Jem J. Rowland

A functional genomics strategy that uses metabolome data to reveal the phenotype of silent mutations

A large proportion of the 6,000 genes present in the genome of Saccharomyces cerevisiae, and of those sequenced in other organisms, encode proteins of unknown function. Many of these genes are “silent,” that is, they show no overt phenotype, in terms of growth rate or other fluxes, when they are deleted from the genome. We demonstrate how the intracellular concentrations of metabolites can reveal phenotypes for proteins active in metabolic regulation. Quantification of the change of several metabolite concentrations relative to the concentration change of one selected metabolite can reveal the site of action, in the metabolic network, of a silent gene. In the same way, comprehensive analyses of metabolite concentrations in mutants, providing “metabolic snapshots,” can reveal functions when snapshots from strains deleted for unstudied genes are compared to those deleted for known genes. This approach to functional analysis, using comparative metabolomics, we call FANCY—an abbreviation for functional analysis by co-responses in yeast.

Genetic algorithms as a method for variable selection in multiple linear regression and partial least squares regression, with applications to pyrolysis mass spectrometry

Abstract Four optimising methods for variable selection in multivariate calibration have been described: one for determining the optimal subset of variables to give the best possible root-mean-square error of prediction (RMSEP) in a multiple linear regression (MLR) model, the second for determining the optimal subset of variables which produce a model with RMSEP less than or equal to a given value. Algorithms three and four were identical to algorithms one and two, respectively, except that this time they use a cost function derived from a partial least squares (PLS) model rather than an MLR model. Applied to a typical set of pyrolysis mass spectrometry data the first variable selection method is shown to reduce the RMSEP of the optimal MLR or PLS model significantly when the number of variables is decreased by approximately half. Alternatively, the number of variables may be reduced substantially (> 10-fold) with no loss in RMSEP.

Primary and secondary metabolism, and post‐translational protein modifications, as portrayed by proteomic analysis of Streptomyces coelicolor

The newly sequenced genome of Streptomyces coelicolor is estimated to encode 7825 theoretical proteins. We have mapped approximately 10% of the theoretical proteome experimentally using two‐dimensional gel electrophoresis and matrix‐assisted laser desorption ionization time‐of‐flight (MALDI‐TOF) mass spectrometry. Products from 770 different genes were identified, and the types of proteins represented are discussed in terms of their anno‐tated functional classes. An average of 1.2 proteins per gene was observed, indicating extensive post‐translational regulation. Examples of modification by N‐acetylation, adenylylation and proteolytic processing were characterized using mass spectrometry. Proteins from both primary and certain secondary metabolic pathways are strongly represented on the map, and a number of these enzymes were identified at more than one two‐dimensional gel location. Post‐translational modification mechanisms may therefore play a significant role in the regulation of these pathways. Unexpectedly, one of the enzymes for synthesis of the actinorhodin polyketide antibiotic appears to be located outside the cytoplasmic compartment, within the cell wall matrix. Of 20 gene clusters encoding enzymes characteristic of secondary metabolism, eight are represented on the proteome map, including three that specify the production of novel metabolites. This information will be valuable in the characterization of the new metabolites.

Wavelet Denoising of Infrared Spectra

The application of wavelet denoising to infrared spectra was investigated. Six different wavelet denoising methods (SURE, VISU, HYBRID, MINMAX, MAD and WAVELET PACKETS) were applied to pure infrared spectra with various added levels of homo- and heteroscedastic noise. The performances of the wavelet denoising methods were compared with the standard Fourier and moving mean filtering in terms of root mean square errors between the pure and denoised spectra and visual quality of the denoised spectrum. The use of predictive ability as a possible objective criterion for denoising performance was also investigated. The main conclusion is that for very low signal-to-noise ratios (S/N) the standard denoising methods (Fourier and moving mean) are comparable to the more sophisticated methods. At higher S/N levels the wavelet denoising methods, in particular the HYBRID and VISU methods, are better. Wavelet methods are also better in restoring the visual quality of the denoised infrared spectra.

