Automatic learning for the classification of chemical reactions and in statistical thermodynamics

This Thesis describes the application of automatic learning methods for a) the classification of organic and metabolic reactions, and b) the mapping of Potential Energy Surfaces(PES). The classification of reactions was approached with two distinct methodologies: a representation of chemical reactions based on NMR data, and a representation of chemical reactions from the reaction equation based on the physico-chemical and topological features of chemical bonds. NMR-based classification of photochemical and enzymatic reactions. Photochemical and metabolic reactions were classified by Kohonen Self-Organizing Maps (Kohonen SOMs) and Random Forests (RFs) taking as input the difference between the 1H NMR spectra of the products and the reactants. The development of such a representation can be applied in automatic analysis of changes in the 1H NMR spectrum of a mixture and their interpretation in terms of the chemical reactions taking place. Examples of possible applications are the monitoring of reaction processes, evaluation of the stability of chemicals, or even the interpretation of metabonomic data. A Kohonen SOM trained with a data set of metabolic reactions catalysed by transferases was able to correctly classify 75% of an independent test set in terms of the EC number subclass. Random Forests improved the correct predictions to 79%. With photochemical reactions classified into 7 groups, an independent test set was classified with 86-93% accuracy. The data set of photochemical reactions was also used to simulate mixtures with two reactions occurring simultaneously. Kohonen SOMs and Feed-Forward Neural Networks (FFNNs) were trained to classify the reactions occurring in a mixture based on the 1H NMR spectra of the products and reactants. Kohonen SOMs allowed the correct assignment of 53-63% of the mixtures (in a test set). Counter-Propagation Neural Networks (CPNNs) gave origin to similar results. The use of supervised learning techniques allowed an improvement in the results. They were improved to 77% of correct assignments when an ensemble of ten FFNNs were used and to 80% when Random Forests were used. This study was performed with NMR data simulated from the molecular structure by the SPINUS program. In the design of one test set, simulated data was combined with experimental data. The results support the proposal of linking databases of chemical reactions to experimental or simulated NMR data for automatic classification of reactions and mixtures of reactions. Genome-scale classification of enzymatic reactions from their reaction equation. The MOLMAP descriptor relies on a Kohonen SOM that defines types of bonds on the basis of their physico-chemical and topological properties. The MOLMAP descriptor of a molecule represents the types of bonds available in that molecule. The MOLMAP descriptor of a reaction is defined as the difference between the MOLMAPs of the products and the reactants, and numerically encodes the pattern of bonds that are broken, changed, and made during a chemical reaction. The automatic perception of chemical similarities between metabolic reactions is required for a variety of applications ranging from the computer validation of classification systems, genome-scale reconstruction (or comparison) of metabolic pathways, to the classification of enzymatic mechanisms. Catalytic functions of proteins are generally described by the EC numbers that are simultaneously employed as identifiers of reactions, enzymes, and enzyme genes, thus linking metabolic and genomic information. Different methods should be available to automatically compare metabolic reactions and for the automatic assignment of EC numbers to reactions still not officially classified. In this study, the genome-scale data set of enzymatic reactions available in the KEGG database was encoded by the MOLMAP descriptors, and was submitted to Kohonen SOMs to compare the resulting map with the official EC number classification, to explore the possibility of predicting EC numbers from the reaction equation, and to assess the internal consistency of the EC classification at the class level. A general agreement with the EC classification was observed, i.e. a relationship between the similarity of MOLMAPs and the similarity of EC numbers. At the same time, MOLMAPs were able to discriminate between EC sub-subclasses. EC numbers could be assigned at the class, subclass, and sub-subclass levels with accuracies up to 92%, 80%, and 70% for independent test sets. The correspondence between chemical similarity of metabolic reactions and their MOLMAP descriptors was applied to the identification of a number of reactions mapped into the same neuron but belonging to different EC classes, which demonstrated the ability of the MOLMAP/SOM approach to verify the internal consistency of classifications in databases of metabolic reactions. RFs were also used to assign the four levels of the EC hierarchy from the reaction equation. EC numbers were correctly assigned in 95%, 90%, 85% and 86% of the cases (for independent test sets) at the class, subclass, sub-subclass and full EC number level,respectively. Experiments for the classification of reactions from the main reactants and products were performed with RFs - EC numbers were assigned at the class, subclass and sub-subclass level with accuracies of 78%, 74% and 63%, respectively. In the course of the experiments with metabolic reactions we suggested that the MOLMAP / SOM concept could be extended to the representation of other levels of metabolic information such as metabolic pathways. Following the MOLMAP idea, the pattern of neurons activated by the reactions of a metabolic pathway is a representation of the reactions involved in that pathway - a descriptor of the metabolic pathway. This reasoning enabled the comparison of different pathways, the automatic classification of pathways, and a classification of organisms based on their biochemical machinery. The three levels of classification (from bonds to metabolic pathways) allowed to map and perceive chemical similarities between metabolic pathways even for pathways of different types of metabolism and pathways that do not share similarities in terms of EC numbers. Mapping of PES by neural networks (NNs). In a first series of experiments, ensembles of Feed-Forward NNs (EnsFFNNs) and Associative Neural Networks (ASNNs) were trained to reproduce PES represented by the Lennard-Jones (LJ) analytical potential function. The accuracy of the method was assessed by comparing the results of molecular dynamics simulations (thermal, structural, and dynamic properties) obtained from the NNs-PES and from the LJ function. The results indicated that for LJ-type potentials, NNs can be trained to generate accurate PES to be used in molecular simulations. EnsFFNNs and ASNNs gave better results than single FFNNs. A remarkable ability of the NNs models to interpolate between distant curves and accurately reproduce potentials to be used in molecular simulations is shown. The purpose of the first study was to systematically analyse the accuracy of different NNs. Our main motivation, however, is reflected in the next study: the mapping of multidimensional PES by NNs to simulate, by Molecular Dynamics or Monte Carlo, the adsorption and self-assembly of solvated organic molecules on noble-metal electrodes. Indeed, for such complex and heterogeneous systems the development of suitable analytical functions that fit quantum mechanical interaction energies is a non-trivial or even impossible task. The data consisted of energy values, from Density Functional Theory (DFT) calculations, at different distances, for several molecular orientations and three electrode adsorption sites. The results indicate that NNs require a data set large enough to cover well the diversity of possible interaction sites, distances, and orientations. NNs trained with such data sets can perform equally well or even better than analytical functions. Therefore, they can be used in molecular simulations, particularly for the ethanol/Au (111) interface which is the case studied in the present Thesis. Once properly trained, the networks are able to produce, as output, any required number of energy points for accurate interpolations.

