Environmental data mining and modeling based on machine learning algorithms and geostatistics

The paper presents some contemporary approaches to spatial environmental data analysis. The main topics are concentrated on the decision-oriented problems of environmental spatial data mining and modeling: valorization and representativity of data with the help of exploratory data analysis, spatial predictions, probabilistic and risk mapping, development and application of conditional stochastic simulation models. The innovative part of the paper presents integrated/hybrid model-machine learning (ML) residuals sequential simulations-MLRSS. The models are based on multilayer perceptron and support vector regression ML algorithms used for modeling long-range spatial trends and sequential simulations of the residuals. NIL algorithms deliver non-linear solution for the spatial non-stationary problems, which are difficult for geostatistical approach. Geostatistical tools (variography) are used to characterize performance of ML algorithms, by analyzing quality and quantity of the spatially structured information extracted from data with ML algorithms. Sequential simulations provide efficient assessment of uncertainty and spatial variability. Case study from the Chernobyl fallouts illustrates the performance of the proposed model. It is shown that probability mapping, provided by the combination of ML data driven and geostatistical model based approaches, can be efficiently used in decision-making process. (C) 2003 Elsevier Ltd. All rights reserved.

Modeling sequencing errors by combining Hidden Markov models.

Among the largest resources for biological sequence data is the large amount of expressed sequence tags (ESTs) available in public and proprietary databases. ESTs provide information on transcripts but for technical reasons they often contain sequencing errors. Therefore, when analyzing EST sequences computationally, such errors must be taken into account. Earlier attempts to model error prone coding regions have shown good performance in detecting and predicting these while correcting sequencing errors using codon usage frequencies. In the research presented here, we improve the detection of translation start and stop sites by integrating a more complex mRNA model with codon usage bias based error correction into one hidden Markov model (HMM), thus generalizing this error correction approach to more complex HMMs. We show that our method maintains the performance in detecting coding sequences.

Fusion of Multi-Atlas Segmentations with Spatial Distribution Modeling

In recent years, multi-atlas fusion methods have gainedsignificant attention in medical image segmentation. Inthis paper, we propose a general Markov Random Field(MRF) based framework that can perform edge-preservingsmoothing of the labels at the time of fusing the labelsitself. More specifically, we formulate the label fusionproblem with MRF-based neighborhood priors, as an energyminimization problem containing a unary data term and apairwise smoothness term. We present how the existingfusion methods like majority voting, global weightedvoting and local weighted voting methods can be reframedto profit from the proposed framework, for generatingmore accurate segmentations as well as more contiguoussegmentations by getting rid of holes and islands. Theproposed framework is evaluated for segmenting lymphnodes in 3D head and neck CT images. A comparison ofvarious fusion algorithms is also presented.

Evaluating the influence of spatial uncertainty in locality points for species distributional modeling

1. Species distribution modelling is used increasingly in both applied and theoretical research to predict how species are distributed and to understand attributes of species' environmental requirements. In species distribution modelling, various statistical methods are used that combine species occurrence data with environmental spatial data layers to predict the suitability of any site for that species. While the number of data sharing initiatives involving species' occurrences in the scientific community has increased dramatically over the past few years, various data quality and methodological concerns related to using these data for species distribution modelling have not been addressed adequately. 2. We evaluated how uncertainty in georeferences and associated locational error in occurrences influence species distribution modelling using two treatments: (1) a control treatment where models were calibrated with original, accurate data and (2) an error treatment where data were first degraded spatially to simulate locational error. To incorporate error into the coordinates, we moved each coordinate with a random number drawn from the normal distribution with a mean of zero and a standard deviation of 5 km. We evaluated the influence of error on the performance of 10 commonly used distributional modelling techniques applied to 40 species in four distinct geographical regions. 3. Locational error in occurrences reduced model performance in three of these regions; relatively accurate predictions of species distributions were possible for most species, even with degraded occurrences. Two species distribution modelling techniques, boosted regression trees and maximum entropy, were the best performing models in the face of locational errors. The results obtained with boosted regression trees were only slightly degraded by errors in location, and the results obtained with the maximum entropy approach were not affected by such errors. 4. Synthesis and applications. To use the vast array of occurrence data that exists currently for research and management relating to the geographical ranges of species, modellers need to know the influence of locational error on model quality and whether some modelling techniques are particularly robust to error. We show that certain modelling techniques are particularly robust to a moderate level of locational error and that useful predictions of species distributions can be made even when occurrence data include some error.

