Congress detail

Authors: Larese, Mónica G.; Granitto, Pablo M.; Gómez, Juan C.

Resumen: Spotted microarray images present high variability in their quality due to intrinsic factors arising at the manufacturing process. Bad quality spots must be filtered at early steps to avoid wrong conclusions in subsequent data analysis. Microarray analysis tools let human experts manually  flag out bad spots, which is a tedious and error prone procedure. These packages also provide automatic  flagging, but this is limited to computing several morphological and statistical measures which are intended to be later combined and thresholded. Only a few works propose to find a computational model for classification. In this work, ensemble algorithms are proposed to perform bad/good quality spots discrimination. Four ensemble methods are implemented to solve this binary classification problem, namely Discrete, Real and Gentle AdaBoost, and Random Forests. The same seven features per channel (Cy3 and Cy5) proposed by Hautaniemi et al. and used by Bicego et al. are computed for each spot. These features are: spot intensity, background intensity, alignment error, roundness, spot size, background noise and bleeding. A publicly available dataset is used, which consists of 155 spots extracted from two diu000bfferent microarray images. These spots were unanimously labeled by three human experts which have several years of experience dealing with microarray experiments. The classification performances are compared to those obtained by Hautaniemi et al. and Bicego et al., and show that ensembles can improve the accuracy of the discrimination. .

Meeting type: Congreso.

Production: Detection of bad quality spots in cDNA microarray images.

Scientific meeting: 2do. Congreso Argentino de Bioinformática y Biología Computacional.

Meeting place: Córdoba.

Organizing Institution: Asociación Argentina de Bioinformática y Biología Computacional.

It's published?: No

Meeting month: 5

Year: 2011.