Evaluating the effectiveness of Bayesian feature selection

A practical Bayesian approach for inference in neural network models has been available for ten years, and yet it is not used frequently in medical applications. In this chapter we show how both regularisation and feature selection can bring significant benefits in diagnostic tasks through two case studies: heart arrhythmia classification based on ECG data and the prognosis of lupus. In the first of these, the number of variables was reduced by two thirds without significantly affecting performance, while in the second, only the Bayesian models had an acceptable accuracy. In both tasks, neural networks outperformed other pattern recognition approaches.

Approximate Bayesian techniques for inference in stochastic dynamical systems

This thesis is concerned with approximate inference in dynamical systems, from a variational Bayesian perspective. When modelling real world dynamical systems, stochastic differential equations appear as a natural choice, mainly because of their ability to model the noise of the system by adding a variant of some stochastic process to the deterministic dynamics. Hence, inference in such processes has drawn much attention. Here two new extended frameworks are derived and presented that are based on basis function expansions and local polynomial approximations of a recently proposed variational Bayesian algorithm. It is shown that the new extensions converge to the original variational algorithm and can be used for state estimation (smoothing). However, the main focus is on estimating the (hyper-) parameters of these systems (i.e. drift parameters and diffusion coefficients). The new methods are numerically validated on a range of different systems which vary in dimensionality and non-linearity. These are the Ornstein-Uhlenbeck process, for which the exact likelihood can be computed analytically, the univariate and highly non-linear, stochastic double well and the multivariate chaotic stochastic Lorenz '63 (3-dimensional model). The algorithms are also applied to the 40 dimensional stochastic Lorenz '96 system. In this investigation these new approaches are compared with a variety of other well known methods such as the ensemble Kalman filter / smoother, a hybrid Monte Carlo sampler, the dual unscented Kalman filter (for jointly estimating the systems states and model parameters) and full weak-constraint 4D-Var. Empirical analysis of their asymptotic behaviour as a function of observation density or length of time window increases is provided.

Estimating parameters in stochastic systems:a variational Bayesian approach

This work is concerned with approximate inference in dynamical systems, from a variational Bayesian perspective. When modelling real world dynamical systems, stochastic differential equations appear as a natural choice, mainly because of their ability to model the noise of the system by adding a variation of some stochastic process to the deterministic dynamics. Hence, inference in such processes has drawn much attention. Here a new extended framework is derived that is based on a local polynomial approximation of a recently proposed variational Bayesian algorithm. The paper begins by showing that the new extension of this variational algorithm can be used for state estimation (smoothing) and converges to the original algorithm. However, the main focus is on estimating the (hyper-) parameters of these systems (i.e. drift parameters and diffusion coefficients). The new approach is validated on a range of different systems which vary in dimensionality and non-linearity. These are the Ornstein–Uhlenbeck process, the exact likelihood of which can be computed analytically, the univariate and highly non-linear, stochastic double well and the multivariate chaotic stochastic Lorenz ’63 (3D model). As a special case the algorithm is also applied to the 40 dimensional stochastic Lorenz ’96 system. In our investigation we compare this new approach with a variety of other well known methods, such as the hybrid Monte Carlo, dual unscented Kalman filter, full weak-constraint 4D-Var algorithm and analyse empirically their asymptotic behaviour as a function of observation density or length of time window increases. In particular we show that we are able to estimate parameters in both the drift (deterministic) and the diffusion (stochastic) part of the model evolution equations using our new methods.

Neural Network Modelling with Input Uncertainty: Theory and Application

It is generally assumed when using Bayesian inference methods for neural networks that the input data contains no noise. For real-world (errors in variable) problems this is clearly an unsafe assumption. This paper presents a Bayesian neural network framework which accounts for input noise provided that a model of the noise process exists. In the limit where the noise process is small and symmetric it is shown, using the Laplace approximation, that this method adds an extra term to the usual Bayesian error bar which depends on the variance of the input noise process. Further, by treating the true (noiseless) input as a hidden variable, and sampling this jointly with the network’s weights, using a Markov chain Monte Carlo method, it is demonstrated that it is possible to infer the regression over the noiseless input. This leads to the possibility of training an accurate model of a system using less accurate, or more uncertain, data. This is demonstrated on both the, synthetic, noisy sine wave problem and a real problem of inferring the forward model for a satellite radar backscatter system used to predict sea surface wind vectors.

