PrognoScan: a new database for meta-analysis of the prognostic value of genes

Abstract

Background: In cancer research, the association between a gene and clinical outcome suggests the underlying etiology of the disease and consequently can motivate further studies. The recent availability of published cancer microarray datasets with clinical annotation provides the opportunity for linking gene expression to prognosis. However, the data are not easy to access and analyze without an effective analysis platform.

Description: To take advantage of public resources in full, a database named "PrognoScan" has been developed. This is 1) a large collection of publicly available cancer microarray datasets with clinical annotation, as well as 2) a tool for assessing the biological relationship between gene expression and prognosis. PrognoScan employs the minimum P-value approach for grouping patients for survival analysis that finds the optimal cutpoint in continuous gene expression measurement without prior biological knowledge or assumption and, as a result, enables systematic meta-analysis of multiple datasets.

Conclusion: PrognoScan provides a powerful platform for evaluating potential tumor markers and therapeutic targets and would accelerate cancer research. The database is publicly accessible at http://gibk21.bse.kyutech.ac.jp/PrognoScan/index.html.

Figure 1

PrognoScan screenshot and sample search results (part 1). (A) The top page is quite simple and only requires entering the gene identifier(s). (B) Summary table for MKI67, shown here in part (See Additional file 1 for the full table.). Column headings include dataset, cancer type, subtype, endpoint, cohort, contributor, array type, probe ID, number of patients, optimal cutpoint, P_min and P_cor. A statistically significant value of P_cor is given in red font. Each dataset has a link to the public domain where the raw data is archived. By clicking a probe ID in the summary table, a detailed report for the test is displayed. The table can be downloaded in a tab delimited file from the button at bottom.

Figure 2

PrognoScan screenshot and sample search results (part 2). (A) Annotation table. Row headings are color-coded. For example, headings of details such as therapy history, sample type and pathological parameters are highlighted in yellow and basic attributes in blue. (B) Expression plot. Patients are ordered by the expression values of the given gene. The X-axis represents the accumulative number of patients and the Y-axis represents the expression value. Straight lines (cyan) show the optimal cutpoints that dichotomize patients into high (red) and low (blue) expression groups. (C) Expression histogram. The distribution of the expression value is presented where the X-axis represents the number of patients and the Y-axis represents the expression value on the same scale as the expression plot. The line of the optimal cutpoint is also shown (cyan). (D) P-value plot. For each potential cutpoint of expression measurement, patients are dichotomized and survival difference between high and low expression groups is calculated by log-rank test. The X-axis represents the accumulative number of patients on the same scale as the expression plot and the Y-axis represents raw P-values on a log scale. The cutpoint to minimize the P-value is determined and indicated by the cyan line. The gray line indicates the 5% significance level. (E) Kaplan-Meier plot. Survival curves for high (red) and low (blue) expression groups dichotomized at the optimal cutpoint are plotted. The X-axis represents time and the Y-axis represents survival rate. 95% confidence intervals for each group are also indicated by dotted lines.

Figure 3

Kaplan-Meier plots for high and low SIX1-expressing groups in breast cancers.

Figure 4

Kaplan-Meier plots for high and low MCTS1-expressing groups in breast, lung, blood and brain cancers.