The prediction of protein function at CASP6

In the CASP6 experiment, the new "Function Prediction" category was tentatively introduced. Predictors were asked to provide functional information on the CASP targets, many of which were of unknown function. This article describes the setup of the experiment and its results, highlighting what was learned from it, and suggesting modifications to its format for the next rounds. The obvious limitation of such an experiment is that the results cannot be assessed in the standard CASP fashion, as all targets remain of unknown function. Furthermore, we had to face the expected difficulties due to the novelty of the experiment and to the problems connected with function definition. Nevertheless, and even with a limited number of participating groups, we believe that the results of the experiment can be useful both for its future and for experimentalists working on the functional assignment of the CASP6 targets. We found that, in a few cases, a consensus functional prediction could be derived for targets of unknown function. However, our analysis suggests that a general description of the method used should be made available together with the predictions so that a higher reliability can be assigned to cases where completely independent methods give the same or similar predictions.