Selection bias in gene extraction on the basis of microarray gene-expression data

In the context of cancer diagnosis and treatment, we consider the problem of constructing an accurate prediction rule on the basis of a relatively small number of tumor tissue samples of known type containing the expression data on very many (possibly thousands) genes. Recently, results have been presented in the literature suggesting that it is possible to construct a prediction rule from only a few genes such that it has a negligible prediction error rate. However, in these results the test error or the leave-one-out cross-validated error is calculated without allowance for the selection bias. There is no allowance because the rule is either tested on tissue samples that were used in the first instance to select the genes being used in the rule or because the cross-validation of the rule is not external to the selection process; that is, gene selection is not performed in training the rule at each stage of the cross-validation process. We describe how in practice the selection bias can be assessed and corrected for by either performing a cross-validation or applying the bootstrap external to the selection process. We recommend using 10-fold rather than leave-one-out cross-validation, and concerning the bootstrap, we suggest using the so-called. 632+ bootstrap error estimate designed to handle overfitted prediction rules. Using two published data sets, we demonstrate that when correction is made for the selection bias, the cross-validated error is no longer zero for a subset of only a few genes.

ANNA: A new prediction method for bioassessment programs

1. Cluster analysis of reference sites with similar biota is the initial step in creating River Invertebrate Prediction and Classification System (RIVPACS) and similar river bioassessment models such as Australian River Assessment System (AUSRIVAS). This paper describes and tests an alternative prediction method, Assessment by Nearest Neighbour Analysis (ANNA), based on the same philosophy as RIVPACS and AUSRIVAS but without the grouping step that some people view as artificial. 2. The steps in creating ANNA models are: (i) weighting the predictor variables using a multivariate approach analogous to principal axis correlations, (ii) calculating the weighted Euclidian distance from a test site to the reference sites based on the environmental predictors, (iii) predicting the faunal composition based on the nearest reference sites and (iv) calculating an observed/expected (O/E) analogous to RIVPACS/AUSRIVAS. 3. The paper compares AUSRIVAS and ANNA models on 17 datasets representing a variety of habitats and seasons. First, it examines each model's regressions for Observed versus Expected number of taxa, including the r(2), intercept and slope. Second, the two models' assessments of 79 test sites in New Zealand are compared. Third, the models are compared on test and presumed reference sites along a known trace metal gradient. Fourth, ANNA models are evaluated for western Australia, a geographically distinct region of Australia. The comparisons demonstrate that ANNA and AUSRIVAS are generally equivalent in performance, although ANNA turns out to be potentially more robust for the O versus E regressions and is potentially more accurate on the trace metal gradient sites. 4. The ANNA method is recommended for use in bioassessment of rivers, at least for corroborating the results of the well established AUSRIVAS- and RIVPACS-type models, if not to replace them.

Prediction of MHC class II-binding peptides using an evolutionary algorithm and artificial neural network

Motivation: Prediction methods for identifying binding peptides could minimize the number of peptides required to be synthesized and assayed, and thereby facilitate the identification of potential T-cell epitopes. We developed a bioinformatic method for the prediction of peptide binding to MHC class II molecules. Results: Experimental binding data and expert knowledge of anchor positions and binding motifs were combined with an evolutionary algorithm (EA) and an artificial neural network (ANN): binding data extraction --> peptide alignment --> ANN training and classification. This method, termed PERUN, was implemented for the prediction of peptides that bind to HLA-DR4(B1*0401). The respective positive predictive values of PERUN predictions of high-, moderate-, low- and zero-affinity binder-a were assessed as 0.8, 0.7, 0.5 and 0.8 by cross-validation, and 1.0, 0.8, 0.3 and 0.7 by experimental binding. This illustrates the synergy between experimentation and computer modeling, and its application to the identification of potential immunotheraaeutic peptides.

