Personalized modeling for drug concentration prediction using Support Vector Machine

Building a personalized model to describe the drug concentration inside the human body for each patient is highly important to the clinical practice and demanding to the modeling tools. Instead of using traditional explicit methods, in this paper we propose a machine learning approach to describe the relation between the drug concentration and patients' features. Machine learning has been largely applied to analyze data in various domains, but it is still new to personalized medicine, especially dose individualization. We focus mainly on the prediction of the drug concentrations as well as the analysis of different features' influence. Models are built based on Support Vector Machine and the prediction results are compared with the traditional analytical models.

Prediction of infant drug exposure through breastfeeding: population PK modeling and simulation of fluoxetine

BACKGROUND: Risks of significant infant drug exposurethrough breastmilk are poorly defined for many drugs, and largescalepopulation data are lacking. We used population pharmacokinetics(PK) modeling to predict fluoxetine exposure levels ofinfants via mother's milk in a simulated population of 1000 motherinfantpairs.METHODS: Using our original data on fluoxetine PK of 25breastfeeding women, a population PK model was developed withNONMEM and parameters, including milk concentrations, wereestimated. An exponential distribution model was used to account forindividual variation. Simulation random and distribution-constrainedassignment of doses, dosing time, feeding intervals and milk volumewas conducted to generate 1000 mother-infant pairs with characteristicssuch as the steady-state serum concentrations (Css) and infantdose relative to the maternal weight-adjusted dose (relative infantdose: RID). Full bioavailability and a conservative point estimate of1-month-old infant CYP2D6 activity to be 20% of the adult value(adjusted by weigth) according to a recent study, were assumed forinfant Css calculations.RESULTS: A linear 2-compartment model was selected as thebest model. Derived parameters, including milk-to-plasma ratios(mean: 0.66; SD: 0.34; range, 0 - 1.1) were consistent with the valuesreported in the literature. The estimated RID was below 10% in >95%of infants. The model predicted median infant-mother Css ratio was0.096 (range 0.035 - 0.25); literature reported mean was 0.07 (range0-0.59). Moreover, the predicted incidence of infant-mother Css ratioof >0.2 was less than 1%.CONCLUSION: Our in silico model prediction is consistent withclinical observations, suggesting that substantial systemic fluoxetineexposure in infants through human milk is rare, but further analysisshould include active metabolites. Our approach may be valid forother drugs. [supported by CIHR and Swiss National Science Foundation(SNSF)]

The challenge of modeling nuclear receptor regulatory networks in mammalian cells.

Nuclear receptors are a major component of signal transduction in animals. They mediate the regulatory activities of many hormones, nutrients and metabolites on the homeostasis and physiology of cells and tissues. It is of high interest to model the corresponding regulatory networks. While molecular and cell biology studies of individual promoters have provided important mechanistic insight, a more complex picture is emerging from genome-wide studies. The regulatory circuitry of nuclear receptor regulated gene expression networks, and their response to cellular signaling, appear highly dynamic, and involve long as well as short range chromatin interactions. We review how progress in understanding the kinetics and regulation of cofactor recruitment, and the development of new genomic methods, provide opportunities but also a major challenge for modeling nuclear receptor mediated regulatory networks.

Aerodynamic drag modeling of alpine skiers performing giant slalom turns.

PURPOSE: Aerodynamic drag plays an important role in performance for athletes practicing sports that involve high-velocity motions. In giant slalom, the skier is continuously changing his/her body posture, and this affects the energy dissipated in aerodynamic drag. It is therefore important to quantify this energy to understand the dynamic behavior of the skier. The aims of this study were to model the aerodynamic drag of alpine skiers in giant slalom simulated conditions and to apply these models in a field experiment to estimate energy dissipated through aerodynamic drag. METHODS: The aerodynamic characteristics of 15 recreational male and female skiers were measured in a wind tunnel while holding nine different skiing-specific postures. The drag and the frontal area were recorded simultaneously for each posture. Four generalized and two individualized models of the drag coefficient were built, using different sets of parameters. These models were subsequently applied in a field study designed to compare the aerodynamic energy losses between a dynamic and a compact skiing technique. RESULTS: The generalized models estimated aerodynamic drag with an accuracy of between 11.00% and 14.28%, and the individualized models estimated aerodynamic drag with an accuracy between 4.52% and 5.30%. The individualized model used for the field study showed that using a dynamic technique led to 10% more aerodynamic drag energy loss than using a compact technique. DISCUSSION: The individualized models were capable of discriminating different techniques performed by advanced skiers and seemed more accurate than the generalized models. The models presented here offer a simple yet accurate method to estimate the aerodynamic drag acting upon alpine skiers while rapidly moving through the range of positions typical to turning technique.

