Timothy Hancock

Mining metabolic pathways through gene expression

Motivation: An observed metabolic response is the result of the coordinated activation and interaction between multiple genetic pathways. However, the complex structure of metabolism has meant that a compete understanding of which pathways are required to produce an observed metabolic response is not fully understood. In this article, we propose an approach that can identify the genetic pathways which dictate the response of metabolic network to specific experimental conditions. Results: Our approach is a combination of probabilistic models for pathway ranking, clustering and classification. First, we use a non-parametric pathway extraction method to identify the most highly correlated paths through the metabolic network. We then extract the defining structure within these top-ranked pathways using both Markov clustering and classification algorithms. Furthermore, we define detailed node and edge annotations, which enable us to track each pathway, not only with respect to its genetic dependencies, but also allow for an analysis of the interacting reactions, compounds and KEGG sub-networks. We show that our approach identifies biologically meaningful pathways within two microarray expression datasets using entire KEGG metabolic networks. Availability and implementation: An R package containing a full implementation of our proposed method is currently available from http://www.bic.kyoto-u.ac.jp/pathway/timhancock Contact: timhancock@kuicr.kyoto-u.ac.jp Supplementary information: Supplementary data are available at Bioinformatics online.

A markov classification model for metabolic pathways

BackgroundThis paper considers the problem of identifying pathways through metabolic networks that relate to a specific biological response. Our proposed model, HME3M, first identifies frequently traversed network paths using a Markov mixture model. Then by employing a hierarchical mixture of experts, separate classifiers are built using information specific to each path and combined into an ensemble prediction for the response.ResultsWe compared the performance of HME3M with logistic regression and support vector machines (SVM) for both simulated pathways and on two metabolic networks, glycolysis and the pentose phosphate pathway for Arabidopsis thaliana. We use AltGenExpress microarray data and focus on the pathway differences in the developmental stages and stress responses of Arabidopsis. The results clearly show that HME3M outperformed the comparison methods in the presence of increasing network complexity and pathway noise. Furthermore an analysis of the paths identified by HME3M for each metabolic network confirmed known biological responses of Arabidopsis.ConclusionsThis paper clearly shows HME3M to be an accurate and robust method for classifying metabolic pathways. HME3M is shown to outperform all comparison methods and further is capable of identifying known biologically active pathways within microarray data.

Boosted Network Classifiers for Local Feature Selection

Like all models, network feature selection models require that assumptions be made on the size and structure of the desired features. The most common assumption is sparsity, where only a small section of the entire network is thought to produce a specific phenomenon. The sparsity assumption is enforced through regularized models, such as the lasso. However, assuming sparsity may be inappropriate for many real-world networks, which possess highly correlated modules. In this paper, we illustrate two novel optimization strategies, namely, boosted expectation propagation (BEP) and boosted message passing (BMP), which directly use the network structure to estimate the parameters of a network classifier. BEP and BMP are ensemble methods that seek to optimize classification performance by combining individual models built upon local network features. Neither BEP nor BMP assumes a sparse solution, but instead they seek a weighted average of all network features where the weights are used to emphasize all features that are useful for classification. In this paper, we compare BEP and BMP with network-regularized logistic regression models on simulated and real biological networks. The results show that, where highly correlated network structure exists, assuming sparsity adversely effects the accuracy and feature selection power of the network classifier.

Unsupervised Data Mining: Introduction

This chapter focuses on cluster analysis in the context of unsupervised data mining. Various facets of cluster analysis, including proximities, are discussed in detail. Techniques of determining the natural number of clusters are described. Finally, techniques of assessing cluster accuracy and reproducibility are detailed. Techniques mentioned in this chapter are expanded upon in the following chapters.

Identifying neighborhoods of coordinated gene expression and metabolite profiles.

In this paper we investigate how metabolic network structure affects any coordination between transcript and metabolite profiles. To achieve this goal we conduct two complementary analyses focused on the metabolic response to stress. First, we investigate the general size of any relationship between metabolic network gene expression and metabolite profiles. We find that strongly correlated transcript-metabolite profiles are sustained over surprisingly long network distances away from any target metabolite. Secondly, we employ a novel pathway mining method to investigate the structure of this transcript-metabolite relationship. The objective of this method is to identify a minimum set of metabolites which are the target of significantly correlated gene expression pathways. The results reveal that in general, a global regulation signature targeting a small number of metabolites is responsible for a large scale metabolic response. However, our method also reveals pathway specific effects that can degrade this global regulation signature and complicates the observed coordination between transcript-metabolite profiles.