Synchronous versus asynchronous modeling of gene regulatory networks.

MOTIVATION: In silico modeling of gene regulatory networks has gained some momentum recently due to increased interest in analyzing the dynamics of biological systems. This has been further facilitated by the increasing availability of experimental data on gene-gene, protein-protein and gene-protein interactions. The two dynamical properties that are often experimentally testable are perturbations and stable steady states. Although a lot of work has been done on the identification of steady states, not much work has been reported on in silico modeling of cellular differentiation processes. RESULTS: In this manuscript, we provide algorithms based on reduced ordered binary decision diagrams (ROBDDs) for Boolean modeling of gene regulatory networks. Algorithms for synchronous and asynchronous transition models have been proposed and their corresponding computational properties have been analyzed. These algorithms allow users to compute cyclic attractors of large networks that are currently not feasible using existing software. Hereby we provide a framework to analyze the effect of multiple gene perturbation protocols, and their effect on cell differentiation processes. These algorithms were validated on the T-helper model showing the correct steady state identification and Th1-Th2 cellular differentiation process. AVAILABILITY: The software binaries for Windows and Linux platforms can be downloaded from http://si2.epfl.ch/~garg/genysis.html.

Treeline and climate change : analyzing and modeling patterns and shifts in the Swiss Alps

Summary: Global warming has led to an average earth surface temperature increase of about 0.7 °C in the 20th century, according to the 2007 IPCC report. In Switzerland, the temperature increase in the same period was even higher: 1.3 °C in the Northern Alps anal 1.7 °C in the Southern Alps. The impacts of this warming on ecosystems aspecially on climatically sensitive systems like the treeline ecotone -are already visible today. Alpine treeline species show increased growth rates, more establishment of young trees in forest gaps is observed in many locations and treelines are migrating upwards. With the forecasted warming, this globally visible phenomenon is expected to continue. This PhD thesis aimed to develop a set of methods and models to investigate current and future climatic treeline positions and treeline shifts in the Swiss Alps in a spatial context. The focus was therefore on: 1) the quantification of current treeline dynamics and its potential causes, 2) the evaluation and improvement of temperaturebased treeline indicators and 3) the spatial analysis and projection of past, current and future climatic treeline positions and their respective elevational shifts. The methods used involved a combination of field temperature measurements, statistical modeling and spatial modeling in a geographical information system. To determine treeline shifts and assign the respective drivers, neighborhood relationships between forest patches were analyzed using moving window algorithms. Time series regression modeling was used in the development of an air-to-soil temperature transfer model to calculate thermal treeline indicators. The indicators were then applied spatially to delineate the climatic treeline, based on interpolated temperature data. Observation of recent forest dynamics in the Swiss treeline ecotone showed that changes were mainly due to forest in-growth, but also partly to upward attitudinal shifts. The recent reduction in agricultural land-use was found to be the dominant driver of these changes. Climate-driven changes were identified only at the uppermost limits of the treeline ecotone. Seasonal mean temperature indicators were found to be the best for predicting climatic treelines. Applying dynamic seasonal delimitations and the air-to-soil temperature transfer model improved the indicators' applicability for spatial modeling. Reproducing the climatic treelines of the past 45 years revealed regionally different attitudinal shifts, the largest being located near the highest mountain mass. Modeling climatic treelines based on two IPCC climate warming scenarios predicted major shifts in treeline altitude. However, the currently-observed treeline is not expected to reach this limit easily, due to lagged reaction, possible climate feedback effects and other limiting factors. Résumé: Selon le rapport 2007 de l'IPCC, le réchauffement global a induit une augmentation de la température terrestre de 0.7 °C en moyenne au cours du 20e siècle. En Suisse, l'augmentation durant la