Evaluation of a new bioelectrical impedance instrument for the prediction of body cell mass independently of height or weight

The objective of the present study was to evaluate the performance of a new bioelectrical impedance instrument, the Soft Tissue Analyzer (STA), which predicts a subject's body composition. A cross-sectional population study in which the impedance of 205 healthy adult subjects was measured using the STA. Extracellular water (ECW) volume (as a percentage of total body water, TBW) and fat-free mass (FFM) were predicted by both the STA and a compartmental model, and compared according to correlation and limits of agreement analysis, with the equivalent data obtained by independent reference methods of measurement (TBW measured by D2O dilution, and FFM measured by dual-energy X-ray absorptiometry). There was a small (2.0 kg) but significant (P < 0.02) difference in mean FFM predicted by the STA, compared with the reference technique in the males, but not in the females (-0.4 kg) or in the combined group (0.8 kg). Both methods were highly correlated. Similarly, small but significant differences for predicted mean ECW volume were observed. The limits of agreement for FFM and ECW were -7.5-9.9 and -4.1-3.0 kg, respectively. Both FFM and ECW (as a percentage of TBW) are well predicted by the STA on a population basis, but the magnitude of the limits of agreement with reference methods may preclude its usefulness for predicting body composition in an individual. In addition, the theoretical basis of an impedance method that does not include a measure of conductor length requires further validation. (C) Elsevier Science Inc. 2000.

Methods for prediction of peptide binding to MHC molecules: A comparative study

Background: A variety of methods for prediction of peptide binding to major histocompatibility complex (MHC) have been proposed. These methods are based on binding motifs, binding matrices, hidden Markov models (HMM), or artificial neural networks (ANN). There has been little prior work on the comparative analysis of these methods. Materials and Methods: We performed a comparison of the performance of six methods applied to the prediction of two human MHC class I molecules, including binding matrices and motifs, ANNs, and HMMs. Results: The selection of the optimal prediction method depends on the amount of available data (the number of peptides of known binding affinity to the MHC molecule of interest), the biases in the data set and the intended purpose of the prediction (screening of a single protein versus mass screening). When little or no peptide data are available, binding motifs are the most useful alternative to random guessing or use of a complete overlapping set of peptides for selection of candidate binders. As the number of known peptide binders increases, binding matrices and HMM become more useful predictors. ANN and HMM are the predictive methods of choice for MHC alleles with more than 100 known binding peptides. Conclusion: The ability of bioinformatic methods to reliably predict MHC binding peptides, and thereby potential T-cell epitopes, has major implications for clinical immunology, particularly in the area of vaccine design.

Inter-subband redundancy prediction using neural network for video coding

High performance video codec is mandatory for multimedia applications such as video-on-demand and video conferencing. Recent research has proposed numerous video coding techniques to meet the requirement in bandwidth, delay, loss and Quality-of-Service (QoS). In this paper, we present our investigations on inter-subband self-similarity within the wavelet-decomposed video frames using neural networks, and study the performance of applying the spatial network model to all video frames over time. The goal of our proposed method is to restore the highest perceptual quality for video transmitted over a highly congested network. Our contributions in this paper are: (1) A new coding model with neural network based, inter-subband redundancy (ISR) prediction for video coding using wavelet (2) The performance of 1D and 2D ISR prediction, including multiple levels of wavelet decompositions. Our result shows a short-term quality enhancement may be obtained using both 1D and 2D ISR prediction.

Prediction of carbon monoxide in fires by conditional moment closure

Carbon monoxide is the chief killer in fires. Dangerous levels of CO can occur when reacting combustion gases are quenched by heat transfer, or by mixing of the fire plume in a cooled under- or overventilated upper layer. In this paper, carbon monoxide predictions for enclosure fires are modeled by the conditional moment closure (CMC) method and are compared with laboratory data. The modeled fire situation is a buoyant, turbulent, diffusion flame burning under a hood. The fire plume entrains fresh air, and the postflame gases are cooled considerably under the hood by conduction and radiation, emulating conditions which occur in enclosure fires and lead to the freezing of CO burnout. Predictions of CO in the cooled layer are presented in the context of a complete computational fluid dynamics solution of velocity, temperature, and major species concentrations. A range of underhood equivalence ratios, from rich to lean, are investigated. The CMC method predicts CO in very good agreement with data. In particular, CMC is able to correctly predict CO concentrations in lean cooled gases, showing its capability in conditions where reaction rates change considerably.

