Survival Time Definition: Define right censored outcomes in terms of start and end dates/times, and a censor date/time or indicator.

Kaplan-Meier Plots:

• Plot one or more survival curves, based on categorical sample covariates or continuous-valued data, such as gene expression values. For the latter, the expression range may be flexibly split into two or more ranges (such as tertiles) to define groups.
• View sample information for selected individuals (corresponding to censored tick marks, or events) from the plot, or to all individuals in a group.
• Customize with color-coded line, standard errors (Peto or Greenwood), medians, numbers at risk, time scale in days, weeks, months or years, and plot labels. Invert the plot or convert to a log scale.
• Divide patients into strata (e.g. gender-specific) for separate Kaplan-Meier analysis.

Logrank Tests: Tests of homogeneity and trend across groups.

Cox Regression: Multivariate analysis of survival outcome based on expression profiles and/or clinical covariates. Allows for testing of prognostic significance of genes while adjusting for other known prognostic factors. Stepwise Cox regression selects genes that are independently predictive of outcome.

GeneScreen: Screen all genes for association with survival outcome, using Cox regression. Optionally adjust analysis for known prognostic factors. Base selection of significant genes on a defined false discovery rate. Use a leave-n-out approach to predict outcome for cases not in the analysis data set.

2-Group and k-Group: Supervised clustering:

• Selection of predictive genes using univariate statistics (e.g. t-tests, ANOVA) with GeneScreen, followed by unsupervised clustering with k-mean, self-organizing maps, Bayesian classification or hierarchical clustering.
• Nearest shrunken centroids method for classification on a training set, an independent test set and/or a leave-n-out set.
• Support Vector Machines for k-way classification and testing on an independent set.

Metacluster: Cumulation of results from repetitive clustering to average over many local minima, or to use a leave-n-out strategy for independently testing classification performance.

Link to Outcome: The results of clustering or classification can be directly assessed with respect to outcome using thumbnail Kaplan-Meier survival plots for each cluster or class.

Line Graphs: Show sequential changes in expression during a trial or experiment

• Optionally stratified by treatment or other characteristic.
• Optional standard error bars.
• Find genes with similar patterns.
• Display multiple genes in single plot, or as a tiled display with one gene per tile.

General Tools:

• Pathways: Displays pathways (derived from KEGG, GenMAPP or custom built), with gene expression (or covariate) data superimposed. Click for information on a gene or a metabolic product. Annotated with text or thumbnail graphics that summarize key properties of the genes.
• Attribute analysis: Attributes are gene classifications that include GO codes, protein classification and pathway membership, in tree structure. Genes in a list are compared to all genes to determine the attributes over-represented (with odds ratio and statistical significance).
• SNPs and Genomics: Download genomic sequences, select genomic regions to display (5' promoter, exons, introns and/or 3' tail), align two or more sequences or probe sets, BLAST against full genomes, find matches to a motif, compare occurrences of a selected motif in a set of genes to a control set, highlight conserved and/or repeat genomic regions, display known SNPs.

Cancer-Specific Tools:

• Karyotype description is automatically interpreted to mark up a chromosome ideogram to indicate regions of loss, duplication and sites of translocation.
• Loss-of-heterozygosity analysis using SNP array data, showing regions of LOH for individual samples or aggregated over a group.
• Genomic amplification based on CGH data shown on the chromosome ideogram.

Diagnoses/Classification: Results from classification/clustering methods, and metacluster runs can be applied to a separate set of cases to predict class.

Outcome: Results from multivariate Cox regressions, with optimal gene selection, can be applied to a separate set of cases to predict outcome.