Stages Of Model Building Research Articles

Microarray‐based prediction of Parkinson's disease using clinical data as additional response variables

Recent discoveries and developments in the field of genomics have led to the commercialization of novel diagnostic devices for studying disease or estimating therapeutic outcomes in individual patients. With this emerging field, the emphasis is shifting to integration of clinical research into product development. Data acquisition is primary in the initial exploratory phase of product development, and during the process of sample collection and data generation in clinical microarray studies, great amounts of additional information, such as demographic, clinical, and study design variables associated with the data, are often accumulated and made available. Including additional information in classification has been addressed in many different ways. However, in previous studies, the additional information have consistently been treated as extra predictors, which can be a problem for future prediction if such information are not available or collectable for the new samples. We instead propose to adopt a method called canonical partial least squares, which for our purpose, only uses the additional information at the model building stage to stabilize the construction of a classifier for disease status from microarray data. The canonical partial least squares method is compared with regular partial least squares for the classification of Parkinson's disease from gene expression in peripheral blood samples and also through computer simulations. The present study showed that including clinical data in the model building produces simpler and more stable models for prediction of Parkinson's disease from gene expression data.

A goal programming-TOPSIS approach to multiple response optimization using the concepts of non-dominated solutions and prediction intervals

Multiple response problems include three stages: data gathering, modeling and optimization. Most approaches to multiple response optimization ignore the effects of the modeling stage; the model is taken as given and subjected to multi-objective optimization. Moreover, these approaches use subjective methods for the trade off between responses to obtain one or more solutions. In contradistinction, in this paper we use the Prediction Intervals (PIs) from the model building stage to trade off between responses in an objective manner. Our new method combines concepts from the goal programming approach with normalization based on negative and positive ideal solutions as well as the use of prediction intervals for obtaining a set of non-dominated and efficient solutions. Then, the non-dominated solutions (alternatives) are ranked by the TOPSIS (Technique for Order Preference by Similarity to the Ideal Solution) approach. Since some suggested settings of the input variables may not be possible in practice or may lead to unstable operating conditions, this ranking can be extremely helpful to Decision Makers (DMs). The consideration of statistical results together with the selection of the preferred solution among the efficient solutions by Multiple Attribute Decision Making (MADM) distinguishes our approach from others in the literature. We also show, through a numerical example, how the solutions of other methods can be obtained by modifying the relevant approach according to the DM’s requirements.

Minimum Complexity Echo State Network

Reservoir computing (RC) refers to a new class of state-space models with a fixed state transition structure (the reservoir) and an adaptable readout form the state space. The reservoir is supposed to be sufficiently complex so as to capture a large number of features of the input stream that can be exploited by the reservoir-to-output readout mapping. The field of RC has been growing rapidly with many successful applications. However, RC has been criticized for not being principled enough. Reservoir construction is largely driven by a series of randomized model-building stages, with both researchers and practitioners having to rely on a series of trials and errors. To initialize a systematic study of the field, we concentrate on one of the most popular classes of RC methods, namely echo state network, and ask: What is the minimal complexity of reservoir construction for obtaining competitive models and what is the memory capacity (MC) of such simplified reservoirs? On a number of widely used time series benchmarks of different origin and characteristics, as well as by conducting a theoretical analysis we show that a simple deterministically constructed cycle reservoir is comparable to the standard echo state network methodology. The (short-term) MC of linear cyclic reservoirs can be made arbitrarily close to the proved optimal value.

The K5 Lyase KflA Combines a Viral Tail Spike Structure with a Bacterial Polysaccharide Lyase Mechanism

K5 lyase A (KflA) is a tail spike protein (TSP) encoded by a K5A coliphage, which cleaves K5 capsular polysaccharide, a glycosaminoglycan with the repeat unit [-4)-betaGlcA-(1,4)- alphaGlcNAc(1-], displayed on the surface of Escherichia coli K5 strains. The crystal structure of KflA reveals a trimeric arrangement, with each monomer containing a right-handed, single-stranded parallel beta-helix domain. Stable trimer formation by the intertwining of strands in the C-terminal domain, followed by proteolytic maturation, is likely to be catalyzed by an autochaperone as described for K1F endosialidase. The structure of KflA represents the first bacteriophage tail spike protein combining polysaccharide lyase activity with a single-stranded parallel beta-helix fold. We propose a catalytic site and mechanism representing convergence with the syn-beta-elimination site of heparinase II from Pedobacter heparinus.

