生科系三乙专题讨论进度应遵守的规范及软体和器材的使用.PPT

Extracting tree-way gene interactions from microarray data 郭士逢 輔仁大學生命科學系 9 / 25 / 2008 Outlines Introduction Method of extracting gene interactions Verify the triplets by using testing data Potential causes for the interaction Discussion Motivation High-throughput genomic data Large amount of information hidden there Microarray gene expression profiling data Too simplistic to access gene interactions by clustering using pairwise correlation as distance metric Co-expression Co-expressed gene pairs are rare. Restricted to special condition or process Ex. Special organ, specific developmental stage, after a drug treatment, or particular disease state. Some of them only exist in cancer cell A dynamic network Each link represents a significant pairwise correlation between two genes. Alternative approach Assuming a third gene (controller) that can affect co-expression relationship. Looking a three-way interaction Why not four or more? Number of combinations too large Most nodes in network with sparse connection, except some hubs Main issue How to screen and select gene triplets, in which correlation of two genes significantly associated with the expression level of a controller gene? Data Microarray dataset containing 678 samples from human cancer tissue and cell lines GEO, raw data, human, oncology, Affymetrix, quantile normalization Splitting data randomly into training set and testing set of equal sizes Triplets identified by training data validated in testing data Identify bimodal gene Determine whether log-expression of a gene is a single normal or a mixture of two normal distribution. If bimodal model fit better, find the threshold value to split the sample. At least 60 samples in subgroup Evaluate interactions Consider a gene triplet A, B and C According bimodal expression of controller C, split sample to two sub-groups: High and Low. Correlations between A and B in Low C group is r1, and r2 in High C. Transform ri to normal zi, and test if z1=z2. If z1≠Z2, we got