même période a été plus importante: 1.3 °C dans les Alpes du nord et 1.7 °C dans les Alpes du sud. Les impacts de ce réchauffement sur les écosystèmes - en particuliers les systèmes sensibles comme l'écotone de la limite des arbres - sont déjà visibles aujourd'hui. Les espèces de la limite alpine des forêts ont des taux de croissance plus forts, on observe en de nombreux endroits un accroissement du nombre de jeunes arbres s'établissant dans les trouées et la limite des arbres migre vers le haut. Compte tenu du réchauffement prévu, on s'attend à ce que ce phénomène, visible globalement, persiste. Cette thèse de doctorat visait à développer un jeu de méthodes et de modèles pour étudier dans un contexte spatial la position présente et future de la limite climatique des arbres, ainsi que ses déplacements, au sein des Alpes suisses. L'étude s'est donc focalisée sur: 1) la quantification de la dynamique actuelle de la limite des arbres et ses causes potentielles, 2) l'évaluation et l'amélioration des indicateurs, basés sur la température, pour la limite des arbres et 3) l'analyse spatiale et la projection de la position climatique passée, présente et future de la limite des arbres et des déplacements altitudinaux de cette position. Les méthodes utilisées sont une combinaison de mesures de température sur le terrain, de modélisation statistique et de la modélisation spatiale à l'aide d'un système d'information géographique. Les relations de voisinage entre parcelles de forêt ont été analysées à l'aide d'algorithmes utilisant des fenêtres mobiles, afin de mesurer les déplacements de la limite des arbres et déterminer leurs causes. Un modèle de transfert de température air-sol, basé sur les modèles de régression sur séries temporelles, a été développé pour calculer des indicateurs thermiques de la limite des arbres. Les indicateurs ont ensuite été appliqués spatialement pour délimiter la limite climatique des arbres, sur la base de données de températures interpolées. L'observation de la dynamique forestière récente dans l'écotone de la limite des arbres en Suisse a montré que les changements étaient principalement dus à la fermeture des trouées, mais aussi en partie à des déplacements vers des altitudes plus élevées. Il a été montré que la récente déprise agricole était la cause principale de ces changements. Des changements dus au climat n'ont été identifiés qu'aux limites supérieures de l'écotone de la limite des arbres. Les indicateurs de température moyenne saisonnière se sont avérés le mieux convenir pour prédire la limite climatique des arbres. L'application de limites dynamiques saisonnières et du modèle de transfert de température air-sol a amélioré l'applicabilité des indicateurs pour la modélisation spatiale. La reproduction des limites climatiques des arbres durant ces 45 dernières années a mis en évidence des changements d'altitude différents selon les régions, les plus importants étant situés près du plus haut massif montagneux. La modélisation des limites climatiques des arbres d'après deux scénarios de réchauffement climatique de l'IPCC a prédit des changements majeurs de l'altitude de la limite des arbres. Toutefois, l'on ne s'attend pas à ce que la limite des arbres actuellement observée atteigne cette limite facilement, en raison du délai de réaction, d'effets rétroactifs du climat et d'autres facteurs limitants.

MAMOT: hidden Markov modeling tool.

Hidden Markov models (HMMs) are probabilistic models that are well adapted to many tasks in bioinformatics, for example, for predicting the occurrence of specific motifs in biological sequences. MAMOT is a command-line program for Unix-like operating systems, including MacOS X, that we developed to allow scientists to apply HMMs more easily in their research. One can define the architecture and initial parameters of the model in a text file and then use MAMOT for parameter optimization on example data, decoding (like predicting motif occurrence in sequences) and the production of stochastic sequences generated according to the probabilistic model. Two examples for which models are provided are coiled-coil domains in protein sequences and protein binding sites in DNA. A wealth of useful features include the use of pseudocounts, state tying and fixing of selected parameters in learning, and the inclusion of prior probabilities in decoding. AVAILABILITY: MAMOT is implemented in C++, and is distributed under the GNU General Public Licence (GPL). The software, documentation, and example model files can be found at http://bcf.isb-sib.ch/mamot

COMMIT: Convex Optimization Modeling for Microstructure Informed Tractography.

