Automating assistance for safety critical decisions

Computer systems are increasingly being introduced to assist in decision making, including hazardous decision making. To ensure effective assistance, decision procedures should be theoretically sound, flexible in operation (particularly in unpredictable environments) and effectively accountable to human supervisors and auditors. Strengths and weaknesses of classical statistical decision models are discussed from these perspectives. It is argued that more can be learned from human decision behaviour than has traditionally been assumed, and this motivates the concept of a symbolic decision procedure (SDP). The SDP is defined, described in terms of first-order (predicate) logic, and its use illustrated in a decision support system for medicine. We point out that the classical numerical decision procedure is a special case of a generalized symbolic procedure, and discuss the potential for rigorous formalization of the latter. We conclude that symbolic decision procedures may meet requirements for assisting human operators in hazardous situations more satisfactorily than classical decision procedures.     

Experimental optimization of a real time fed-batch fermentation process using Markov decision process

This article describes a methodology that implements a Markov decision process (MDP) optimization technique in a real time fed-batch experiment. Biological systems can be better modeled under the stochastic framework and MDP is shown to be a suitable technique for their optimization. A nonlinear input/output model is used to calculate the probability transitions. All elements of the MDP are identified according to physical parameters. Finally, this study compares the results obtained when optimizing ethanol production using the infinite horizon problem, with total expected discount policy, to previous experimental results aimed at optimizing ethanol production using a recombinant Escherichia coli fed-batch cultivation. (c) 1997 John Wiley & Sons, Inc. Biotechnol Bioeng 55: 317-327, 1997.     

What Is the Optimal Threshold at Which to Recommend Breast Biopsy?

Background

A 2% threshold, traditionally used as a level above which breast biopsy recommended, has been generalized to all patients from several specific situations analyzed in the literature. We use a sequential decision analytic model considering clinical and mammography features to determine the optimal general threshold for image guided breast biopsy and the sensitivity of this threshold to variation of these features.

Methodology/Principal Findings

We built a decision analytical model called a Markov Decision Process (MDP) model, which determines the optimal threshold of breast cancer risk to perform breast biopsy in order to maximize a patient’s total quality-adjusted life years (QALYs). The optimal biopsy threshold is determined based on a patient’s probability of breast cancer estimated by a logistic regression model (LRM) which uses demographic risk factors (age, family history, and hormone use) and mammographic findings (described using the established lexicon–BI-RADS). We estimate the MDP model''s parameters using SEER data (prevalence of invasive vs. in situ disease, stage at diagnosis, and survival), US life tables (all cause mortality), and the medical literature (biopsy disutility and treatment efficacy) to determine the optimal “base case” risk threshold for breast biopsy and perform sensitivity analysis. The base case MDP model reveals that 2% is the optimal threshold for breast biopsy for patients between 42 and 75 however the thresholds below age 42 is lower (1%) and above age 75 is higher (range of 3–5%). Our sensitivity analysis reveals that the optimal biopsy threshold varies most notably with changes in age and disutility of biopsy.

Conclusions/Significance

Our MDP model validates the 2% threshold currently used for biopsy but shows this optimal threshold varies substantially with patient age and biopsy disutility.     

Timing of testing and treatment for asymptomatic diseases

Many papers in the medical literature analyze the cost-effectiveness of screening for diseases by comparing a limited number of a priori testing policies under estimated problem parameters. However, this may be insufficient to determine the best timing of the tests or incorporate changes over time. In this paper, we develop and solve a Markov Decision Process (MDP) model for a simple class of asymptomatic diseases in order to provide the building blocks for analysis of a more general class of diseases. We provide a computationally efficient method for determining a cost-effective dynamic intervention strategy that takes into account (i) the results of the previous test for each individual and (ii) the change in the individual’s behavior based on awareness of the disease. We demonstrate the usefulness of the approach by applying the results to screening decisions for Hepatitis C (HCV) using medical data, and compare our findings to current HCV screening recommendations.     