Bayesian analysis of the error correction model

This paper presents a method for estimating the posterior probability density of the cointegrating rank of a multivariate error correction model. A second contribution is the careful elicitation of the prior for the cointegrating vectors derived from a prior on the cointegrating space. This prior obtains naturally from treating the cointegrating space as the parameter of interest in inference and overcomes problems previously encountered in Bayesian cointegration analysis. Using this new prior and Laplace approximation, an estimator for the posterior probability of the rank is given. The approach performs well compared with information criteria in Monte Carlo experiments. (C) 2003 Elsevier B.V. All rights reserved.

Prediction of impact forces in a vibratory ball mill using an inverse technique

To understand the dynamic mechanisms of the mechanical milling process in a vibratory mill, it is necessary to determine the characteristics of the impact forces associated with the collision events. However, it is difficult to directly measure the impact force in an operating mill. This paper describes an inverse technique for the prediction of impact forces from acceleration measurements on a vibratory ball mill. The characteristics of the vibratory mill have been investigated by the modal testing technique, and its system modes have been identified. In the modelling of the system vibration response to the impact forces, two modal equations have been used to describe the modal responses. The superposition of the modal responses gives rise to the total response of the system. A method based on an optimisation approach has been developed to predict the impact forces by minimising the difference between the measured acceleration of the vibratory ball mill and the predicted acceleration from the solution of the modal equations. The predicted and measured impact forces are in good agreement. Copyright (C) 1996 Elsevier Science Ltd.

A systematic approach to error isolation in computerized wastewater simulation models

Activated sludge models are used extensively in the study of wastewater treatment processes. While various commercial implementations of these models are available, there are many people who need to code models themselves using the simulation packages available to them, Quality assurance of such models is difficult. While benchmarking problems have been developed and are available, the comparison of simulation data with that of commercial models leads only to the detection, not the isolation of errors. To identify the errors in the code is time-consuming. In this paper, we address the problem by developing a systematic and largely automated approach to the isolation of coding errors. There are three steps: firstly, possible errors are classified according to their place in the model structure and a feature matrix is established for each class of errors. Secondly, an observer is designed to generate residuals, such that each class of errors imposes a subspace, spanned by its feature matrix, on the residuals. Finally. localising the residuals in a subspace isolates coding errors. The algorithm proved capable of rapidly and reliably isolating a variety of single and simultaneous errors in a case study using the ASM 1 activated sludge model. In this paper a newly coded model was verified against a known implementation. The method is also applicable to simultaneous verification of any two independent implementations, hence is useful in commercial model development.

Prediction of coal ash fusion temperatures with the F*A*C*T thermodynamic computer package

This paper outlines research on the processes taking place within the coal mineral matter at high temperatures and development of the relationship between ash fusion temperatures (AFT) and phase equilibria of the coal ash slags. A new thermodynamic database for the Al-Ca-Fe-O-Si system developed by the author was used in conjunction with the thermodynamic computer package F*A*C*T for these purposes. In addition, high temperature experimental studies were undertaken that involved heat treatment and quenching of the ash cones followed by the analyses using different techniques. The study provided new information on the processes taking place during AFT test and demonstrated the validity of the AFTs predictions with F*A*C*T. Examples of practical applications of the AFT prediction method are given in the paper. The results of this study are important not only for the AFT predictions, but also in general for the application of phase equilibrium science to the characterisation of the coal mineral matter interactions at high temperature. (C) 2002 Elsevier Science Ltd. All rights reserved.

