|To submit Genes in Experiment and Ranks either upload a text file or paste into the text area:
Format: Each gene id is followed by a number indicating differential expression, a p-value, rank, or a number that can be converted into a rank. Genes are separated by LINEBREAK or SEMICOLON, gene and rank are separtated by WHITESPACE or COMMA.
|Available GO gene-association databases |
Type of statistics:
Minimal length of considered GO paths:
e.g. "biological_process"=1, "biological_process%behavior"=2
Subset of GO hierarchy:
Limit search to subset of GO hierarchy that contains a keyword,
e.g. "biological_process", "molecular_function", "cellular_component"
Maximal p-value in GO output list:
Maximum number of GOs/groups to display:
Show GOs with constistant LOW/HIGH ranks:
-1 => do not cluster.
Merge GOs if indicating gene lists are inclusions or differ by less than # genes
Correct for multiple testing:
Find significant GOs in ranked gene list.|
The input is a list of genes of a complete microarray or experiment with an accompanying number. The number specifies a rank or a p-value and indicates, for example, differential expression. If the number is not a rank, it is converted into one. The output is similar to GOstat. The Total represents the mean rank or mean input p-value. A P-Value for each GO is calculated
There is three different options for test statistics:
- (recommended) performs a Wilcoxon Signed Rank test. This test will determine whether the ranks for a particular GO are significantly higher or lower than usual.
- uses a Kolmogorov-Smirnov test to test whether the ranks of all genes associated with a GO follow uniform distribution, i.e. are randomly distributed within the complete list of genes or are biased towards low or high ranks/p-values.
This approach is similar to the "Gene set enrichment analysis" described by Mootha et al (Nature Genetics, 2003).
The main difference to 1 is that KS will test for any deviations from uniformity (i.e. also those GOs with ranks in the middle) while the Wilcoxon test only considders high or low ranks.
- you may do a Kolmogorov-Smirnov test based on p-values, in this case the input has to be p-values and is left as it is. In this case it is assumed that p-values are usually uniformly distributed between 0 and 1 and Kolmogorov-Smirnov tests are performed to find GOs where the distribution of p-values differs from this. However, care should be taken with this option as results of significance tests on microarray data do not often fulfill the requirement of producing uniformly distributed p-values, but may be biased towards 0 or 1.