Machine learning algorithms for spatial data. Case studies: Environmental pollution, natural hazards, renewable resources

This paper presents general problems and approaches for the spatial data analysis using machine learning algorithms. Machine learning is a very powerful approach to adaptive data analysis, modelling and visualisation. The key feature of the machine learning algorithms is that they learn from empirical data and can be used in cases when the modelled environmental phenomena are hidden, nonlinear, noisy and highly variable in space and in time. Most of the machines learning algorithms are universal and adaptive modelling tools developed to solve basic problems of learning from data: classification/pattern recognition, regression/mapping and probability density modelling. In the present report some of the widely used machine learning algorithms, namely artificial neural networks (ANN) of different architectures and Support Vector Machines (SVM), are adapted to the problems of the analysis and modelling of geo-spatial data. Machine learning algorithms have an important advantage over traditional models of spatial statistics when problems are considered in a high dimensional geo-feature spaces, when the dimension of space exceeds 5. Such features are usually generated, for example, from digital elevation models, remote sensing images, etc. An important extension of models concerns considering of real space constrains like geomorphology, networks, and other natural structures. Recent developments in semi-supervised learning can improve modelling of environmental phenomena taking into account on geo-manifolds. An important part of the study deals with the analysis of relevant variables and models' inputs. This problem is approached by using different feature selection/feature extraction nonlinear tools. To demonstrate the application of machine learning algorithms several interesting case studies are considered: digital soil mapping using SVM, automatic mapping of soil and water system pollution using ANN; natural hazards risk analysis (avalanches, landslides), assessments of renewable resources (wind fields) with SVM and ANN models, etc. The dimensionality of spaces considered varies from 2 to more than 30. Figures 1, 2, 3 demonstrate some results of the studies and their outputs. Finally, the results of environmental mapping are discussed and compared with traditional models of geostatistics.

Simple estimation of incident HIV infection rates in notification cohorts based on window periods of algorithms for evaluation of line-immunoassay result patterns.

BACKGROUND: Tests for recent infections (TRIs) are important for HIV surveillance. We have shown that a patient's antibody pattern in a confirmatory line immunoassay (Inno-Lia) also yields information on time since infection. We have published algorithms which, with a certain sensitivity and specificity, distinguish between incident (< = 12 months) and older infection. In order to use these algorithms like other TRIs, i.e., based on their windows, we now determined their window periods. METHODS: We classified Inno-Lia results of 527 treatment-naïve patients with HIV-1 infection < = 12 months according to incidence by 25 algorithms. The time after which all infections were ruled older, i.e. the algorithm's window, was determined by linear regression of the proportion ruled incident in dependence of time since infection. Window-based incident infection rates (IIR) were determined utilizing the relationship 'Prevalence = Incidence x Duration' in four annual cohorts of HIV-1 notifications. Results were compared to performance-based IIR also derived from Inno-Lia results, but utilizing the relationship 'incident = true incident + false incident' and also to the IIR derived from the BED incidence assay. RESULTS: Window periods varied between 45.8 and 130.1 days and correlated well with the algorithms' diagnostic sensitivity (R(2) = 0.962; P<0.0001). Among the 25 algorithms, the mean window-based IIR among the 748 notifications of 2005/06 was 0.457 compared to 0.453 obtained for performance-based IIR with a model not correcting for selection bias. Evaluation of BED results using a window of 153 days yielded an IIR of 0.669. Window-based IIR and performance-based IIR increased by 22.4% and respectively 30.6% in 2008, while 2009 and 2010 showed a return to baseline for both methods. CONCLUSIONS: IIR estimations by window- and performance-based evaluations of Inno-Lia algorithm results were similar and can be used together to assess IIR changes between annual HIV notification cohorts.

Sélection et suivi prétransplantation des patients candidats à la greffe cardiaque en Suisse romande [Selection and preoperative follow-up of heart transplantation candidates in the French part of Switzerland].