Variable selection in wavelet regression models

Variable selection and compression are often used to produce more parsimonious regression models. But when they are applied directly to the original spectrum domain, it is not easy to determine the type of feature the selected variables represent. By performing variable selection in the wavelet domain we show that it is possible to identify important variables as being part of short- or large-scale features. Therefore, the suggested method is to extract information about the selected variables that otherwise would have been inaccessible. We are also able to obtain information about the location of these features in the original domain. In this article we demonstrate three types of variable selection methods applied to the wavelet domain: selection of optimal combination of scales, thresholding based on mutual information and truncation of weight vectors in the partial least squares (PLS) regression algorithm. We found that truncation of weight vectors in PLS was the most effective method for selecting variables. For the two experimental data sets tested we obtained approximately the same prediction error using less than 1% (for Data set 1) and 10% (for Data set 2) of the original variables. We also discovered that the selected variables were restricted to a limited number of wavelet scales. This information can be used to suggest whether the underlying features may be dominated by narrow (selective) peaks (indicated by variables in short wavelet scale regions) or by broader regions (indicated by variables in long wavelet scale regions). Thus, wavelet regression is here used as an extension of the more traditional Fourier regression (where the modelling is performed in the frequency domain without taking into consideration any of the information in the time domain).

Discrimination of the variety and region of origin of extra virgin olive oils using 13C NMR and multivariate calibration with variable reduction

Copyright (c) 1997 Elsevier Science B.V. All rights reserved. There is strong evidence that consumption of olive oil, especially extra virgin olive oil, reduces the risk of circulatory system diseases. Such oil is generally more expensive than other edible oils, Italian - and in particular Tuscan - oils being particularly favoured by connoisseurs, and commanding an even higher price. There is therefore a great temptation to adulterate olive oil with a cheaper oil, or falsify its origin or grade. An easy and reliable method to identify different types of olive oil is required. Our work has focused on discriminating extra virgin olive oils by their region and variety. We have applied Principal Components Analysis (PCA), Principal Components Regression (PCR) and Partial Least Squares (PLS) to discriminate olive oils on the basis of their 13 C NMR spectra. Variable Selection was used in order to reduce the number of variables in the data. Two main methods of variable selection have been used; these are the Fisher Ratio, and the ratio of Inner Variance to Outer Variance or Characteristicity [W. Eshuis, P.G. Kistemaker and H.L.C. Meuzelaar, in C.E.R. Jones and C.A. Cramers (Eds.), Analytical Pyrolysis, Elsevier, Amsterdam, 1977, pp. 151-156.]. Both these methods proved successful in improving the PCA clustering, and the prediction results of PCR and PLS, although the optimal number of variables varied between datasets. PCR2 and PLS2 models, in which a single model is used to predict each variety or each region simultaneously, achieved a successful prediction rate of some 70%. However, multiple PLS1 models routinely achieved successful predictions of over 90% and in many cases 100% of the data in test sets. Indeed the variety of all but 1 of 66 samples was correctly predicted. It is clear that multiple, specialised models perform much better than global ones, and that the inclusion of certain variables can be highly detrimental to the multivariate calibration process.

Diffuse reflectance absorbance spectroscopy taking in chemometrics (DRASTIC). A hyperspectral FT-IR-based approach to rapid screening for metabolite overproduction

We introduce diffuse-reflectance absorbance spectroscopy in the mid-infrared as a novel method of chemical imaging for the rapid screening of biological samples for metabolite overproduction, using mixtures of ampicillin with Escherichia coli and Staphylococcus aureus as model systems. Deconvolution of the hyperspectral information provided by the raw diffuse reflectance-absorbance mid-infrared spectra was achieved using a combination of principal components analysis (PCA), artificial neural networks (ANNs) and partial least squares regression (PLS). Whereas a univariate approach necessitates appropriate data selection to remove any interferences, the chemometrics/hyperspectral approach could be employed to permit filtering of undesired components to give accurate quantification by PLS and ANNs without any preprocessing. The use of PCs as inputs to the ANNs decreased the training time from some 12 h to ca. 5 min. Equivalent concentrations of ampicillin between 0.05 and 20 mM in an E. coli or S. aureus background were quantified with >95% accuracy using this approach.

