Find related research by combining models of experimental data

In a proof of concept study, HIIT researchers have shown that it is possible to find relevant earlier research direcly based on the experimental data, better than by the standard keyword searches on annotations of the data. The results corresponded well with citations between the journal papers in which the data sets had been first publicized, to the extent of even finding errors in the data. The search is scalable, with linear complexity, given models learned separately for each data set. The work was done with European Bioinformatics Institute EBI of EMBL, and published in PLOS ONE.

Read the paper