Tractography is a class of algorithms aiming at in vivo mapping the major neuronal pathways in the white matter from diffusion magnetic resonance imaging (MRI) data. These techniques offer a powerful tool to noninvasively investigate at the macroscopic scale the architecture of the neuronal connections of the brain. However, unfortunately, the reconstructions recovered with existing tractography algorithms are not really quantitative even though diffusion MRI is a quantitative modality by nature. As a matter of fact, several techniques have been proposed in recent years to estimate, at the voxel level, intrinsic microstructural features of the tissue, such as axonal density and diameter, by using multicompartment models. In this paper, we present a novel framework to reestablish the link between tractography and tissue microstructure. Starting from an input set of candidate fiber-tracts, which are estimated from the data using standard fiber-tracking techniques, we model the diffusion MRI signal in each voxel of the image as a linear combination of the restricted and hindered contributions generated in every location of the brain by these candidate tracts. Then, we seek for the global weight of each of them, i.e., the effective contribution or volume, such that they globally fit the measured signal at best. We demonstrate that these weights can be easily recovered by solving a global convex optimization problem and using efficient algorithms. The effectiveness of our approach has been evaluated both on a realistic phantom with known ground-truth and in vivo brain data. Results clearly demonstrate the benefits of the proposed formulation, opening new perspectives for a more quantitative and biologically plausible assessment of the structural connectivity of the brain.

Modeling stochasticity and robustness in gene regulatory networks.

MOTIVATION: Understanding gene regulation in biological processes and modeling the robustness of underlying regulatory networks is an important problem that is currently being addressed by computational systems biologists. Lately, there has been a renewed interest in Boolean modeling techniques for gene regulatory networks (GRNs). However, due to their deterministic nature, it is often difficult to identify whether these modeling approaches are robust to the addition of stochastic noise that is widespread in gene regulatory processes. Stochasticity in Boolean models of GRNs has been addressed relatively sparingly in the past, mainly by flipping the expression of genes between different expression levels with a predefined probability. This stochasticity in nodes (SIN) model leads to over representation of noise in GRNs and hence non-correspondence with biological observations. RESULTS: In this article, we introduce the stochasticity in functions (SIF) model for simulating stochasticity in Boolean models of GRNs. By providing biological motivation behind the use of the SIF model and applying it to the T-helper and T-cell activation networks, we show that the SIF model provides more biologically robust results than the existing SIN model of stochasticity in GRNs. AVAILABILITY: Algorithms are made available under our Boolean modeling toolbox, GenYsis. The software binaries can be downloaded from http://si2.epfl.ch/ approximately garg/genysis.html.

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.

Modeling of the TCR-MHC-peptide complex.

The methodology for generating a homology model of the T1 TCR-PbCS-K(d) class I major histocompatibility complex (MHC) class I complex is presented. The resulting model provides a qualitative explanation of the effect of over 50 different mutations in the region of the complementarity determining region (CDR) loops of the T cell receptor (TCR), the peptide and the MHC's alpha(1)/alpha(2) helices. The peptide is modified by an azido benzoic acid photoreactive group, which is part of the epitope recognized by the TCR. The construction of the model makes use of closely related homologs (the A6 TCR-Tax-HLA A2 complex, the 2C TCR, the 14.3.d TCR Vbeta chain, the 1934.4 TCR Valpha chain, and the H-2 K(b)-ovalbumine peptide), ab initio sampling of CDR loops conformations and experimental data to select from the set of possibilities. The model shows a complex arrangement of the CDR3alpha, CDR1beta, CDR2beta and CDR3beta loops that leads to the highly specific recognition of the photoreactive group. The protocol can be applied systematically to a series of related sequences, permitting the analysis at the structural level of the large TCR repertoire specific for a given peptide-MHC complex.