Clustering-based hybrid inundation model for forecasting flood inundation depths

Estimation of flood depths and extents may provide disaster information for dealing with contingency and alleviating risk and loss of life and property. We present a two-stage procedure underlying CHIM (clustering-based hybrid inundation model), which is composed of linear regression models and ANNs (artificial neural networks) to build the regional flood inundation forecasting model. The two-stage procedure mainly includes data preprocessing and model building stages. In the data preprocessing stage, K-means clustering is used to categorize the data points of the different flooding characteristics in the study area and to identify the control point(s) from individual flooding cluster(s). In the model building stage, three classes of flood depth forecasting models are built in each cluster: the back-propagation neural network (BPNN) for each control point, the linear regression models for the grids that have highly linear correlation with the control point, and a multi-grid BPNN for the grids that do not have highly linear correlation with the control point. The practicability and effectiveness of the proposed approach is tested in the Dacun Township, Changhua County in Central Taiwan. The results show that the proposed CHIM can continuously and adequately provide 1-h-ahead flood inundation maps that well match the simulation flood inundation results and very effectively reduce 99% CPU time.

The data-based mechanistic approach to the modelling, forecasting and control of environmental systems

The paper presents a unified approach to the modelling, forecasting and control of natural and man-made environmental systems. The modelling approach exploits the author’s Data-Based Mechanistic (DBM) modelling philosophy, combined with powerful methods of recursive statistical estimation. These provide the basis for two major stages of model building: first, the critical evaluation of the over-parametrized simulation models that are currently the most common vehicle used in environmental planning and management studies; and second, the adaptive, data-based estimation of parsimonious, ‘top-down’ models that can be used for adaptive forecasting and data assimilation, as well as operational control and management system design. The associated control system design methodology is based on the Non-Minimal State Space (NMSS) approach to the design of Proportional-Integral-Plus (PIP) control systems, based on the DBM models obtained at the previous modelling stage. The paper includes a case study concerned with the modelling and control of globally averaged levels of CO 2 in the atmosphere.

Improving experimental phasing: the role of strongest reflections

Very strong reflections have a dominant impact on the initial phasing and model-building stages of structure determination. However, experimental phasing (MIR, SAD or MAD) fails on some of the strongest reflections when the heavy-atom contribution to scattering is relatively weak or absent. It is shown that when just a few (approximately 50-100) of these reflections are assigned low-error phases, the entire set of isomorphous replacement phases becomes significantly improved after density modification. This improvement is indicated by higher map correlation coefficients and reduced mean phase errors of the updated data. The problem of phasing the strongest reflections may be solved by the direct measurement of triplet phases in a three-beam diffraction experiment. The analysis shows that merging isomorphous replacement data with a limited number of highly accurate phases from the reference-beam diffraction experiment would significantly improve conventional experimental phasing.

Segmentation of quantitative ultrasonographic images of the calcaneus using elastic deformation of the flexible Fourier contour.