Heart transplantation (HTx) started in 1987 at two university hospitals (CHUV, HUG) in the western part of Switzerland, with 223 HTx performed at the CHUV until December 2010. Between 1987 and 2003, 106 HTx were realized at the HUG resulting in a total of 329 HTx in the western part of Switzerland. After the relocation of organ transplantation activity in the western part of Switzerland in 2003, the surgical part and the early postoperative care of HTx remained limited to the CHUV. However, every other HTx activity are pursued at the two university hospitals (CHUV, HUG). This article summarizes the actual protocols for selection and pre-transplant follow-up of HTx candidates in the western part of Switzerland, permitting a uniform structure of pretransplant follow-up in the western part of Switzerland.

Combined T2 -preparation and two-dimensional pencil-beam inner volume selection.

PURPOSE: To improve coronary magnetic resonance angiography (MRA) by combining a two-dimensional (2D) spatially selective radiofrequency (RF) pulse with a T2 -preparation module ("2D-T2 -Prep"). METHODS: An adiabatic T2 -Prep was modified so that the first and last pulses were of differing spatial selectivity. The first RF pulse was replaced by a 2D pulse, such that a pencil-beam volume is excited. The last RF pulse remains nonselective, thus restoring the T2 -prepared pencil-beam, while tipping the (formerly longitudinal) magnetization outside of the pencil-beam into the transverse plane, where it is then spoiled. Thus, only a cylinder of T2 -prepared tissue remains for imaging. Numerical simulations were followed by phantom validation and in vivo coronary MRA, where the technique was quantitatively evaluated. Reduced field-of-view (rFoV) images were similarly studied. RESULTS: In vivo, full field-of-view 2D-T2 -Prep significantly improved vessel sharpness as compared to conventional T2 -Prep, without adversely affecting signal-to-noise (SNR) or contrast-to-noise ratios (CNR). It also reduced respiratory motion artifacts. In rFoV images, the SNR, CNR, and vessel sharpness decreased, although scan time reduction was 60%. CONCLUSION: When compared with conventional T2 -Prep, the 2D-T2 -Prep improves vessel sharpness and decreases respiratory ghosting while preserving both SNR and CNR. It may also acquire rFoV images for accelerated data acquisition.

Effects of selection and drift on G matrix evolution in a heterogeneous environment: a multivariate Qst-Fst Test with the freshwater snail Galba truncatula.

Unraveling the effect of selection vs. drift on the evolution of quantitative traits is commonly achieved by one of two methods. Either one contrasts population differentiation estimates for genetic markers and quantitative traits (the Q(st)-F(st) contrast) or multivariate methods are used to study the covariance between sets of traits. In particular, many studies have focused on the genetic variance-covariance matrix (the G matrix). However, both drift and selection can cause changes in G. To understand their joint effects, we recently combined the two methods into a single test (accompanying article by Martin et al.), which we apply here to a network of 16 natural populations of the freshwater snail Galba truncatula. Using this new neutrality test, extended to hierarchical population structures, we studied the multivariate equivalent of the Q(st)-F(st) contrast for several life-history traits of G. truncatula. We found strong evidence of selection acting on multivariate phenotypes. Selection was homogeneous among populations within each habitat and heterogeneous between habitats. We found that the G matrices were relatively stable within each habitat, with proportionality between the among-populations (D) and the within-populations (G) covariance matrices. The effect of habitat heterogeneity is to break this proportionality because of selection for habitat-dependent optima. Individual-based simulations mimicking our empirical system confirmed that these patterns are expected under the selective regime inferred. We show that homogenizing selection can mimic some effect of drift on the G matrix (G and D almost proportional), but that incorporating information from molecular markers (multivariate Q(st)-F(st)) allows disentangling the two effects.