Assessment of climate change statistical downscaling methods: Application and comparison of two statistical methods to a single site in Lisbon

Dissertação apresentada na Faculdade de Ciências e Tecnologia da Universidade Nova de Lisboa para a obtenção do grau de Mestre em Engenharia do Ambiente

Some reflections on acquisition, processing and analysis of statistical data in forest soils

Beyond the classical statistical approaches (determination of basic statistics, regression analysis, ANOVA, etc.) a new set of applications of different statistical techniques has increasingly gained relevance in the analysis, processing and interpretation of data concerning the characteristics of forest soils. This is possible to be seen in some of the recent publications in the context of Multivariate Statistics. These new methods require additional care that is not always included or refered in some approaches. In the particular case of geostatistical data applications it is necessary, besides to geo-reference all the data acquisition, to collect the samples in regular grids and in sufficient quantity so that the variograms can reflect the spatial distribution of soil properties in a representative manner. In the case of the great majority of Multivariate Statistics techniques (Principal Component Analysis, Correspondence Analysis, Cluster Analysis, etc.) despite the fact they do not require in most cases the assumption of normal distribution, they however need a proper and rigorous strategy for its utilization. In this work, some reflections about these methodologies and, in particular, about the main constraints that often occur during the information collecting process and about the various linking possibilities of these different techniques will be presented. At the end, illustrations of some particular cases of the applications of these statistical methods will also be presented.

Study of the Reliability of Statistical Timing Analysis for Real-Time Systems

Presented at 23rd International Conference on Real-Time Networks and Systems (RTNS 2015). 4 to 6, Nov, 2015, Main Track. Lille, France.