Hierarchical decision‐making balances current and future reproductive success

Parental decisions in animals are often context‐dependent and shaped by fitness trade‐offs between parents and offspring. For example, the selection of breeding habitats can considerably impact the fitness of both offspring and parents, and therefore, parents should carefully weigh the costs and benefits of available options for their current and future reproductive success. Here, we show that resource‐use preferences are shaped by a trade‐off between parental effort and offspring safety in a tadpole‐transporting frog. In a large‐scale in situ experiment, we investigated decision strategies across an entire population of poison frogs that distribute their tadpoles across multiple water bodies. Pool use followed a dynamic and sequential selection process, and transportation became more efficient over time. Our results point to a complex suite of environmental variables that are considered during offspring deposition, which necessitates a highly dynamic and flexible decision‐making process in tadpole‐transporting frogs.     

Minimum distance probability discriminant analysis for mixed variables

Minimum distance probability (MDP) is a robust discriminant algorithm based on a distance function. In this article, we generalize the use of MDP to the case of mixed (continuous and categorical) variables by means of the individual-score (IS) distance. This distance assumes an underlying parametric model and is based on the score transformation of the data. We have adapted it to the usual case of ignoring the distribution of the whole set of observed variables, but assuming that some knowledge about the marginal distributions is available. Finally, MDP with IS distance (IS-MDP) is compared with other discriminant methods (including those designed for mixed data) in several examples and simulations. IS-MDP is shown to be the most efficient method according the leave-one-out criterion.     

Stochastic search variable selection based on two mixture components and continuous‐scale weighting

Stochastic search variable selection (SSVS) is a Bayesian variable selection method that employs covariate‐specific discrete indicator variables to select which covariates (e.g., molecular markers) are included in or excluded from the model. We present a new variant of SSVS where, instead of discrete indicator variables, we use continuous‐scale weighting variables (which take also values between zero and one) to select covariates into the model. The improved model performance is shown and compared to standard SSVS using simulated and real quantitative trait locus mapping datasets. The decision making to decide phenotype‐genotype associations in our SSVS variant is based on median of posterior distribution or using Bayes factors. We also show here that by using continuous‐scale weighting variables it is possible to improve mixing properties of Markov chain Monte Carlo sampling substantially compared to standard SSVS. Also, the separation of association signals and nonsignals (control of noise level) seems to be more efficient compared to the standard SSVS. Thus, the novel method provides efficient new framework for SSVS analysis that additionally provides whole posterior distribution for pseudo‐indicators which means more information and may help in decision making.     

Which States Matter? An Application of an Intelligent Discretization Method to Solve a Continuous POMDP in Conservation Biology

When managing populations of threatened species, conservation managers seek to make the best conservation decisions to avoid extinction. Making the best decision is difficult because the true population size and the effects of management are uncertain. Managers must allocate limited resources between actively protecting the species and monitoring. Resources spent on monitoring reduce expenditure on management that could be used to directly improve species persistence. However monitoring may prevent sub-optimal management actions being taken as a result of observation error. Partially observable Markov decision processes (POMDPs) can optimize management for populations with partial detectability, but the solution methods can only be applied when there are few discrete states. We use the Continuous U-Tree (CU-Tree) algorithm to discretely represent a continuous state space by using only the states that are necessary to maintain an optimal management policy. We exploit the compact discretization created by CU-Tree to solve a POMDP on the original continuous state space. We apply our method to a population of sea otters and explore the trade-off between allocating resources to management and monitoring. We show that accurately discovering the population size is less important than management for the long term survival of our otter population.     

Bayesian latent multi‐state modeling for nonequidistant longitudinal electronic health records

Large amounts of longitudinal health records are now available for dynamic monitoring of the underlying processes governing the observations. However, the health status progression across time is not typically observed directly: records are observed only when a subject interacts with the system, yielding irregular and often sparse observations. This suggests that the observed trajectories should be modeled via a latent continuous‐time process potentially as a function of time‐varying covariates. We develop a continuous‐time hidden Markov model to analyze longitudinal data accounting for irregular visits and different types of observations. By employing a specific missing data likelihood formulation, we can construct an efficient computational algorithm. We focus on Bayesian inference for the model: this is facilitated by an expectation‐maximization algorithm and Markov chain Monte Carlo methods. Simulation studies demonstrate that these approaches can be implemented efficiently for large data sets in a fully Bayesian setting. We apply this model to a real cohort where patients suffer from chronic obstructive pulmonary disease with the outcome being the number of drugs taken, using health care utilization indicators and patient characteristics as covariates.     