Multi-scale support vector algorithms for hot spot detection and modelling

The algorithmic approach to data modelling has developed rapidly these last years, in particular methods based on data mining and machine learning have been used in a growing number of applications. These methods follow a data-driven methodology, aiming at providing the best possible generalization and predictive abilities instead of concentrating on the properties of the data model. One of the most successful groups of such methods is known as Support Vector algorithms. Following the fruitful developments in applying Support Vector algorithms to spatial data, this paper introduces a new extension of the traditional support vector regression (SVR) algorithm. This extension allows for the simultaneous modelling of environmental data at several spatial scales. The joint influence of environmental processes presenting different patterns at different scales is here learned automatically from data, providing the optimum mixture of short and large-scale models. The method is adaptive to the spatial scale of the data. With this advantage, it can provide efficient means to model local anomalies that may typically arise in situations at an early phase of an environmental emergency. However, the proposed approach still requires some prior knowledge on the possible existence of such short-scale patterns. This is a possible limitation of the method for its implementation in early warning systems. The purpose of this paper is to present the multi-scale SVR model and to illustrate its use with an application to the mapping of Cs137 activity given the measurements taken in the region of Briansk following the Chernobyl accident.

Wound infection after excision and primary midline closure for pilonidal disease: risk factor analysis to improve patient selection.

BACKGROUND: Excision and primary midline closure for pilonidal disease (PD) is a simple procedure; however, it is frequently complicated by infection and prolonged healing. The aim of this study was to analyze risk factors for surgical site infection (SSI) in this context. METHODS: All consecutive patients undergoing excision and primary closure for PD from January 2002 through October 2008 were retrospectively assessed. The end points were SSI, as defined by the Center for Disease Control, and time to healing. Univariable and multivariable risk factor analyses were performed. RESULTS: One hundred thirty-one patients were included [97 men (74%), median age = 24 (range 15-66) years]. SSI occurred in 41 (31%) patients. Median time to healing was 20 days (range 12-76) in patients without SSI and 62 days (range 20-176) in patients with SSI (P < 0.0001). In univariable and multivariable analyses, smoking [OR = 2.6 (95% CI 1.02, 6.8), P = 0.046] and lack of antibiotic prophylaxis [OR = 5.6 (95% CI 2.5, 14.3), P = 0.001] were significant predictors for SSI. Adjusted for SSI, age over 25 was a significant predictor of prolonged healing. CONCLUSION: This study suggests that the rate of SSI after excision and primary closure of PD is higher in smokers and could be reduced by antibiotic prophylaxis. SSI significantly prolongs healing time, particularly in patients over 25 years.

Selection of patients with germline MLH1 mutated Lynch syndrome by determination of MLH1 methylation and BRAF mutation.

Lynch syndrome is one of the most common hereditary colorectal cancer (CRC) syndrome and is caused by germline mutations of MLH1, MSH2 and more rarely MSH6, PMS2, MLH3 genes. Whereas the absence of MSH2 protein is predictive of Lynch syndrome, it is not the case for the absence of MLH1 protein. The purpose of this study was to develop a sensitive and cost effective algorithm to select Lynch syndrome cases among patients with MLH1 immunohistochemical silencing. Eleven sporadic CRC and 16 Lynch syndrome cases with MLH1 protein abnormalities were selected. The BRAF c.1799T> A mutation (p.Val600Glu) was analyzed by direct sequencing after PCR amplification of exon 15. Methylation of MLH1 promoter was determined by Methylation-Sensitive Single-Strand Conformation Analysis. In patients with Lynch syndrome, there was no BRAF mutation and only one case showed MLH1 methylation (6%). In sporadic CRC, all cases were MLH1 methylated (100%) and 8 out of 11 cases carried the above BRAF mutation (73%) whereas only 3 cases were BRAF wild type (27%). We propose the following algorithm: (1) no further molecular analysis should be performed for CRC exhibiting MLH1 methylation and BRAF mutation, and these cases should be considered as sporadic CRC; (2) CRC with unmethylated MLH1 and negative for BRAF mutation should be considered as Lynch syndrome; and (3) only a small fraction of CRC with MLH1 promoter methylation but negative for BRAF mutation should be true Lynch syndrome patients. These potentially Lynch syndrome patients should be offered genetic counselling before searching for MLH1 gene mutations.

Pervasive positive selection on duplicated and nonduplicated vertebrate protein coding genes.