Stoneley wave modeling in heterogeneous porous media with viscous pore fluids

We implemented Biot-type porous wave equations in a pseudo-spectral numerical modeling algorithm for the simulation of Stoneley waves in porous media. Fourier and Chebyshev methods are used to compute the spatial derivatives along the horizontal and vertical directions, respectively. To prevent from overly short time steps due to the small grid spacing at the top and bottom of the model as a consequence of the Chebyshev operator, the mesh is stretched in the vertical direction. As a large benefit, the Chebyshev operator allows for an explicit treatment of interfaces. Boundary conditions can be implemented with a characteristics approach. The characteristic variables are evaluated at zero viscosity. We use this approach to model seismic wave propagation at the interface between a fluid and a porous medium. Each medium is represented by a different mesh and the two meshes are connected through the above described characteristics domain-decomposition method. We show an experiment for sealed pore boundary conditions, where we first compare the numerical solution to an analytical solution. We then show the influence of heterogeneity and viscosity of the pore fluid on the propagation of the Stoneley wave and surface waves in general.

GeoKernels: Modeling of Spatial Data on GeoManifolds

This paper presents a review of methodology for semi-supervised modeling with kernel methods, when the manifold assumption is guaranteed to be satisfied. It concerns environmental data modeling on natural manifolds, such as complex topographies of the mountainous regions, where environmental processes are highly influenced by the relief. These relations, possibly regionalized and nonlinear, can be modeled from data with machine learning using the digital elevation models in semi-supervised kernel methods. The range of the tools and methodological issues discussed in the study includes feature selection and semisupervised Support Vector algorithms. The real case study devoted to data-driven modeling of meteorological fields illustrates the discussed approach.

Making sense of AMPA receptor trafficking by modeling molecular mechanisms of synaptic plasticity.

Synaptic plasticity involves a complex molecular machinery with various protein interactions but it is not yet clear how its components give rise to the different aspects of synaptic plasticity. Here we ask whether it is possible to mathematically model synaptic plasticity by making use of known substances only. We present a model of a multistable biochemical reaction system and use it to simulate the plasticity of synaptic transmission in long-term potentiation (LTP) or long-term depression (LTD) after repeated excitation of the synapse. According to our model, we can distinguish between two phases: first, a "viscosity" phase after the first excitation, the effects of which like the activation of NMDA receptors and CaMKII fade out in the absence of further excitations. Second, a "plasticity" phase actuated by an identical subsequent excitation that follows after a short time interval and causes the temporarily altered concentrations of AMPA subunits in the postsynaptic membrane to be stabilized. We show that positive feedback is the crucial element in the core chemical reaction, i.e. the activation of the short-tail AMPA subunit by NEM-sensitive factor, which allows generating multiple stable equilibria. Three stable equilibria are related to LTP, LTD and a third unfixed state called ACTIVE. Our mathematical approach shows that modeling synaptic multistability is possible by making use of known substances like NMDA and AMPA receptors, NEM-sensitive factor, glutamate, CaMKII and brain-derived neurotrophic factor. Furthermore, we could show that the heteromeric combination of short- and long-tail AMPA receptor subunits fulfills the function of a memory tag.

Modeling the forensic two-trace problem with Bayesian networks

The forensic two-trace problem is a perplexing inference problem introduced by Evett (J Forensic Sci Soc 27:375-381, 1987). Different possible ways of wording the competing pair of propositions (i.e., one proposition advanced by the prosecution and one proposition advanced by the defence) led to different quantifications of the value of the evidence (Meester and Sjerps in Biometrics 59:727-732, 2003). Here, we re-examine this scenario with the aim of clarifying the interrelationships that exist between the different solutions, and in this way, produce a global vision of the problem. We propose to investigate the different expressions for evaluating the value of the evidence by using a graphical approach, i.e. Bayesian networks, to model the rationale behind each of the proposed solutions and the assumptions made on the unknown parameters in this problem.

Further insights on the French WISC-IV factor structure through Bayesian structural equation modeling

The interpretation of the Wechsler Intelligence Scale for Children-Fourth Edition (WISC-IV) is based on a 4-factor model, which is only partially compatible with the mainstream Cattell-Horn-Carroll (CHC) model of intelligence measurement. The structure of cognitive batteries is frequently analyzed via exploratory factor analysis and/or confirmatory factor analysis. With classical confirmatory factor analysis, almost all crossloadings between latent variables and measures are fixed to zero in order to allow the model to be identified. However, inappropriate zero cross-loadings can contribute to poor model fit, distorted factors, and biased factor correlations; most important, they do not necessarily faithfully reflect theory. To deal with these methodological and theoretical limitations, we used a new statistical approach, Bayesian structural equation modeling (BSEM), among a sample of 249 French-speaking Swiss children (8-12 years). With BSEM, zero-fixed cross-loadings between latent variables and measures are replaced by approximate zeros, based on informative, small-variance priors. Results indicated that a direct hierarchical CHC-based model with 5 factors plus a general intelligence factor better represented the structure of the WISC-IV than did the 4-factor structure and the higher order models. Because a direct hierarchical CHC model was more adequate, it was concluded that the general factor should be considered as a breadth rather than a superordinate factor. Because it was possible for us to estimate the influence of each of the latent variables on the 15 subtest scores, BSEM allowed improvement of the understanding of the structure of intelligence tests and the clinical interpretation of the subtest scores.