Population pharmacokinetic and pharmacogenetic analysis of tacrolimus in paediatric liver transplant patients

Aims - To build a population pharmacokinetic model that describes the apparent clearance of tacrolimus and the potential demographic, clinical and genetically controlled factors that could lead to inter-patient pharmacokinetic variability within children following liver transplantation. Methods - The present study retrospectively examined tacrolimus whole blood pre-dose concentrations (n = 628) of 43 children during their first year post-liver transplantation. Population pharmacokinetic analysis was performed using the non-linear mixed effects modelling program (nonmem) to determine the population mean parameter estimate of clearance and influential covariates. Results - The final model identified time post-transplantation and CYP3A5*1 allele as influential covariates on tacrolimus apparent clearance according to the following equation: TVCL = 12.9 x (Weight/13.2)0.35 x EXP (-0.0058 x TPT) x EXP (0.428 x CYP3A5) where TVCL is the typical value for apparent clearance, TPT is time post-transplantation in days and the CYP3A5 is 1 where *1 allele is present and 0 otherwise. The population estimate and inter-individual variability (%CV) of tacrolimus apparent clearance were found to be 0.977 l h−1 kg−1 (95% CI 0.958, 0.996) and 40.0%, respectively, while the residual variability between the observed and predicted concentrations was 35.4%. Conclusion Tacrolimus apparent clearance was influenced by time post-transplantation and CYP3A5 genotypes. The results of this study, once confirmed by a large scale prospective study, can be used in conjunction with therapeutic drug monitoring to recommend tacrolimus dose adjustments that take into account not only body weight but also genetic and time-related changes in tacrolimus clearance.

Prophylactic ranitidine treatment in critically ill children - a population pharmacokinetic study

Aims - To characterize the population pharmacokinetics of ranitidine in critically ill children and to determine the influence of various clinical and demographic factors on its disposition. Methods - Data were collected prospectively from 78 paediatric patients (n = 248 plasma samples) who received oral or intravenous ranitidine for prophylaxis against stress ulcers, gastrointestinal bleeding or the treatment of gastro-oesophageal reflux. Plasma samples were analysed using high-performance liquid chromatography, and the data were subjected to population pharmacokinetic analysis using nonlinear mixed-effects modelling. Results - A one-compartment model best described the plasma concentration profile, with an exponential structure for interindividual errors and a proportional structure for intra-individual error. After backward stepwise elimination, the final model showed a significant decrease in objective function value (−12.618; P < 0.001) compared with the weight-corrected base model. Final parameter estimates for the population were 32.1 l h−1 for total clearance and 285 l for volume of distribution, both allometrically modelled for a 70 kg adult. Final estimates for absorption rate constant and bioavailability were 1.31 h−1 and 27.5%, respectively. No significant relationship was found between age and weight-corrected ranitidine pharmacokinetic parameters in the final model, with the covariate for cardiac failure or surgery being shown to reduce clearance significantly by a factor of 0.46. Conclusions - Currently, ranitidine dose recommendations are based on children's weights. However, our findings suggest that a dosing scheme that takes into consideration both weight and cardiac failure/surgery would be more appropriate in order to avoid administration of higher or more frequent doses than necessary.

