Phenotype Association Versioning

Many factors contribute to the identification of phenodeviants by statistical analysis. This includes the number of mutant and baseline mice, the statistical test used, the selected thresholds and changes to the underlying software that runs the analysis. For these reasons, we will be versioning genotype-to-phenotype associations from data release to data release. A given genotype-to-phenotype may change from release to release.

Statistical Tests

In general, we are applying a Fisher Exact Test for categorical data and linear regression for continuous data. In cases where there is no variability in values for a data parameter in a control or mutant mouse group, a rank sum test is applied instead of a linear regression. The statistical test used is always noted when displayed on the portal or when obtained by the API. Documentation on statistical analysis is available here: Statistics help

P-value threshold

In this first release, we are using a p value threshold of ≤ 1 x10-4 for all statistical tests to make a phenotype call. This threshold may be adjusted for some parameters upon further review by statistical experts.

Clinical Blood Chemistry and Hematology

Review of PhenStat calls for clinical blood chemistry and hematology by phenotypers at WTSI suggest our current analysis maybe giving a high false positive rate. Alternative statistical approaches are being considered. We suggest looking at the underlying data that supports a phenotype association if it's critical to your research.