Ways to test stochastic dynamic programming models empirically

Stochastic dynamic programming (SDP) models are widely used to predict optimal behavioural and life history strategies. We discuss a diversity of ways to test SDP models empirically, taking as our main illustration a model of the daily singing routine of birds. One approach to verification is to quantify model parameters, but most SDP models are schematic. Because predictions are therefore qualitative, testing several predictions is desirable. How state determines behaviour (the policy) is a central prediction that should be examined directly if both state and behaviour are measurable. Complementary predictions concern how behaviour and state change through time, but information is discarded by considering behaviour rather than state, by looking only at average state rather than its distribution, and by not following individuals. We identify the various circumstances in which an individual's state/behaviour at one time is correlated with its state/behaviour at a later time. When there are several state variables the relationships between them may be informative. Often model parameters represent environmental conditions that can also be viewed as state variables. Experimental manipulation of the environment has several advantages as a test, but a problem is uncertainty over how much the organism's policy will adjust. As an example we allow birds to use different assumptions about how well past weather predicts future weather. We advocate mirroring planned empirical investigations on the computer to investigate which manipulations and predictions will best test a model. Copyright 2000 The Association for the Study of Animal Behaviour.     

The Bacillus subtilis cannibalism toxin SDP collapses the proton motive force and induces autolysis

Bacillus subtilis SDP is a peptide toxin that kills cells outside the biofilm to support continued growth. We show that purified SDP acts like endogenously produced SDP; it delays sporulation, and the SdpI immunity protein confers SDP resistance. SDP kills a variety of Gram‐positive bacteria in the phylum Firmicutes, as well as Escherichia coli with a compromised outer membrane, suggesting it participates in defence of the B. subtilis biofilm against Gram‐positive bacteria as well as cannibalism. Fluorescence microscopy reveals that the effect of SDP on cells differs from that of nisin, nigericin, valinomycin and vancomycin‐KCl, but resembles that of CCCP, DNP and azide. Indeed, SDP rapidly collapses the PMF as measured by fluorometry and flow cytometry, which triggers the slower process of autolysis. This secondary consequence of SDP treatment is not required for cell death since the autolysin‐defective lytC, lytD, lytE, lytF strain fails to be lysed but is nevertheless killed by SDP. Collapsing the PMF is an ideal mechanism for a toxin involved in cannibalism and biofilm defence, since this would incapacitate neighbouring cells by inhibiting motility and secretion of proteins and toxins. It would also induce autolysis in many Gram‐positive species, thereby releasing nutrients that promote biofilm growth.     

Land‐use Change Dynamics,Soil Type and Species Forming Mono‐dominant Patches: the Case of Pteridium aquilinum in a Neotropical Rain Forest Region

Deforestation and agricultural land degradation in tropical regions can create conditions for growth of perennial plant species forming mono‐dominated patches (MDP). Such species might limit forest regeneration, and their proliferation forces the abandonment of fields and subsequent deforestation to establish new fields. Therefore, identifying factors fostering MDP species is critical for biodiversity conservation in human‐modified landscapes. Here, we propose a conceptual framework to identify such factors and apply it to the case of Pteridium aquilinum (bracken fern), a light‐demanding species, tolerant of low soil fertility and fire. We hypothesize that bracken proliferation is promoted by land‐use changes that increase light availability, especially in sites with low soil fertility and land uses involving fire. We assessed this idea using agricultural fields in southeastern Mexico with different land‐use change histories and quantifying prevalence and cover of bracken. Five different land‐use change histories resulted from transitions among forest, crop, pasture, and fallow field stages. Of the 133 fields sampled, 71 percent had P. aquilinum; regression tree analysis indicated that 65 percent of inter‐field variation in prevalence and 90 percent in cover was explained by land‐use change history and soil type. Maximum prevalence, cover, and rates of increase in bracken were found on fields with low fertility sandy/clay soils, which had been used for crops and pasture, were frequently burned, and had high levels of light. Fields on fertile alluvial soil never used for pasture were bracken‐free. Agriculture promoting high light environments on less fertile soils is a major cause of bracken proliferation and likely that of other MDP species.     

