Multi-class classification of microarray data with repeated measurements: application to cancer Ka Yee Yeung and Roger E. Bumgarner Abstract Prediction of the diagnostic category of a tissue sample from its expression profile and selection of relevant genes for class prediction have important applications in cancer research. We developed the uncorrelated shrunken centroid (USC) and error-weighted, uncorrelated shrunken centroid (EWUSC) algorithms that are applicable to microarray data with any number of classes. We demonstrated that removing highly correlated genes typically improves classification results using a small set of genes. Manuscript: Genome Biology 2003, 4: R83 Correction: Genome Biology 2005, 6:405 (using version 1.0 of our command-line implementation) Supplementary results: pdf file (using version 0.1 of our implementation) Software: MeV+USC and documentation. We have incorporated USC into a user friendly microarray analysis suite MeV, which runs on windows, mac OSX and linux, and requires Java 1.4.2 or above. Executables of our command-line software (including EWUSC and USC) are available as command-line implementations. The java code (written by Ka Yee Yeung) was developed on Red Hat 7.1 using Java SDK1.4. Theoretically, these bytecode files should also run on Windows or MacOS with the Java VM. But we provide no guarantee and offer no support on our software. (You need to install the Java VM before you can run these bytecode files. Download SDK1.4) Documentation (pdf) Bytecode files (version 1.0) updated May 9, 2005. Bytecode files (version 0.1) used to produce results in the original manuscript (2003) Gene expression data sets used: NCI 60 data Original data Ross et al. 2000 Pre-processed data used in the paper: Readme (text) Training set: log expression ratio (tab-delimited text file) Test set: log expression ratio (tab-delimited text file) class file label for each training sample label for each test sample Multiple tumor data Original data Ramaswamy et al. 2001 Pre-processed data used in the paper: Readme (text) Training set: RMA measures and associated standard errors (tab-delimited text file) Test set: RMA measures and associated standard errors (tab-delimited text file) class file Label for each training sample Label for each test sample Breast cancer data van't Veer et al. 2002: We do not have permission to distribute any part of the breast cancer data. The full data (log expression ratios with p-values) are available from Rosetta's web site. Synthetic data sets used: 5 randomly generated synthetic data sets (Windows users: you might want to right click on the links and save the link on your local machine, and then extract the archive.) Readme and visualization (pdf) Table S1 (pdf) at alpha = 2 (high biological noise). Results produced using version 0.1 of our implementation. Table S2 (pdf) at alpha = 0.1 (low biological noise) Results produced using version 0.1 of our implementation. Table S3 (pdf) at alpha = 1 (medium biological noise) Results produced using version 0.1 of our implementation. 4 repeated measurements at alpha = 2: Low technical noise data (lambda = 1) High technical noise data (lambda = 10) 1 (or no) repeated measurement at alpha = 2: Low technical noise data (lambda = 1) High technical noise data (lambda = 10) 20 repeated measurements at alpha = 2: Low technical noise data (lambda = 1) High technical noise data (lambda = 10) 4 repeated measurements at alpha = 0.1: Low technical noise data (lambda = 1) High technical noise data (lambda = 10) 4 repeated measurements at alpha = 1: Low technical noise data (lambda = 1) High technical noise data (lambda = 10) Publicly available software packages used in this work: BioConductor Gist Noble and Pavlidis PAM: prediction analysis for microarrays Tibshirani et al. 2002 RMA package Irizarry et al. 2003