Thermostability of heterophile antibodies from human sera infected with Schistosoma mansoni and geo-helminths: an immuno-metric statistical analysis

Antibody in human sera that induces lysis of sheep erythrocytes in hemolytic assay was investigated. The present study showed that the presence in serum of the thermostable cytolytic anti-sheep red blood cells antibodies is dependent on the Schistosoma mansoni infection, and this is more frequent in adults than in children. The thermostable characteristic of hemolysins in normal sera was not dependent on the presence of Ascaris lumbricoides, Trichuris trichiura or hookworm geo-helminths. Further, thermostable complement-activating heterophile antibodies were noticed in children in association with massive number of S. mansoni eggs. The results were obtained by using the z- and the chi-square tests. The z-test allows us to formulate a one-sided alternative, i.e., a tendency of one of the attributes. On the other hand, the chi-square test analyzes the independence between attributes by using a contingency table. Besides the obtained results being interesting in the field of schistosomiasis mansoni, they can provide a new insight into the use of statistics in medical science.

Deciding Kleene Algebra Terms Equivalence in Coq

This paper presents a mechanically verified implementation of an algorithm for deciding the equivalence of Kleene algebra terms within the Coq proof assistant. The algorithm decides equivalence of two given regular expressions through an iterated process of testing the equivalence of their partial derivatives and does not require the construction of the corresponding automata. Recent theoretical and experimental research provides evidence that this method is, on average, more efficient than the classical methods based on automata. We present some performance tests, comparisons with similar approaches, and also introduce a generalization of the algorithm to decide the equivalence of terms of Kleene algebra with tests. The motivation for the work presented in this paper is that of using the libraries developed as trusted frameworks for carrying out certified program verification.

Exploraty Multivariate Statistical Methods Applied to Pharmaceutical Industry CRM Data

Dissertação apresentada como requisito parcial para obtenção do grau de Mestre em Estatística e Gestão de Informação

Factor analysis in the stock market: An application to statistical arbitrage

A Work Project, presented as part of the requirements for the Award of a Masters Degree in Finance from the NOVA – School of Business and Economics

Estimation of statistical cure from cancer using population-based data

Dissertação de mestrado em Estatística

Desarrollos estadísticos para el procesamiento de información genómica masiva en agro-biotecnología. Statistical developments for processing massive genomic information in agro-biotechnology.

Nuevas biotecnologías, como los marcadores de la molécula de ADN, permiten caracterizar el genoma vegetal. El uso de la información genómica producida para cientos o miles de posiciones cromosómicas permite identificar genotipos superiores en menos tiempo que el requerido por la selección fenotípica tradicional. La mayoría de los caracteres de las especies vegetales cultivadas de importancia agronómica y económica, son controlados por poli-genes causantes de un fenotipo con variación continua, altamente afectados por el ambiente. Su herencia es compleja ya que resulta de la interacción entre genes, del mismo o distinto cromosoma, y de la interacción del genotipo con el ambiente, dificultando la selección. Estas biotecnologías producen bases de datos con gran cantidad de información y estructuras complejas de correlación que requieren de métodos y modelos biométricos específicos para su procesamiento. Los modelos estadísticos focalizados en explicar el fenotipo a partir de información genómica masiva requieren la estimación de un gran número de parámetros. No existen métodos, dentro de la estadística paramétrica capaces de abordar este problema eficientemente. Además los modelos deben contemplar no-aditividades (interacciones) entre efectos génicos y de éstos con el ambiente que son también dificiles de manejar desde la concepción paramétrica. Se hipotetiza que el análisis de la asociación entre caracteres fenotípicos y genotipos moleculares, caracterizados por abundante información genómica, podría realizarse eficientemente en el contexto de los modelos mixtos semiparamétricos y/o de métodos no-paramétricos basados en técnicas de aprendizaje automático. El objetivo de este proyecto es desarrollar nuevos métodos para análisis de datos que permitan el uso eficiente de información genómica masiva en evaluaciones genéticas de interés agro-biotecnológico. Los objetivos específicos incluyen la comparación, respecto a propiedades estadísticas y computacionales, de estrategias analíticas paramétricas con estrategias semiparamétricas y no-paramétricas. Se trabajará con aproximaciones por regresión del análisis de loci de caracteres cuantitativos bajo distintas estrategias y escenarios (reales y simulados) con distinto volúmenes de datos de marcadores moleculares. En el área paramétrica se pondrá especial énfasis en modelos mixtos, mientras que en el área no paramétrica se evaluarán algoritmos de redes neuronales, máquinas de soporte vectorial, filtros multivariados, suavizados del tipo LOESS y métodos basados en núcleos de reciente aparición. La propuesta semiparamétrica se basará en una estrategia de análisis en dos etapas orientadas a: 1) reducir la dimensionalidad de los datos genómicos y 2) modelar el fenotipo introduciendo sólo las señales moleculares más significativas. Con este trabajo se espera poner a disposición de investigadores de nuestro medio, nuevas herramientas y procedimientos de análisis que permitan maximizar la eficiencia en el uso de los recursos asignados a la masiva captura de datos genómicos y su aplicación en desarrollos agro-biotecnológicos.