Genomewide association analysis in diverse inbred mice: power and population structure

The discovery of quantitative trait loci (QTL) in model organisms has relied heavily on the ability to perform controlled breeding to generate genotypic and phenotypic diversity. Recently, we and others have demonstrated the use of an existing set of diverse inbred mice (referred to here as the mouse diversity panel, MDP) as a QTL mapping population. The use of the MDP population has many advantages relative to traditional F(2) mapping populations, including increased phenotypic diversity, a higher recombination frequency, and the ability to collect genotype and phenotype data in community databases. However, these methods are complicated by population structure inherent in the MDP and the lack of an analytical framework to assess statistical power. To address these issues, we measured gene expression levels in hypothalamus across the MDP. We then mapped these phenotypes as quantitative traits with our association algorithm, resulting in a large set of expression QTL (eQTL). We utilized these eQTL, and specifically cis-eQTL, to develop a novel nonparametric method for association analysis in structured populations like the MDP. These eQTL data confirmed that the MDP is a suitable mapping population for QTL discovery and that eQTL results can serve as a gold standard for relative measures of statistical power.     

Time Horizon and Dominance in Dynamic Life Cycle Assessment

Temporal aspects have traditionally not been recognized adequately in life cycle assessment (LCA). The dynamic LCA model recently proposed offers a significant step forward in the dynamic assessment of global warming impacts. The results obtained with dynamic LCA are highly sensitive to the choice of a time horizon. Therefore, decision making between alternative systems can be critical because conclusions are dependent on the specific time horizon. In this article, we develop a decision‐making methodology based on the concept of time dominance. We introduce instantaneous and cumulative time dominance criteria to the dynamic LCA context and argue why the dominance of an alternative should also imply preference. Our approach allows for the rejection of certain alternatives without the determination of a specific time horizon. The number of decision‐relevant alternatives can thereby be reduced and the decision problem facilitated. We demonstrate our methodology by means of a case study of end‐of‐life alternatives for a wooden chair derived from the original authors of dynamic LCA and discuss the implications and limitations of the approach. The methodology based on time dominance criteria is supplementary to the dynamic LCA model, but does not substitute it. The overall value of this article stretches beyond LCA onto more general assessments of global warming, for example, in policy where the choice of a time horizon is equally significant.     

Markov and Semi‐Markov Switching Linear Mixed Models Used to Identify Forest Tree Growth Components

Summary Tree growth is assumed to be mainly the result of three components: (i) an endogenous component assumed to be structured as a succession of roughly stationary phases separated by marked change points that are asynchronous among individuals, (ii) a time‐varying environmental component assumed to take the form of synchronous fluctuations among individuals, and (iii) an individual component corresponding mainly to the local environment of each tree. To identify and characterize these three components, we propose to use semi‐Markov switching linear mixed models, i.e., models that combine linear mixed models in a semi‐Markovian manner. The underlying semi‐Markov chain represents the succession of growth phases and their lengths (endogenous component) whereas the linear mixed models attached to each state of the underlying semi‐Markov chain represent—in the corresponding growth phase—both the influence of time‐varying climatic covariates (environmental component) as fixed effects, and interindividual heterogeneity (individual component) as random effects. In this article, we address the estimation of Markov and semi‐Markov switching linear mixed models in a general framework. We propose a Monte Carlo expectation–maximization like algorithm whose iterations decompose into three steps: (i) sampling of state sequences given random effects, (ii) prediction of random effects given state sequences, and (iii) maximization. The proposed statistical modeling approach is illustrated by the analysis of successive annual shoots along Corsican pine trunks influenced by climatic covariates.     

