The combination of neural estimates in prediction and decision problems

In this dissertation, different ways of combining neural predictive models or neural-based forecasts are discussed. The proposed approaches consider mostly Gaussian radial basis function networks, which can be efficiently identified and estimated through recursive/adaptive methods. Two different ways of combining are explored to get a final estimate – model mixing and model synthesis –, with the aim of obtaining improvements both in terms of efficiency and effectiveness. In the context of model mixing, the usual framework for linearly combining estimates from different models is extended, to deal with the case where the forecast errors from those models are correlated. In the context of model synthesis, and to address the problems raised by heavily nonstationary time series, we propose hybrid dynamic models for more advanced time series forecasting, composed of a dynamic trend regressive model (or, even, a dynamic harmonic regressive model), and a Gaussian radial basis function network. Additionally, using the model mixing procedure, two approaches for decision-making from forecasting models are discussed and compared: either inferring decisions from combined predictive estimates, or combining prescriptive solutions derived from different forecasting models. Finally, the application of some of the models and methods proposed previously is illustrated with two case studies, based on time series from finance and from tourism.

Investigation of urinary volatile organic metabolites as potential cancer biomarkers by solid-phase microextraction in combination with gas chromatography-mass spectrometry

BACKGROUND: Non-invasive diagnostic strategies aimed at identifying biomarkers of cancer are of great interest for early cancer detection. Urine is potentially a rich source of volatile organic metabolites (VOMs) that can be used as potential cancer biomarkers. Our aim was to develop a generally reliable, rapid, sensitive, and robust analytical method for screening large numbers of urine samples, resulting in a broad spectrum of native VOMs, as a tool to evaluate the potential of these metabolites in the early diagnosis of cancer. METHODS: To investigate urinary volatile metabolites as potential cancer biomarkers, urine samples from 33 cancer patients (oncological group: 14 leukaemia, 12 colorectal and 7 lymphoma) and 21 healthy (control group, cancer-free) individuals were qualitatively and quantitatively analysed. Dynamic solid-phase microextraction in headspace mode (dHS-SPME) using a carboxenpolydimethylsiloxane (CAR/PDMS) sorbent in combination with GC-qMS-based metabolomics was applied to isolate and identify the volatile metabolites. This method provides a potential non-invasive method for early cancer diagnosis as a first approach. To fulfil this objective, three important dHS-SPME experimental parameters that influence extraction efficiency (fibre coating, extraction time and temperature of sampling) were optimised using a univariate optimisation design. The highest extraction efficiency was obtained when sampling was performed at 501C for 60min using samples with high ionic strengths (17% sodium chloride, wv 1) and under agitation. RESULTS: A total of 82 volatile metabolites belonging to distinct chemical classes were identified in the control and oncological groups. Benzene derivatives, terpenoids and phenols were the most common classes for the oncological group, whereas ketones and sulphur compounds were the main classes that were isolated from the urine headspace of healthy subjects. The results demonstrate that compound concentrations were dramatically different between cancer patients and healthy volunteers. The positive rates of 16 patients among the 82 identified were found to be statistically different (Po0.05). A significant increase in the peak area of 2-methyl3-phenyl-2-propenal, p-cymene, anisole, 4-methyl-phenol and 1,2-dihydro-1,1,6-trimethyl-naphthalene in cancer patients was observed. On average, statistically significant lower abundances of dimethyl disulphide were found in cancer patients. CONCLUSIONS: Gas chromatographic peak areas were submitted to multivariate analysis (principal component analysis and supervised linear discriminant analysis) to visualise clusters within cases and to detect the volatile metabolites that are able to differentiate cancer patients from healthy individuals. Very good discrimination within cancer groups and between cancer and control groups was achieved.

A fast method using a new hydrophilic–lipophilic balanced sorbent in combination with ultra-high performance liquid chromatography for quantification of significant bioactive metabolites in wines

This manuscript describes the development and validation of an ultra-fast, efficient, and high throughput analytical method based on ultra-high performance liquid chromatography (UHPLC) equipped with a photodiode array (PDA) detection system, for the simultaneous analysis of fifteen bioactive metabolites: gallic acid, protocatechuic acid, (−)-catechin, gentisic acid, (−)-epicatechin, syringic acid, p-coumaric acid, ferulic acid, m-coumaric acid, rutin, trans-resveratrol, myricetin, quercetin, cinnamic acid and kaempferol, in wines. A 50-mm column packed with 1.7-μm particles operating at elevated pressure (UHPLC strategy) was selected to attain ultra-fast analysis and highly efficient separations. In order to reduce the complexity of wine extract and improve the recovery efficiency, a reverse-phase solid-phase extraction (SPE) procedure using as sorbent a new macroporous copolymer made from a balanced ratio of two monomers, the lipophilic divinylbenzene and the hydrophilic N-vinylpyrrolidone (Oasis™ HLB), was performed prior to UHPLC–PDA analysis. The calibration curves of bioactive metabolites showed good linearity within the established range. Limits of detection (LOD) and quantification (LOQ) ranged from 0.006 μg mL−1 to 0.58 μg mL−1, and from 0.019 μg mL−1 to 1.94 μg mL−1, for gallic and gentisic acids, respectively. The average recoveries ± SD for the three levels of concentration tested (n = 9) in red and white wines were, respectively, 89 ± 3% and 90 ± 2%. The repeatability expressed as relative standard deviation (RSD) was below 10% for all the metabolites assayed. The validated method was then applied to red and white wines from different geographical origins (Azores, Canary and Madeira Islands). The most abundant component in the analysed red wines was (−)-epicatechin followed by (−)-catechin and rutin, whereas in white wines syringic and p-coumaric acids were found the major phenolic metabolites. The method was completely validated, providing a sensitive analysis for bioactive phenolic metabolites detection and showing satisfactory data for all the parameters tested. Moreover, was revealed as an ultra-fast approach allowing the separation of the fifteen bioactive metabolites investigated with high resolution power within 5 min.