Motivation: Algorithms predicting microRNA (miR)-mRNA interactions generate high numbers of possible interactions, many of which might be non-existent or irrelevant in a certain biological context. It is desirable to develop a transparent, user-friendly, unbiased tool to enrich miR-mRNA predictions.
Results: The miMsg algorithm uses matched miR/mRNA expression data to enrich miR-mRNA predictions. It grades interactions by the number, magnitude and significance of misplacements in the combined ranking profiles of miR/mRNA expression assessed over multiple biological samples. miMsg requires minimal user input and makes no statistical assumptions. It identified 921 out of 56 262 interactions as top scoring and significant in an actual germ cell cancer dataset. Twenty-eight miR-mRNA pairs were deemed of highest interest based on ranking by miMsg and supported by current knowledge about validated interactions and biological function. To conclude, miMsg is an effective algorithm to reduce a high number of predicted interactions to a small set of high confidence interactions for further study.
Availability and implementation: Matlab source code and datasets available at www.martinrijlaarsdam.nl/mimsg .
Supplementary information: Supplementary data are available at Bioinformatics online.