Comparison of an ELISA and an LC/MS/MS method for measuring tacrolimus concentrations and making dosage decisions in transplant recipients

This study compared an enzyme-linked immunosorbent assay (ELISA) to a liquid chromatography-tandem mass spectrometry (LC/MS/MS) technique for measurement of tacrolimus concentrations in adult kidney and liver transplant recipients, and investigated how assay choice influenced pharmacokinetic parameter estimates and drug dosage decisions. Tacrolimus concentrations measured by both ELISA and LC/MS/MS from 29 kidney (n = 98 samples) and 27 liver (n = 97 samples) transplant recipients were used to evaluate the performance of these methods in the clinical setting. Tacrolimus concentrations measured by the two techniques were compared via regression analysis. Population pharmacokinetic models were developed independently using ELISA and LC/MS/MS data from 76 kidney recipients. Derived kinetic parameters were used to formulate typical dosing regimens for concentration targeting. Dosage recommendations for the two assays were compared. The relation between LC/MS/MS and ELISA measurements was best described by the regression equation ELISA = 1.02 . (LC/MS/MS) + 0.14 in kidney recipients, and ELISA = 1.12 . (LC/MS/MS) - 0.87 in liver recipients. ELISA displayed less accuracy than LC/MS/MS at lower tacrolimus concentrations. Population pharmacokinetic models based on ELISA and LC/MS/MS data were similar with residual random errors of 4.1 ng/mL and 3.7 ng/mL, respectively. Assay choice gave rise to dosage prediction differences ranging from 0% to 30%. ELISA measurements of tacrolimus are not automatically interchangeable with LC/MS/MS values. Assay differences were greatest in adult liver recipients, probably reflecting periods of liver dysfunction and impaired biliary secretion of metabolites. While the majority of data collected in this study suggested assay differences in adult kidney recipients were minimal, findings of ELISA dosage underpredictions of up to 25% in the long term must be investigated further.

Estimation of total body water from bioelectrical impedance analysis in patients with pancreatic cancer - agreement between three methods of prediction

Introduction Bioelectrical impedance analysis (BIA) is a useful field measure to estimate total body water (TBW). No prediction formulae have been developed or validated against a reference method in patients with pancreatic cancer. The aim of this study was to assess the agreement between three prediction equations for the estimation of TBW in cachectic patients with pancreatic cancer. Methods Resistance was measured at frequencies of 50 and 200 kHz in 18 outpatients (10 males and eight females, age 70.2 +/- 11.8 years) with pancreatic cancer from two tertiary Australian hospitals. Three published prediction formulae were used to calculate TBW - TBWs developed in surgical patients, TBWca-uw and TBWca-nw developed in underweight and normal weight patients with end-stage cancer. Results There was no significant difference in the TBW estimated by the three prediction equations - TBWs 32.9 +/- 8.3 L, TBWca-nw 36.3 +/- 7.4 L, TBWca-uw 34.6 +/- 7.6 L. At a population level, there is agreement between prediction of TBW in patients with pancreatic cancer estimated from the three equations. The best combination of low bias and narrow limits of agreement was observed when TBW was estimated from the equation developed in the underweight cancer patients relative to the normal weight cancer patients. When no established BIA prediction equation exists, practitioners should utilize an equation developed in a population with similar critical characteristics such as diagnosis, weight loss, body mass index and/or age. Conclusions Further research is required to determine the accuracy of the BIA prediction technique against a reference method in patients with pancreatic cancer.

Prediction of Golgi Type II membrane proteins based on their transmembrane domains

Motivation: A major issue in cell biology today is how distinct intracellular regions of the cell, like the Golgi Apparatus, maintain their unique composition of proteins and lipids. The cell differentially separates Golgi resident proteins from proteins that move through the organelle to other subcellular destinations. We set out to determine if we could distinguish these two types of transmembrane proteins using computational approaches. Results: A new method has been developed to predict Golgi membrane proteins based on their transmembrane domains. To establish the prediction procedure, we took the hydrophobicity values and frequencies of different residues within the transmembrane domains into consideration. A simple linear discriminant function was developed with a small number of parameters derived from a dataset of Type II transmembrane proteins of known localization. This can discriminate between proteins destined for Golgi apparatus or other locations (post-Golgi) with a success rate of 89.3% or 85.2%, respectively on our redundancy-reduced data sets.