A Markov classification model for metabolic pathways

The size and complexity of metabolic networks has increased past the point where a researcher is able to intuitively understand all interacting components. Confronted with complexity, biologists must now create models of these networks to identify key relationships of specific interest to their experiments. In this paper focus on the problem of identifying pathways through metabolic networks that relate to a specific biological response. Our proposed model, HME3M, first identifies frequently traversed network paths using a Markov mixture model. Then by employing a hierarchical mixture of experts, separate classifiers are built using information specific to each path and combined into an ensemble classifier the response. We compare the performance of HME3M with logistic regression and support vector machines (SVM) in both simulated and realistic environments. These experiments clearly show HME3M is a highly interpretable model that outperforms common classification methods for large realistic networks and high levels of pathway noise.

Nuclear magnetic resonance metabonomic profiling using tO2PLS

Blood plasma collected from adult fish (black bream, Sparidae) exposed to a dose of 5 mg kg(-1) 17β-estradiol underwent metabonomic profiling using nuclear magnetic resonance (NMR). An extension of the orthogonal 2 projection to latent structure (O2PLS) analysis, tO2PLS, was proposed and utilized to classify changes between the control and experimental metabolic profiles. As a bidirectional modeling tool, O2PLS examines the (variable) commonality between two different data blocks, and extracts the joint correlations as well as the unique variations present within each data block. tO2PLS is a proposed matrix transposition of O2PLS to allow for commonality between experiments (spectral profiles) to be observed, rather than between sample variables. tO2PLS analysis highlighted two potential biomarkers, trimethylamine-N-oxide (TMAO) and choline, that distinguish between control and 17β-estradiol exposed fish. This study presents an alternative way of examining spectroscopic (metabolite) data, providing a method for the visual assessment of similarities and differences between control and experimental spectral features in large data sets.

A Toolbox Approach to Flexible and Efficient Data Mining

This paper describes a flexible and efficient toolbox based on the scripting language Python, capable of handling common tasks in data mining. Using either a relational database or flat files the toolbox gives the user a uniform view of a data collection. Two core features of the toolbox are caching of database queries and parallelism within a collection of independent queries. Our toolbox provides a number of routines for basic data mining tasks on top of which the user can add more functions - mainly domain and data collection dependent - for complex and time consuming data mining tasks.

Identifying pathways of coordinated gene expression.

Methods capable of identifying genetic pathways with coordinated expression signatures are critical to advance our understanding of the functions of biological networks. Currently, the most comprehensive and validated biological networks are metabolic networks. Complete metabolic networks are easily sourced from multiple online databases. These databases reveal metabolic networks to be large, highly complex structures. This complexity is sufficient to hide the specific details on which pathways are interacting to produce an observed network response. In this chapter we will outline a complete framework for identifying the metabolic pathways that relate to an observed phenomenon. To illuminate the functional metabolic pathways, we overlay microarray experiments on top of a complete metabolic network. We then extract the functional components within a metabolic network through a combination of novel pathway ranking, clustering, and classification algorithms. This chapter is designed as a simple tutorial which enables this framework to be applied to any metabolic network and microarray data.

Active pathway identification and classification with probabilistic ensembles.

A popular means of modeling metabolic networks is through identifying frequently observed pathways. However the definition of what constitutes an observation of a pathway and how to evaluate the importance of identified pathways remains unclear. In this paper we investigate different methods for defining an observed pathway and evaluate their performance with pathway classification models. We use three methods for defining an observed pathway; a path in gene over-expression, a path in probable gene over-expression and a path of most accurate classification. The performance of each definition is evaluated with three classification models; a probabilistic pathway classifier - HME3M, logistic regression and SVM. The results show that defining pathways using the probability of gene over-expression creates stable and accurate classifiers. Conversely we also show defining pathways of most accurate classification finds a severely biased pathways that are unrepresentative of underlying microarray data structure.