Population pharmacokinetic and pharmacogenetic analysis of 6-mercaptopurine in paediatric patients with acute lymphoblastic leukaemia

WHAT IS ALREADY KNOWN ABOUT THIS SUBJECT • The cytotoxic effects of 6-mercaptopurine (6-MP) were found to be due to drug-derived intracellular metabolites (mainly 6-thioguanine nucleotides and to some extent 6-methylmercaptopurine nucleotides) rather than the drug itself. • Current empirical dosing methods for oral 6-MP result in highly variable drug and metabolite concentrations and hence variability in treatment outcome. WHAT THIS STUDY ADDS • The first population pharmacokinetic model has been developed for 6-MP active metabolites in paediatric patients with acute lymphoblastic leukaemia and the potential demographic and genetically controlled factors that could lead to interpatient pharmacokinetic variability among this population have been assessed. • The model shows a large reduction in interindividual variability of pharmacokinetic parameters when body surface area and thiopurine methyltransferase polymorphism are incorporated into the model as covariates. • The developed model offers a more rational dosing approach for 6-MP than the traditional empirical method (based on body surface area) through combining it with pharmacogenetically guided dosing based on thiopurine methyltransferase genotype. AIMS - To investigate the population pharmacokinetics of 6-mercaptopurine (6-MP) active metabolites in paediatric patients with acute lymphoblastic leukaemia (ALL) and examine the effects of various genetic polymorphisms on the disposition of these metabolites. METHODS - Data were collected prospectively from 19 paediatric patients with ALL (n = 75 samples, 150 concentrations) who received 6-MP maintenance chemotherapy (titrated to a target dose of 75 mg m−2 day−1). All patients were genotyped for polymorphisms in three enzymes involved in 6-MP metabolism. Population pharmacokinetic analysis was performed with the nonlinear mixed effects modelling program (nonmem) to determine the population mean parameter estimate of clearance for the active metabolites. RESULTS - The developed model revealed considerable interindividual variability (IIV) in the clearance of 6-MP active metabolites [6-thioguanine nucleotides (6-TGNs) and 6-methylmercaptopurine nucleotides (6-mMPNs)]. Body surface area explained a significant part of 6-TGNs clearance IIV when incorporated in the model (IIV reduced from 69.9 to 29.3%). The most influential covariate examined, however, was thiopurine methyltransferase (TPMT) genotype, which resulted in the greatest reduction in the model's objective function (P < 0.005) when incorporated as a covariate affecting the fractional metabolic transformation of 6-MP into 6-TGNs. The other genetic covariates tested were not statistically significant and therefore were not included in the final model. CONCLUSIONS - The developed pharmacokinetic model (if successful at external validation) would offer a more rational dosing approach for 6-MP than the traditional empirical method since it combines the current practice of using body surface area in 6-MP dosing with a pharmacogenetically guided dosing based on TPMT genotype.

Potassium canrenoate treatment in paediatric patients:a population pharmacokinetic study using novel dried blood spot sampling

Objective: To characterize the population pharmacokinetics of canrenone following administration of potassium canrenoate (K-canrenoate) in paediatric patients. Methods: Data were collected prospectively from 37 paediatric patients (median weight 2.9 kg, age range 2 days–0.85 years) who received intravenous K-canrenoate for management of retained fluids, for example in heart failure and chronic lung disease. Dried blood spot (DBS) samples (n = 213) from these were analysed for canrenone content and the data subjected to pharmacokinetic analysis using nonlinear mixed-effects modelling. Another group of patients (n = 16) who had 71 matching plasma and DBS samples was analysed separately to compare canrenone pharmacokinetic parameters obtained using the two different matrices. Results: A one-compartment model best described the DBS data. Significant covariates were weight, postmenstrual age (PMA) and gestational age. The final population models for canrenone clearance (CL/F) and volume of distribution (V/F) in DBS were CL/F (l/h) = 12.86 ×  (WT/70.0)0.75 × e [0.066 ×  (PMA - 40]) and V/F (l) = 603.30 ×  (WT/70) × (GA/40)1.89 where weight is in kilograms. The corresponding values of CL/F and V/F in a patient with a median weight of 2.9 kg are 1.11 l/h and 20.48 l, respectively. Estimated half-life of canrenone based on DBS concentrations was similar to that based on matched plasma concentrations (19.99 and 19.37 h, respectively, in 70 kg patient). Conclusion: The range of estimated CL/F in DBS for the study population was 0.12–9.62 l/h; hence, bodyweight-based dosage adjustment of K-canrenoate appears necessary. However, a dosing scheme that takes into consideration both weight and age (PMA/gestational age) of paediatric patients seems more appropriate.