This study investigated a new technique for automatic model-based segmentation of broadband ultrasound attenuation (BUA) images of the calcaneus. We determined whether this technique was able to improve osteoporotic fracture discrimination. The segmentation process included 2 major steps: a model-building stage and the automatic segmentation of new image data sets via an elastic deformation of contour models. Broadband ultrasound attenuation was then averaged within the final contour (BUAwhole). The results of the segmentation were validated on a database of 256 patients by comparison of the clinical results obtained with the automatic circular region of interest (BUAcirc) currently implemented on a commercially available ultrasonography unit. All patients were selected by the same physician, who assessed that the fractures were caused by bone fragility on the basis of the circumstances under which fractures occurred and radiologic data. Short-term reproducibility assessed in 49 women was 3.5% and 3.98% for BUAcirc and BUAwhole, respectively. Both BUAcirc (age-adjusted T score, -3.78; P < .0005; age-adjusted odds ratio, 1.92; 95% confidence interval, 1.34-2.75; area under the receiver operating characteristic curve, 0.70) and BUAwhole (age-adjusted T score,-2.73; P < .01; age-adjusted odds ratio, 1.57; 95% confidence interval, 1.12-2.21; area under the curve, 0.67) performed equally well in discriminating healthy postmenopausal patients (n = 150) from those with fractures (n = 60). Fully automatic segmentation by parametrically deformable elastic models for contour using Fourier descriptors can be achieved with reasonable reproducibility and fracture risk prediction. The method is similar to existing methods (automatic circular region of interest); however, the new contour-based region of interest allows more flexible region of interest geometries and placement and potential adaptation to individual anatomy. The method could also possibly be extended to quantitative ultrasonographic imaging at different skeletal sites.

Wave equation calculation of most energetic traveltimes and amplitudes for Kirchhoff prestack migration

Because of its computational efficiency, Kirchhoff migration is the method of choice for 3D prestack depth migration, particularly in the initial velocity model-building stages where several iterations of migration are necessary. In the beginning, Kirchhoff migration used traveltimes calculated by either eikonal solvers or asymptotic ray theory, while amplitudes were calculated using relatively smooth geometrical spreading and obliquity considerations. More recently, Kirchhoff migration has been generalized to include more than one arrival time, with amplitudes calculated from geometrical optics, solution of the transport equations, or Gaussian beams (e.g., Hill, 2001).

Linking Objectives to Actions: A Decision Support Approach Based on Cause–Effect Linkages

ABSTRACTThe process of translating objectives into actions is a difficult task. This difficulty is due to the wide range of possibilities and the lack of structured information. Managers must take into account relevant information and generate a range of options before a decision is reached. So far, little is available to guide managers in translating a set of objectives into actions. This paper presents a three‐stage action‐planning process to address this gap. The process, supported by a software tool, takes managers through the stages of model building, action generation, and action evaluation and selection. A case study illustrates the application of the process. The paper concludes by discussing the implication of this work for managers and academics.

AVTOP: a full integration of TOPMODEL into GIS

The integration of environmental models and Geographical Information Systems (GIS) usually takes two approaches: loose coupling and tight coupling. This paper, however, presents a full coupling approach within a GIS environment that is achieved by developing the existing hydrological models with the macro language of GIS. Such an approach makes it easy to capitalize on the GIS visualization and spatial analysis functions, thereby significantly supporting the dynamic simulation process of hydrological modelling. The full coupling approach is illustrated by AVTOP — a realization of TOPMODEL (TOPography-based hydrological MODEL) with the macro language Avenue of ArcView — within which all stages of model building are undertaken, from the initial parameter input and transformation of datasets, to the visualization of intermediate maps and the final presentation of results in various graphical formats in their geographic context. Furthermore, this approach provides a convenient and single environment for users to visually interact with the model, e.g. by adjusting parameters and observing the corresponding results simultaneously, which significantly facilitates users in exploratory data analysis and decision making in relation to the model applications.