Noninvasive, On-Line Monitoring of the Biotransformation by Yeast of Glucose to Ethanol Using Dispersive Raman Spectroscopy and Chemometrics

We describe the first application of dispersive Raman spectroscopy using a diode laser exciting at 780 nm and a charge-coupled device (CCD) detector to the noninvasive, on-line determination of the biotransformation by yeast of glucose to ethanol. Software was developed which automatically removed the effects of cosmic rays and other noise, normalized the spectra to an invariant peak, then removed the “baseline” arising from interference by fluorescent impurities, to obtain the “true” Raman spectra. Variable selection was automatically performed on the parameters of relevant Raman peaks (height, width, position of top and center, area and skewness), and a small subset used as the input to cross-validated models based on partial least-squares (PLS) regression. The multivariate calibration models so formed were sufficiently robust to be able to predict the concentration of glucose and ethanol in a completely different fermentation with a precision better than 5%. Dispersive Raman spectroscopy, when coupled with the appropriate chemometrics, is a very useful approach to the noninvasive, on-line determination of the progress of microbial fermentations.

Explanatory Analysis of the Metabolome Using Genetic Programming of Simple, Interpretable Rules

Genetic programming, in conjunction with advanced analytical instruments, is a novel tool for the investigation of complex biological systems at the whole-tissue level. In this study, samples from tomato fruit grown hydroponically under both high- and low-salt conditions were analysed using Fourier-transform infrared spectroscopy (FTIR), with the aim of identifying spectral and biochemical features linked to salinity in the growth environment. FTIR spectra of whole tissue extracts are not amenable to direct visual analysis, so numerical modelling methods were used to generate models capable of classifying the samples based on their spectral characteristics. Genetic programming (GP) provided models with a better prediction accuracy to the conventional data modelling methods used, whilst being much easier to interpret in terms of the variables used. Examination of the GP-derived models showed that there were a small number of spectral regions that were consistently being used. In particular, the spectral region containing absorbances potentially due to a cyanide/nitrile functional group was identified as discriminatory. The explanatory power of the GP models enabled a chemical interpretation of the biochemical differences to be proposed. The combination of FTIR and GP is therefore a powerful and novel analytical tool that, in this study, improves our understanding of the biochemistry of salt tolerance in tomato plants.

Towards Robot Scientists for autonomous scientific discovery

We review the main components of autonomous scientific discovery, and how they lead to the concept of a Robot Scientist. This is a system which uses techniques from artificial intelligence to automate all aspects of the scientific discovery process: it generates hypotheses from a computer model of the domain, designs experiments to test these hypotheses, runs the physical experiments using robotic systems, analyses and interprets the resulting data, and repeats the cycle. We describe our two prototype Robot Scientists: Adam and Eve. Adam has recently proven the potential of such systems by identifying twelve genes responsible for catalysing specific reactions in the metabolic pathways of the yeast Saccharomyces cerevisiae. This work has been formally recorded in great detail using logic. We argue that the reporting of science needs to become fully formalised and that Robot Scientists can help achieve this. This will make scientific information more reproducible and reusable, and promote the integration of computers in scientific reasoning. We believe the greater automation of both the physical and intellectual aspects of scientific investigations to be essential to the future of science. Greater automation improves the accuracy and reliability of experiments, increases the pace of discovery and, in common with conventional laboratory automation, removes tedious and repetitive tasks from the human scientist.