Population pharmacokinetics of single-dose intravenous paracetamol in children

Background To determine the pharmacokinetics (PK) of a new i.v. formulation of paracetamol (Perfalgan) in children ≤15 yr of age. Methods After obtaining written informed consent, children under 16 yr of age were recruited to this study. Blood samples were obtained at 0, 15, 30 min, 1, 2, 4, 6, and 8 h after administration of a weight-dependent dose of i.v. paracetamol. Paracetamol concentration was measured using a validated high-performance liquid chromatographic assay with ultraviolet detection method, with a lower limit of quantification (LLOQ) of 900 pg on column and an intra-day coefficient of variation of 14.3% at the LLOQ. Population PK analysis was performed by non-linear mixed-effect modelling using NONMEM. Results One hundred and fifty-nine blood samples from 33 children aged 1.8–15 yr, weight 13.7–56 kg, were analysed. Data were best described by a two-compartment model. Only body weight as a covariate significantly improved the goodness of fit of the model. The final population models for paracetamol clearance (CL), V1 (central volume of distribution), Q (inter-compartmental clearance), and V2 (peripheral volume of distribution) were: 16.51×(WT/70)0.75, 28.4×(WT/70), 11.32×(WT/70)0.75, and 13.26×(WT/70), respectively (CL, Q in litres per hour, WT in kilograms, and V1 and V2 in litres). Conclusions In children aged 1.8–15 yr, the PK parameters for i.v. paracetamol were not influenced directly by age but were by total body weight and, using allometric size scaling, significantly affected the clearances (CL, Q) and volumes of distribution (V1, V2).

A weakly-supervised Bayesian model for violence detection from social media

Social streams have proven to be the mostup-to-date and inclusive information on cur-rent events. In this paper we propose a novelprobabilistic modelling framework, called violence detection model (VDM), which enables the identification of text containing violent content and extraction of violence-related topics over social media data. The proposed VDM model does not require any labeled corpora for training, instead, it only needs the in-corporation of word prior knowledge which captures whether a word indicates violence or not. We propose a novel approach of deriving word prior knowledge using the relative entropy measurement of words based on the in-tuition that low entropy words are indicative of semantically coherent topics and therefore more informative, while high entropy words indicates words whose usage is more topical diverse and therefore less informative. Our proposed VDM model has been evaluated on the TREC Microblog 2011 dataset to identify topics related to violence. Experimental results show that deriving word priors using our proposed relative entropy method is more effective than the widely-used information gain method. Moreover, VDM gives higher violence classification results and produces more coherent violence-related topics compared toa few competitive baselines.

Alpha helical trans-membrane proteins:enhanced prediction using a Bayesian approach

Membrane proteins, which constitute approximately 20% of most genomes, are poorly tractable targets for experimental structure determination, thus analysis by prediction and modelling makes an important contribution to their on-going study. Membrane proteins form two main classes: alpha helical and beta barrel trans-membrane proteins. By using a method based on Bayesian Networks, which provides a flexible and powerful framework for statistical inference, we addressed alpha-helical topology prediction. This method has accuracies of 77.4% for prokaryotic proteins and 61.4% for eukaryotic proteins. The method described here represents an important advance in the computational determination of membrane protein topology and offers a useful, and complementary, tool for the analysis of membrane proteins for a range of applications.

