|
|
Methods for class prediction with high-dimensional gene expression data
Šilhavá, Jana ; Matula, Petr (oponent) ; Železný, Filip (oponent) ; Smrž, Pavel (vedoucí práce)
This thesis deals with class prediction with high-dimensional gene expression data. During the last decade, an increasing amount of genomic data has become available. Combining gene expression data with other data can be useful in clinical management, where it can improve the prediction of disease prognosis. The main part of this thesis is aimed at combining gene expression data with clinical data. We use logistic regression models that can be built through various regularized techniques. Generalized linear models enable us to combine models with different structure of data. It is shown that such a combination may yield more accurate predictions than those obtained based on the use of gene expression or clinical data alone. Suggested approaches are not computationally intensive. Evaluations are performed with simulated data sets in different settings and then with real benchmark data sets. The work also characterizes an additional predictive value of microarrays. The thesis includes a comparison of selected features of gene expression classifiers built up in five different breast cancer data sets. Finally, a feature selection that combines gene expression data with gene ontology information is proposed.
|
|
|
Nowcasting the Czech Trade Balance
Babecká Kucharčuková, Oxana ; Brůha, Jan
V tomto článku se zabýváme "teďpovědí" (nowcasting) a krátkodobou předpovědí hlavních veličin zahraničního obchodu. Uvažujeme čtyři empirické metody: regresi založenou na hlavních komponentách, elastickou síť, dynamický faktorový model a metodu částečných nejmenších čtverců. Diskutujeme, jak lze tyto metody adaptovat na situaci řad s různou frekvencí a nesynchronní publikací. Srovnáváme tyto metody s jednorozměrnými metodami. Ukazujeme, že pro nominální a reálné dovozy i vývozy dává nejpřesnější predikce metoda elastické sítě následovaná dynamickým faktorovým modelem a metodou hlavních komponent. Pro dovozní a vývozní ceny dávají nejlepší predikce pro "zpětpověď" (backcasting) a teďpověď jednorozměrné metody, pro krátkodobou předpověď však přesnější výsledky poskytují sofistikovanější metody. I zde dominuje nad ostatními metodami elastická síť.
Plný text:  PDF
|
|
|
Methods for class prediction with high-dimensional gene expression data
Šilhavá, Jana ; Matula, Petr (oponent) ; Železný, Filip (oponent) ; Smrž, Pavel (vedoucí práce)
This thesis deals with class prediction with high-dimensional gene expression data. During the last decade, an increasing amount of genomic data has become available. Combining gene expression data with other data can be useful in clinical management, where it can improve the prediction of disease prognosis. The main part of this thesis is aimed at combining gene expression data with clinical data. We use logistic regression models that can be built through various regularized techniques. Generalized linear models enable us to combine models with different structure of data. It is shown that such a combination may yield more accurate predictions than those obtained based on the use of gene expression or clinical data alone. Suggested approaches are not computationally intensive. Evaluations are performed with simulated data sets in different settings and then with real benchmark data sets. The work also characterizes an additional predictive value of microarrays. The thesis includes a comparison of selected features of gene expression classifiers built up in five different breast cancer data sets. Finally, a feature selection that combines gene expression data with gene ontology information is proposed.
|
host ::
RSS.