Estrategias estadístico-computacionales avanzadas para análisis de expresión de proteínas/genes en cáncer. Advanced statistical and computational strategies for protein/gene expression analysis in cancer.

El objetivo de este proyecto, enmarcado en el área de metodología de análisis en bioingeniería-biotecnología aplicadas al estudio del cancer, es el análisis y caracterización a través modelos estadísticos con efectos mixtos y técnicas de aprendizaje automático, de perfiles de expresión de proteínas y genes de las vías metabolicas asociadas a progresión tumoral. Dicho estudio se llevará a cabo mediante la utilización de tecnologías de alto rendimiento. Las mismas permiten evaluar miles de genes/proteínas en forma simultánea, generando así una gran cantidad de datos de expresión. Se hipotetiza que para un análisis e interpretación de la información subyacente, caracterizada por su abundancia y complejidad, podría realizarse mediante técnicas estadístico-computacionales eficientes en el contexto de modelos mixtos y técnias de aprendizaje automático. Para que el análisis sea efectivo es necesario contemplar los efectos ocasionados por los diferentes factores experimentales ajenos al fenómeno biológico bajo estudio. Estos efectos pueden enmascarar la información subycente y así perder informacion relavante en el contexto de progresión tumoral. La identificación de estos efectos permitirá obtener, eficientemente, los perfiles de expresión molecular que podrían permitir el desarrollo de métodos de diagnóstico basados en ellos. Con este trabajo se espera poner a disposición de investigadores de nuestro medio, herramientas y procedimientos de análisis que maximicen la eficiencia en el uso de los recursos asignados a la masiva captura de datos genómicos/proteómicos que permitan extraer información biológica relevante pertinente al análisis, clasificación o predicción de cáncer, el diseño de tratamientos y terapias específicos y el mejoramiento de los métodos de detección como así tambien aportar al entendimieto de la progresión tumoral mediante análisis computacional intensivo.

DESARROLLO DE APLICACIONES ESTADÍSTICAS PARA LA AGRICULTURA DE PRECISIÓN. STATISTICAL APPLICATIONS FOR PRECISION AGRICULTURE