Expanding molecular modeling and design tools to non-natural sidechains.

Protein-protein interactions encode the wiring diagram of cellular signaling pathways and their deregulations underlie a variety of diseases, such as cancer. Inhibiting protein-protein interactions with peptide derivatives is a promising way to develop new biological and therapeutic tools. Here, we develop a general framework to computationally handle hundreds of non-natural amino acid sidechains and predict the effect of inserting them into peptides or proteins. We first generate all structural files (pdb and mol2), as well as parameters and topologies for standard molecular mechanics software (CHARMM and Gromacs). Accurate predictions of rotamer probabilities are provided using a novel combined knowledge and physics based strategy. Non-natural sidechains are useful to increase peptide ligand binding affinity. Our results obtained on non-natural mutants of a BCL9 peptide targeting beta-catenin show very good correlation between predicted and experimental binding free-energies, indicating that such predictions can be used to design new inhibitors. Data generated in this work, as well as PyMOL and UCSF Chimera plug-ins for user-friendly visualization of non-natural sidechains, are all available at http://www.swisssidechain.ch. Our results enable researchers to rapidly and efficiently work with hundreds of non-natural sidechains.

Comment on recent modeling studies of astrocyte-neuron metabolic interactions.

Recent years have seen a surge in mathematical modeling of the various aspects of neuron-astrocyte interactions, and the field of brain energy metabolism is no exception in that regard. Despite the advent of biophysical models in the field, the long-lasting debate on the role of lactate in brain energy metabolism is still unresolved. Quite the contrary, it has been ported to the world of differential equations. Here, we summarize the present state of this discussion from the modeler's point of view and bring some crucial points to the attention of the non-mathematically proficient reader.

Modeling complex wells with the multi-scale finite-volume method

In this paper, an extension of the multi-scale finite-volume (MSFV) method is devised, which allows to Simulate flow and transport in reservoirs with complex well configurations. The new framework fits nicely into the data Structure of the original MSFV method,and has the important property that large patches covering the whole well are not required. For each well. an additional degree of freedom is introduced. While the treatment of pressure-constraint wells is trivial (the well-bore reference pressure is explicitly specified), additional equations have to be solved to obtain the unknown well-bore pressure of rate-constraint wells. Numerical Simulations of test cases with multiple complex wells demonstrate the ability of the new algorithm to capture the interference between the various wells and the reservoir accurately. (c) 2008 Elsevier Inc. All rights reserved.

Carbon and oxygen isotope study of hydrothermal carbonates in the zinc-lead deposits of the San Vicente district, central Peru: A quantitative modeling on mixing processes and CO2 degassing

Carbon and oxygen isotope studies of the host and gangue carbonates of Mississippi Valley-type zinc-lead deposits in the San Vicente District hosted in the Upper Triassic to Lower Jurassic dolostones of the Pucara basin (central Peru) were used to constrain models of the ore formation. A mixing model between an incoming hot saline slightly acidic radiogenic (Pb, Sr) fluid and the native formation water explains the overall isotopic variation (delta(13)C = - 11.5 to + 2.5 parts per thousand relative to PDB and delta(18)O = + 18.0 to + 24.3 parts per thousand relative to SMOW) of the carbonate generations. The dolomites formed during the main ore stage show a narrower range (delta(13)C = - 0.1 to + 1.7 parts per thousand and delta(18)O = + 18.7 to + 23.4 parts per thousand) which is explained by exchange between the mineralizing fluids and the host carbonates combined with changes in temperature and pressure. This model of fluid-rock interaction explains the pervasive alteration of the host dolomite I and precipitation of sphalerite I. The open-space filling hydrothermal white sparry dolomite and the coexisting sphalerite II formed by prolonged fluid-host dolomite interaction and limited CO2 degassing. Late void-filling dolomite III (or calcite) and the associated sphalerite III formed as the consequence of CO2 degassing and concomitant pH increase of a slightly acidic ore fluid. Widespread brecciation is associated to CO2 outgassing. Consequently, pressure variability plays a major role in the ore precipitation during the late hydrothermal events in San Vicente. The presence of native sulfur associated with extremely carbon-light calcites replacing evaporitic sulfates (e.g., delta(13)C = - 11.5 parts per thousand), altered native organic matter and heavier hydrothermal bitumen (from - 27.0 to - 23.0 parts per thousand delta(13)C) points to thermochemical reduction of sulfate and/or thiosulfate. The delta(13)C- and delta(18)O-values of the altered host dolostone and hydrothermal carbonates, and the carbon isotope composition of the associated organic matter show a strong regional homogeneity. These results coupled with the strong mineralogical and petrographic similarities of the different MVT occurrences perhaps reflects the fact that the mineralizing processes were similar in the whole San Vicente belt, suggesting the existence of a common regional mineralizing hydrothermal system with interconnected plumbing.