Canal-U

3.9. Benchmarking the prediction methods

It is necessary to underline that gene predictors produce predictions. Predictions mean that you have no guarantees that the coding sequences, the coding regions,the genes you get when applying your algorithm, are true genes, thatis genes which have a biological existence. Only experimental analysiscan confirm or infirm your predictions. Nevertheless it is interesting and also important to be able to evaluate your algorithm, thisis the role of benchmarking. Benchmarking means measuring the capacity of your algorithm to produce good predictions. How can we make thiskind of measurement? We need a reference, an idealreference would be a genome which is well annotated and for whichall of the annotations have been confirmed through experimental results. Unfortunately, there are very few genomes for which we have this experimental confirmation.Even, for example, the E. coli genome, which is a well-knownorganism and well annotated genome, is not the ideal reference. However, it is also interesting to compare prediction algorithm and method between them, to do acompetition, to apply several predictors on the same genomes and to compare the results of this algorithm.