CFD-modelling of insulation debris transport phenomena in water flow

The investigation of insulation debris generation, transport and sedimentation becomes important with regard to reactor safety research for PWR and BWR, when considering the long-term behavior of emergency core cooling systems during all types of loss of coolant accidents (LOCA). The insulation debris released near the break during a LOCA incident consists of a mixture of disparate particle population that varies with size, shape, consistency and other properties. Some fractions of the released insulation debris can be transported into the reactor sump, where it may perturb/impinge on the emergency core cooling systems. Open questions of generic interest are the sedimentation of the insulation debris in a water pool, its possible re-suspension and transport in the sump water flow and the particle load on strainers and corresponding pressure drop. A joint research project on such questions is being performed in cooperation between the University of Applied Sciences Zittau/Görlitz and the Forschungszentrum Dresden-Rossendorf. The project deals with the experimental investigation of particle transport phenomena in coolant flow and the development of CFD models for its description. While the experiments are performed at the University at Zittau/Görlitz, the theoretical modeling efforts are concentrated at Forschungszentrum Dresden-Rossendorf. Whereas the paper Alt et al. is focused on the experiments in the present paper the basic concepts for CFD modeling are described and feasibility studies including the conceptual design of the experiments are presented.

Computational modelling approaches to vaccinology

Excepting the Peripheral and Central Nervous Systems, the Immune System is the most complex of somatic systems in higher animals. This complexity manifests itself at many levels from the molecular to that of the whole organism. Much insight into this confounding complexity can be gained through computational simulation. Such simulations range in application from epitope prediction through to the modelling of vaccination strategies. In this review, we evaluate selectively various key applications relevant to computational vaccinology: these include technique that operates at different scale that is, from molecular to organisms and even to population level.

The enantioselective population pharmacokinetics of intravenous ketorolac in children using a stereoselective assay suitable for microanalysis

Objective: To describe the effect of age and body size on enantiomer selective pharmacokinetic (PK) of intravenous ketorolac in children using a microanalytical assay. Methods: Blood samples were obtained at 0, 15 and 30 min and at 1, 2, 4, 6, 8 and 12 h after a weight-dependent dose of ketorolac. Enantiomer concentration was measured using a liquid chromatography tandem mass spectrometry method. Non-linear mixed-effect modelling was used to assess PK parameters. Key findings: Data from 11 children (1.7–15.6 years, weight 10.7–67.4 kg) were best described by a two-compartment model for R(+), S(−) and racemic ketorolac. Only weight (WT) significantly improved the goodness of fit. The final population models were CL = 1.5 × (WT/46)0.75, V1 = 8.2 × (WT/46), Q = 3.4 × (WT/46)0.75, V2 = 7.9 × (WT/46), CL = 2.98 × (WT/46), V1 = 13.2 × (WT/46), Q = 2.8 × (WT/46)0.75, V2 = 51.5 × (WT/46), and CL = 1.1 × (WT/46)0.75, V1 = 4.9 × (WT/46), Q = 1.7 × (WT/46)0.75 and V2 = 6.3 × (WT/46)for R(+), S(−) and racemic ketorolac. Conclusions: Only body weight influenced the PK parameters for R(+) and S(−) ketorolac. Using allometric size scaling significantly affected the clearances (CL, Q) and volumes of distribution (V1, V2).

Evaluation of climate change scenarios based on aquatic food web modelling

In the years 2004 and 2005 we collected samples of phytoplankton, zooplankton and macroinvertebrates in an artificial small pond in Budapest. We set up a simulation model predicting the abundance of the cyclopoids, Eudiaptomus zachariasi and Ischnura pumilio by considering only temperature as it affects the abundance of population of the previous day. Phytoplankton abundance was simulated by considering not only temperature, but the abundance of the three mentioned groups. This discrete-deterministic model could generate similar patterns like the observed one and testing it on historical data was successful. However, because the model was overpredicting the abundances of Ischnura pumilio and Cyclopoida at the end of the year, these results were not considered. Running the model with the data series of climate change scenarios, we had an opportunity to predict the individual numbers for the period around 2050. If the model is run with the data series of the two scenarios UKHI and UKLO, which predict drastic global warming, then we can observe a decrease in abundance and shift in the date of the maximum abundance occurring (excluding Ischnura pumilio, where the maximum abundance increases and it occurs later), whereas under unchanged climatic conditions (BASE scenario) the change in abundance is negligible. According to the scenarios GFDL 2535, GFDL 5564 and UKTR, a transition could be noticed.