|
AVVISO DI SEMINARIO La Prof. Regina LIU (Department of Statistics, Rutgers University) il giorno 6 Luglio 2004 terrà presso il Dipartimento di Statistica, Probabilità e Statistiche Applicate dell'Università di Roma "La Sapienza", un seminario sul tema: "CLUSTERING GENES BASED ON P-VALUES" 6 Luglio 2004 - Sala 34 – ore 12,00 Tutti gli interessati sono invitati a partecipare. abstract: Clustering genes based on p-values Regina Liu (rliu@stat.rutgers.edu) Department of Statistics, Rutgers University, USA Clustering is an important task in the analysis of microarray gene expression data. It helps identifying groups of co-regulated genes and assigning functions to genes. In many microarray experiments, gene expression data consists of repeated measures in multiple conditions. However, this experimental setup is often ignored in most existing approaches for clustering genes. In this talk we propose a new clustering methodology which can account for both the exact experimental setup and the variability of the data. This methodology compares genes by testing the equality of the condition-mean vectors and condition-variances across experimental conditions, and uses the resulting P-value as a measure of similarity between two (or two groups of) genes for the purpose of clustering. The proposal of using P-value as a measure of similarity is new and can be easily understood as a standardized assessment of the similarity between genes. It is less arbitrary than existing choices such as Euclidean distance, correlation, etc. Our P-value based clustering method provides an intuitive statistical framework for the problem of clustering gene expression data with repeated measurements. This is an advantage over most existing clustering algorithm which are generally exploratory in nature. In addition, the P-value approach can be easily calibrated for different separating criteria by adopting different tests. Moreover, our clustering uses the combined P-values as validations for the final clusters, and hence has the built-in ability to carry out automatic model selection. We apply our clustering methodology to both simulated and real data sets. The preliminary findings appear to be quite supportive of the proposed methodology. This is joint work with Rebecka Jornsten and Jun Li, Rutgers University. |