Peptidoglycan in combination with muramyldipeptide synergistically induces an interleukin‐10‐dependent T helper 2‐dominant immune response

In this study, peptidoglycan (PEG) from Staphylococcus aureus‐stimulated, but not muramyldipeptide (MDP)‐stimulated, Langerhans cells (LCs) induced a dose‐dependent Th2‐prone immune response. However, when LCs were stimulated with PEG in combination with MDP, the strength of Th2 immune responses was synergistically augmented by MDP. Furthermore, it was found that production of IL‐10, but not of IL‐12 p40, by PEG‐stimulated LCs was also enhanced in the presence of MDP. These results suggest that MDP enhances Th2 cell development through up‐regulation of IL‐10 production from PEG‐stimulated LCs, increase the importance of S. aureus colonization in patients with atopic dermatitis.     

Structural insights into the MDP binding and CARD–CARD interaction in zebrafish (Danio rerio) NOD2: a molecular dynamics approach

Nucleotide binding and oligomerization domain (NOD2) is a key component of innate immunity that is highly specific for muramyl dipeptide (MDP)—a peptidoglycan component of bacterial cell wall. MDP recognition by NOD2–leucine rich repeat (LRR) domain activates NF‐κB signaling through a protein–protein interaction between caspase activating and recruitment domains (CARDs) of NOD2 and downstream receptor interacting and activating protein kinase 2 (RIP2). Due to the lack of crystal/NMR structures, MDP recognition and CARD–CARD interaction are poorly understood. Herein, we have predicted the probable MDP and CARD–CARD binding surfaces in zebrafish NOD2 (zNOD2) using various in silico methodologies. The results show that the conserved residues Phe819, Phe871, Trp875, Trp929, Trp899, and Arg845 located at the concave face of zNOD2–LRR confer MDP recognition by hydrophobic and hydrogen bond (H‐bond) interactions. Molecular dynamics simulations reveal a stable association between the electropositive surface on zNOD2–CARDa and the electronegative surface on zRIP2–CARD reinforced mostly by H‐bonds and electrostatic interactions. Importantly, a 3.5 Å salt bridge is observed between Arg60 of zNOD2–CARDa and Asp494 of zRIP2–CARD. Arg11 and Lys53 of zNOD2–CARDa and Ser498 and Glu508 of zRIP2–CARD are critical residues for CARD–CARD interaction and NOD2 signaling. The 2.7 Å H‐bond between Lys104 of the linker and Glu508 of zRIP2–CARD suggests a possible role of the linker for shaping CARD–CARD interaction. These findings are consistent with existing mutagenesis data. We provide first insight into MDP recognition and CARD–CARD interaction in the zebrafish that will be useful to understand the molecular basis of NOD signaling in a broader perspective. Copyright © 2014 John Wiley & Sons, Ltd.     

Biogenic Carbon and Temporary Storage Addressed with Dynamic Life Cycle Assessment

A growing tendency in policy making and carbon footprint estimation gives value to temporary carbon storage in biomass products or to delayed greenhouse gas (GHG) emissions. Some life cycle‐based methods, such as the British publicly available specification (PAS) 2050 or the recently published European Commission's International Reference Life Cycle Data System (ILCD) Handbook, address this issue. This article shows the importance of consistent consideration of biogenic carbon and timing of GHG emissions in life cycle assessment (LCA) and carbon footprint analysis. We use a fictitious case study assessing the life cycle of a wooden chair for four end‐of‐life scenarios to compare different approaches: traditional LCA with and without consideration of biogenic carbon, the PAS 2050 and ILCD Handbook methods, and a dynamic LCA approach. Reliable results require accounting for the timing of every GHG emission, including biogenic carbon flows, as soon as a benefit is given for temporarily storing carbon or delaying GHG emissions. The conclusions of a comparative LCA can change depending on the time horizon chosen for the analysis. The dynamic LCA approach allows for a consistent assessment of the impact, through time, of all GHG emissions (positive) and sequestration (negative). The dynamic LCA is also a valuable approach for decision makers who have to understand the sensitivity of the conclusions to the chosen time horizon.