A partir de las últimas décadas se ha impulsado el desarrollo y la utilización de los Sistemas de Información Geográficos (SIG) y los Sistemas de Posicionamiento Satelital (GPS) orientados a mejorar la eficiencia productiva de distintos sistemas de cultivos extensivos en términos agronómicos, económicos y ambientales. Estas nuevas tecnologías permiten medir variabilidad espacial de propiedades del sitio como conductividad eléctrica aparente y otros atributos del terreno así como el efecto de las mismas sobre la distribución espacial de los rendimientos. Luego, es posible aplicar el manejo sitio-específico en los lotes para mejorar la eficiencia en el uso de los insumos agroquímicos, la protección del medio ambiente y la sustentabilidad de la vida rural. En la actualidad, existe una oferta amplia de recursos tecnológicos propios de la agricultura de precisión para capturar variación espacial a través de los sitios dentro del terreno. El óptimo uso del gran volumen de datos derivado de maquinarias de agricultura de precisión depende fuertemente de las capacidades para explorar la información relativa a las complejas interacciones que subyacen los resultados productivos. La covariación espacial de las propiedades del sitio y el rendimiento de los cultivos ha sido estudiada a través de modelos geoestadísticos clásicos que se basan en la teoría de variables regionalizadas. Nuevos desarrollos de modelos estadísticos contemporáneos, entre los que se destacan los modelos lineales mixtos, constituyen herramientas prometedoras para el tratamiento de datos correlacionados espacialmente. Más aún, debido a la naturaleza multivariada de las múltiples variables registradas en cada sitio, las técnicas de análisis multivariado podrían aportar valiosa información para la visualización y explotación de datos georreferenciados. La comprensión de las bases agronómicas de las complejas interacciones que se producen a la escala de lotes en producción, es hoy posible con el uso de éstas nuevas tecnologías. Los objetivos del presente proyecto son: (l) desarrollar estrategias metodológicas basadas en la complementación de técnicas de análisis multivariados y geoestadísticas, para la clasificación de sitios intralotes y el estudio de interdependencias entre variables de sitio y rendimiento; (ll) proponer modelos mixtos alternativos, basados en funciones de correlación espacial de los términos de error que permitan explorar patrones de correlación espacial de los rendimientos intralotes y las propiedades del suelo en los sitios delimitados. From the last decades the use and development of Geographical Information Systems (GIS) and Satellite Positioning Systems (GPS) is highly promoted in cropping systems. Such technologies allow measuring spatial variability of site properties including electrical conductivity and others soil features as well as their impact on the spatial variability of yields. Therefore, site-specific management could be applied to improve the efficiency in the use of agrochemicals, the environmental protection, and the sustainability of the rural life. Currently, there is a wide offer of technological resources to capture spatial variation across sites within field. However, the optimum use of data coming from the precision agriculture machineries strongly depends on the capabilities to explore the information about the complex interactions underlying the productive outputs. The covariation between spatial soil properties and yields from georeferenced data has been treated in a graphical manner or with standard geostatistical approaches. New statistical modeling capabilities from the Mixed Linear Model framework are promising to deal with correlated data such those produced by the precision agriculture. Moreover, rescuing the multivariate nature of the multiple data collected at each site, several multivariate statistical approaches could be crucial tools for data analysis with georeferenced data. Understanding the basis of complex interactions at the scale of production field is now within reach the use of these new techniques. Our main objectives are: (1) to develop new statistical strategies, based on the complementarities of geostatistics and multivariate methods, useful to classify sites within field grown with grain crops and analyze the interrelationships of several soil and yield variables, (2) to propose mixed linear models to predict yield according spatial soil variability and to build contour maps to promote a more sustainable agriculture.

Application of multivariate statistical process control to fuel cell manufacturing

Univariate statistical control charts, such as the Shewhart chart, do not satisfy the requirements for process monitoring on a high volume automated fuel cell manufacturing line. This is because of the number of variables that require monitoring. The risk of elevated false alarms, due to the nature of the process being high volume, can present problems if univariate methods are used. Multivariate statistical methods are discussed as an alternative for process monitoring and control. The research presented is conducted on a manufacturing line which evaluates the performance of a fuel cell. It has three stages of production assembly that contribute to the final end product performance. The product performance is assessed by power and energy measurements, taken at various time points throughout the discharge testing of the fuel cell. The literature review performed on these multivariate techniques are evaluated using individual and batch observations. Modern techniques using multivariate control charts on Hotellings T2 are compared to other multivariate methods, such as Principal Components Analysis (PCA). The latter, PCA, was identified as the most suitable method. Control charts such as, scores, T2 and DModX charts, are constructed from the PCA model. Diagnostic procedures, using Contribution plots, for out of control points that are detected using these control charts, are also discussed. These plots enable the investigator to perform root cause analysis. Multivariate batch techniques are compared to individual observations typically seen on continuous processes. Recommendations, for the introduction of multivariate techniques that would be appropriate for most high volume processes, are also covered.