Are you an EPFL student looking for a semester project?
Work with us on data science and visualisation projects, and deploy your project as an app on top of Graph Search.
It is generally accepted that filtering microRNA (miRNA) target predictions by conservation or by accessibility can reduce the false discovery rate. However, these two strategies are usually not exploited in a combined and flexible manner. Here, we introduce PACCMIT, a flexible method that filters miRNA binding sites by their conservation, accessibility, or both. The improvement in performance obtained with each of these three filters is demonstrated on the prediction of targets for both i) highly and ii) weakly conserved miRNAs, i.e., in two scenarios in which the miRNA-target interactions are subjected to different evolutionary pressures. We show that in the first scenario conservation is a better filter than accessibility (as both sensitivity and precision are higher among the top predictions) and that the combined filter improves performance of PACCMIT even further. In the second scenario, on the other hand, the accessibility filter performs better than both the conservation and combined filters, suggesting that the site conservation is not equally effective in rejecting false positive predictions for all miRNAs. Regarding the quality of the ranking criterion proposed by Robins and Press and used in PACCMIT, it is shown that top ranking interactions correspond to more downregulated proteins than do the lower ranking interactions. Comparison with several other target prediction algorithms shows that the ranking of predictions provided by PACCMIT is at least as good as the ranking generated by other conservation-based methods and considerably better than the energy-based ranking used in other accessibility-based methods.
Romain Christophe Rémy Fleury, Maliheh Khatibi Moghaddam
, , ,
Patrick Gerber, Mélanie Thérèse Marie Hannebelle, Nitin Arora