Modeling the impact of lesions in the human brain.

Lesions of anatomical brain networks result in functional disturbances of brain systems and behavior which depend sensitively, often unpredictably, on the lesion site. The availability of whole-brain maps of structural connections within the human cerebrum and our increased understanding of the physiology and large-scale dynamics of cortical networks allow us to investigate the functional consequences of focal brain lesions in a computational model. We simulate the dynamic effects of lesions placed in different regions of the cerebral cortex by recording changes in the pattern of endogenous ("resting-state") neural activity. We find that lesions produce specific patterns of altered functional connectivity among distant regions of cortex, often affecting both cortical hemispheres. The magnitude of these dynamic effects depends on the lesion location and is partly predicted by structural network properties of the lesion site. In the model, lesions along the cortical midline and in the vicinity of the temporo-parietal junction result in large and widely distributed changes in functional connectivity, while lesions of primary sensory or motor regions remain more localized. The model suggests that dynamic lesion effects can be predicted on the basis of specific network measures of structural brain networks and that these effects may be related to known behavioral and cognitive consequences of brain lesions.

Cost-effective experimental design to support modeling of concentration-response functions

Modeling concentration-response function became extremely popular in ecotoxicology during the last decade. Indeed, modeling allows determining the total response pattern of a given substance. However, reliable modeling is consuming in term of data, which is in contradiction with the current trend in ecotoxicology, which aims to reduce, for cost and ethical reasons, the number of data produced during an experiment. It is therefore crucial to determine experimental design in a cost-effective manner. In this paper, we propose to use the theory of locally D-optimal designs to determine the set of concentrations to be tested so that the parameters of the concentration-response function can be estimated with high precision. We illustrated this approach by determining the locally D-optimal designs to estimate the toxicity of the herbicide dinoseb on daphnids and algae. The results show that the number of concentrations to be tested is often equal to the number of parameters and often related to the their meaning, i.e. they are located close to the parameters. Furthermore, the results show that the locally D-optimal design often has the minimal number of support points and is not much sensitive to small changes in nominal values of the parameters. In order to reduce the experimental cost and the use of test organisms, especially in case of long-term studies, reliable nominal values may therefore be fixed based on prior knowledge and literature research instead of on preliminary experiments

Estimating dormant and active hematopoietic stem cell kinetics through extensive modeling of bromodeoxyuridine label-retaining cell dynamics.

Bone marrow hematopoietic stem cells (HSCs) are responsible for both lifelong daily maintenance of all blood cells and for repair after cell loss. Until recently the cellular mechanisms by which HSCs accomplish these two very different tasks remained an open question. Biological evidence has now been found for the existence of two related mouse HSC populations. First, a dormant HSC (d-HSC) population which harbors the highest self-renewal potential of all blood cells but is only induced into active self-renewal in response to hematopoietic stress. And second, an active HSC (a-HSC) subset that by and large produces the progenitors and mature cells required for maintenance of day-to-day hematopoiesis. Here we present computational analyses further supporting the d-HSC concept through extensive modeling of experimental DNA label-retaining cell (LRC) data. Our conclusion that the presence of a slowly dividing subpopulation of HSCs is the most likely explanation (amongst the various possible causes including stochastic cellular variation) of the observed long term Bromodeoxyuridine (BrdU) retention, is confirmed by the deterministic and stochastic models presented here. Moreover, modeling both HSC BrdU uptake and dilution in three stages and careful treatment of the BrdU detection sensitivity permitted improved estimates of HSC turnover rates. This analysis predicts that d-HSCs cycle about once every 149-193 days and a-HSCs about once every 28-36 days. We further predict that, using LRC assays, a 75%-92.5% purification of d-HSCs can be achieved after 59-130 days of chase. Interestingly, the d-HSC proportion is now estimated to be around 30-45% of total HSCs - more than twice that of our previous estimate.