstrate that simple motifs can successfully be coupled through artificial evolution to form more complex, modular regulatory networks. These findings support, in principle, the above-mentioned hypothesis on evolutionary mechanisms in biological systems.

Single and Multi-objective in Silico Evolution of Tunable Genetic Oscillators

We compare the ability of single and multi-objective evolutionary algorithms to evolve tunable self-sustained genetic oscillators. Our research is focused on the influence of objective setup on the success rate of evolving self-sustained oscillations and the tunability of the evolved oscillators. We compare temporal and frequency domain fitness functions for single and multi-objective evolution of the parameters in a three-gene genetic regulatory network. We observe that multiobjectivization can hinder convergence when decomposing a period specific based single objective setup in to a multi-objective setup that includes a frequency specific objective. We also find that the objective decomposition from a frequency specified single objective setup to a multi-objective setup, which also specifies period, enable the synthesis of oscillatory dynamics. However this does not help to enhance tunability. We reveal that the use of a helper function in the frequency domain improves the tunability of the oscillators, compared to a time domain based single objective, even if no desired frequency is specified.

Dimensionality reduction of mass spectrometry imaging data using autoencoders

The use of mass spectrometry imaging (MSI) techniques has become a powerful tool in the fields of biology, pharmacology and healthcare. Next generation experimental techniques are able to generate 100s of gigabytes of data from a single image acquisition and thus require advanced algorithms in order to analyse these data. At present, analytical work-flows begin with pre-processing of the data to reduce its size. However, the pre-processed data is also high in dimensionality and requires reduction techniques in order to analyse the data. At present, mostly linear dimensionality reduction techniques are used for hyper-spectral data. Here we successfully apply an autoencoder to MSI data with over 165,000 pixels and more than 7,000 spectral channels reducing it into a few core features. Our unsupervised method provides the MSI community with an effective non-linear dimensionality reduction technique which includes the mapping to and from the reduced dimensional space. This method has added benefits over methods such as PCA by removing the need to select meaningful features from the entire list of components, reducing subjectivity and significant human interaction from the analysis.

Reconstructing regulatory networks in Streptomyces using evolutionary algorithms

Reconstructing biological networks is vital in developing our understanding of nature. Biological systems of particular interest are bacteria that can produce antibiotics during their life cycle. Such an organism is the soil dwelling bacterium Streptomyces coelicolor. Although some of the genes involved in the production of antibiotics in the bacterium have been identified, how these genes are regulated and their specific role in antibiotic production is unknown. By understanding the network structure and gene regulation involved it may be possible to improve the production of antibiotics from this bacterium. Here we use an evolutionary algorithm to optimise parameters in the gene regulatory network of a sub-set of genes in S. coelicolor involved in antibiotic production. We present some of our preliminary results based on real gene expression data for continuous and discrete modelling techniques.

Study of the hindrance effect in sub-barrier fusion reactions

Abstract We have measured the fusion cross sections of the 12C(13C, p)24Na reaction through off-line measurement of the β decay of 24Na using the β-γ coincidence method. Our new measurements in the energy range of E c . m . = 2.6 – 3.0 MeV do not show an obvious S-factor maximum but a plateau. Comparison between this work and various models is presented.

Recent results from the carbon fusion project at Notre Dame

The carbon fusion project at Notre Dame is aimed towards measuring the 12C+12C fusion cross section and its decay branches relevant to astrophysics down to the lowest possible energies. To complement this approach, we are also exploring new techniques for providing more reliable extrapolations of the cross sections in the energy ranges where experimental data are unavailable. In this paper, we report two recent results: 1) an upper limit for the 12C+12C fusion cross section, and 2) a new measurement of 12C(12C,n) along with an improved extrapolation technique based on the mirror reaction channel, 12C(12C,p). The outlook for astrophysical heavy-ion fusion studies at Notre Dame is also discussed.The carbon fusion project at Notre Dame is aimed towards measuring the 12C+12C fusion cross section and its decay branches relevant to astrophysics down to the lowest possible energies. To complement this approach, we are also exploring new techniques for providing more reliable extrapolations of the cross sections in the energy ranges where experimental data are unavailable. In this paper, we report two recent results: 1) an upper limit for the 12C+12C fusion cross section, and 2) a new measurement of 12C(12C,n) along with an improved extrapolation technique based on the mirror reaction channel, 12C(12C,p). The outlook for astrophysical heavy-ion fusion studies at Notre Dame is also discussed.