Assignment of homology to genome sequences using a library of hidden Markov models that represent all proteins of known structure

Of the sequence comparison methods, profile-based methods perform with greater selectively than those that use pairwise comparisons. Of the profile methods, hidden Markov models (HMMs) are apparently the best. The first part of this paper describes calculations that (i) improve the performance of HMMs and (ii) determine a good procedure for creating HMMs for sequences of proteins of known structure. For a family of related proteins, more homologues are detected using multiple models built from diverse single seed sequences than from one model built from a good alignment of those sequences. A new procedure is described for detecting and correcting those errors that arise at the model-building stage of the procedure. These two improvements greatly increase selectivity and coverage. The second part of the paper describes the construction of a library of HMMs, called SUPERFAMILY, that represent essentially all proteins of known structure. The sequences of the domains in proteins of known structure, that have identities less than 95 %, are used as seeds to build the models. Using the current data, this gives a library with 4894 models. The third part of the paper describes the use of the SUPERFAMILY model library to annotate the sequences of over 50 genomes. The models match twice as many target sequences as are matched by pairwise sequence comparison methods. For each genome, close to half of the sequences are matched in all or in part and, overall, the matches cover 35 % of eukaryotic genomes and 45 % of bacterial genomes. On average roughly 15% of genome sequences are labelled as being hypothetical yet homologous to proteins of known structure. The annotations derived from these matches are available from a public web server at: http://stash.mrc-lmb.cam.ac.uk/SUPERFAMILY. This server also enables users to match their own sequences against the SUPERFAMILY model library.

Inclusion of risk factor covariates in a segregation analysis of a population-based sample of 426 breast cancer families

Although many segregation analyses of breast cancer have been published, few have included risk factor covariates. Maximum likelihood segregation analyses examining age-at-onset (model 1) and susceptibility (model 2) models of breast cancer were performed on 426 four-generation families originally ascertained between 1944 and 1952 through a breast cancer proband. Cancer status and risk factor data were collected through interviews of participants or surrogates. When segregation analyses were performed on 10,791 women, without estimation of any covariates, all hypotheses under both models were rejected. Model 1, which required estimation of fewer parameters than model 2, provided a better fit to the data according to Akaike's Information Criterion. Further segregation analyses were performed under model 1 on a subset of women with complete data on education, age at first birth (nulliparous women included), and alcohol use, covariates that were found to significantly (P<0.05) improve the fit over the addition of exam age alone in logistic regression models. All three covariates improved the fit of the models, as did year of birth, but at all stages of model building, all of the hypotheses were still rejected. After the allele frequency was fixed at 0.0033, a subset of families appeared to fit a dominant model. Using this model, risk estimates were calculated based on inferred genotype, age, and covariate values. The penetrance was estimated to be 0.15, much lower than previous estimates based on families ascertained through breast cancer probands with early onset. Moreover, the estimates of penetrance were not greatly influenced by incorporation of the measured risk factors.

Ethnocultural predictors of postpartum infant‐care behaviours among immigrants in Canada

Objectives. To determine, among immigrants, what ethnocultural variables predict postpartum infant‐care behaviours over and above other maternal and infant characteristics. Methods. We recruited a cohort of immigrant women on hospital postpartum units. Data on predictor variables were collected during face‐to‐face interviews in the early postpartum period. Follow‐up telephone interviews occurred at 3 months postpartum to assess infant‐care behaviours. We identified 411 women meeting our eligibility criteria, of these 77.3% agreed to participate and 94% received follow‐up interviews. Results. In the first stage of a multiple linear regression analysis, maternal and infant predictors accounted for 24.2% of the variance in the ‘Infant‐Care Behaviours’ score. In the second stage of model building, ethnocultural variables explained an additional 5.8% of the variance. Separate analyses for recent immigrants (resident in Canada less than 3 years) and less recent immigrants (3 or more years) yielded some differences in predictors. Among recent immigrants, worries about the infant's health, mother's education and an interaction term (current immigration status by parity) were significant predictors, explaining 23% of the variance. Among less recent immigrants, worries about the baby's health, prenatal class attendance, marital status and official language comprehension ability explained 37.7% of the variance. Conclusions. Our results indicate that assessing ethnocultural characteristics, in addition to maternal and infant characteristics, improves the prediction of ‘Infant‐Care Behaviours’ scores.