A stringent branch-site codon model was used to detect positive selection in vertebrate evolution. We show that the test is robust to the large evolutionary distances involved. Positive selection was detected in 77% of 884 genes studied. Most positive selection concerns a few sites on a single branch of the phylogenetic tree: Between 0.9% and 4.7% of sites are affected by positive selection depending on the branches. No functional category was overrepresented among genes under positive selection. Surprisingly, whole genome duplication had no effect on the prevalence of positive selection, whether the fish-specific genome duplication or the two rounds at the origin of vertebrates. Thus positive selection has not been limited to a few gene classes, or to specific evolutionary events such as duplication, but has been pervasive during vertebrate evolution.

Effect of training-sample size and classification difficulty on the accuracy of genomic predictors.

Introduction: As part of the MicroArray Quality Control (MAQC)-II project, this analysis examines how the choice of univariate feature-selection methods and classification algorithms may influence the performance of genomic predictors under varying degrees of prediction difficulty represented by three clinically relevant endpoints. Methods: We used gene-expression data from 230 breast cancers (grouped into training and independent validation sets), and we examined 40 predictors (five univariate feature-selection methods combined with eight different classifiers) for each of the three endpoints. Their classification performance was estimated on the training set by using two different resampling methods and compared with the accuracy observed in the independent validation set. Results: A ranking of the three classification problems was obtained, and the performance of 120 models was estimated and assessed on an independent validation set. The bootstrapping estimates were closer to the validation performance than were the cross-validation estimates. The required sample size for each endpoint was estimated, and both gene-level and pathway-level analyses were performed on the obtained models. Conclusions: We showed that genomic predictor accuracy is determined largely by an interplay between sample size and classification difficulty. Variations on univariate feature-selection methods and choice of classification algorithm have only a modest impact on predictor performance, and several statistically equally good predictors can be developed for any given classification problem.

Aging and motor programming: differential effects according to the selection of action

There are controversial reports about the effect of aging on movement preparation, and it is unclear to which extent cognitive and/or motor related cerebral processes may be affected. This study examines the age effects on electro-cortical oscillatory patterns during various motor programming tasks, in order to assess potential differences according to the mode of action selection. Twenty elderly (EP, 60-84 years) and 20 young (YP, 20-29 years) participants with normal cognition underwent 3 pre-cued response tasks (S1-S2 paradigm). S1 carried either complete information on response side (Full; stimulus-driven motor preparation), no information (None; general motor alertness), or required free response side selection (Free; internally-driven motor preparation). Electroencephalogram (EEG) was recorded using 64 surface electrodes. Alpha (8-12 Hz) desynchronization (ERD)/synchronization (ERS) and motor-related amplitude asymmetries (MRAA) were analyzed during the S1-S2 interval. Reaction times (RTs) to S2 were slower in EP than YP, and in None than in the other 2 tasks. There was an Age x Task interaction due to increased RTs in Free compared to Full in EP only. Central bilateral and midline activation (alpha ERD) was smaller in EP than YP in None. In Full just before S2, readiness to move was reflected by posterior midline inhibition (alpha ERS) in both groups. In Free, such inhibition was present only in YP. Moreover, MRAA showed motor activity lateralization in both groups in Full, but only in YP in Free. The results indicate reduced recruitment of motor regions for motor alertness in the elderly. They further show less efficient cerebral processes subtending free selection of movement in elders, suggesting reduced capacity for internally-driven action with age.

Inconsistency between different measures of sexual selection.

Measuring the intensity of sexual selection is of fundamental importance to the study of sexual dimorphism, population dynamics, and speciation. Several indices, pools of individuals, and fitness proxies are used in the literature, yet their relative performances are strongly debated. Using 12 independent common lizard populations, we manipulated the adult sex ratio, a potentially important determinant of the intensity of sexual selection at a particular time and place. We investigated differences in the intensity of sexual selection, as estimated using three standard indices of sexual selection-the standardized selection gradient (β'), the opportunity of selection (I), and the Bateman gradient (βss)--calculated for different pools of individuals and different fitness proxies. We show that results based on estimates of I were the opposite of those derived from the other indices, whereas results based on estimates of β' were consistent with predictions derived from knowledge about the species' mating system. In addition, our estimates of the strength and direction of sexual selection depended on both the fitness proxy used and the pool of individuals included in the analysis. These observations demonstrate inconsistencies in distinct measures of sexual selection and underscore the need for caution when comparing studies and species.