An implementation of prediction calculation using APL and Clipper

This paper describes the algorithm for the prediction calculation based upon the non-linear algebraic models' self-organization. Special block for a function recognition provides the decrease of the predicted equations exhaustive search at every stage of model building. Another peculiarity is the bootstrap use as supplementary external criterion. The implementation of this algorithm is a complex software system (calculating module is built in APL*PLUS and user-friendly interface is built in Clipper 5.01). This system works as DOS-application for Windows with dynamic data exchange. We present results of testing the system on real currency exchange rate.

Machine Learning and Simulation: Application in Queuing Systems

Machine learning algorithins have application potential in simulation at stages of model building, simulation execution monitoring in real-time, and output analysis and model refinement. This article presents the basics of the inductive learning paradigm and the ID3 algorithm, highlights the potential of learning in simulation as means for extracting heuristic rules to be used in expert systems, and demonstrates a sample application in simulation of queuing systems.

Mise en œuvre des budgets économiques au Niger

Developing economic budgets in Niger, by Jean Mathis, Guy de Monchy et Marc Rocca. Under a United Nations Development Programm (Undp) project to develop short-term economic analysis and forecasting in the Niger Republic, several teams of experts have visited this country since the end of 1983. The aim of this article is to show through a tentative experiment earned out in Niger the stages which can be reached in developing this type of forecasting depending on the economic information available and the existing administrative structures. The stages described here do not take into account the most recent missions and therefore do not reflect the latest situation with regard to the Undp's work to develop its short- term forecasting policy in the Niger Republic. After an introductory section explaining the approach used, three other sections show the transition process from analysis of how the economy functions to various stages of formalization thereof. First of all, therefore, the physical characteristics of the economy of Niger and its major macroeconomic circuits are described. The authors then explain the model-building stage and what can be learned in this way when statistical information is lacking. The last section presents a short-term forecasting model derived from an analogous pedagogical model developed for a macroeconomic forecast seminar organized by the Un at Niamey in July 1984. The conclusion touches on possible ways of improving on the work presented and certain conditions to be fulfilled if it is to be integrated into the economic policy development process.

The asymptotic distribution of the sample inverse autocorrelations of an autoregressive-moving average process

SUMMARY The inverse autocorrelations of a time series are useful at the identification stage of model building. In practice these quantities must be estimated from the data and so the distributional properties of the estimators are of great importance. In this paper the asymptotic distribution of the sample inverse autocorrelations is derived, making use of the duality between the autocorrelations and inverse autocorrelations of two autoregressive-moving average processes.

A Robust Method for Multiple Linear Regression

Techniques of fitting are said to be resistant when the result is not greatly altered in the case a small fraction of the data is altered: techniques of fitting are said to be robust of efficiency when their statistical efficiency remains high for conditions more realistic than the utopian cases of Gaussian distributions with errors of equal variance. These properties are particularly important in the formative stages of model building when the form of the response is not known exactly. Techniques with these properties are proposed and discussed.

A Statistical Analysis of Telephone Noise

The results of a statistical analysis of telephone noise are presented. The analysis consists of two stages: an exploratory data analysis stage, where the data are characterized through various nonparametric statistics and a model-building stage, where the data are matched to models. The exploratory data analysis stage involved examination of noise waveforms, power spectra, and covariance estimates. The results show that telephone noise consists of a deterministic component (sinusoids at various frequencies) and a stochastic component. It is assumed that these components add. The data are filtered to remove the deterministic component. Next, central moment estimates are presented, as well as first-order amplitude statistics (histograms and empirical cumulative distributions) for these filtered data. The results indicate that the filtered data appear wide-sense stationary over short periods of time (typically 1 second) and, although close to gaussian, are distinctly nongaussian. The model-building stage involved fitting the filtered data to two classes of models. The first class of models is based on symmetric stable distributions that arise from the central limit theorem. The second class of models assumes two different physical processes that contribute to the random component of telephone noise: The low-variance process is assumed to be gaussian, while the high-variance component is assumed to be a filtered Poisson process. Both classes of models agree intuitively with the physical processes generating telephone noise and are mathematically tractable. Based largely on graphical tests, both models appear to fit the filtered data reasonably well.