Number of errors from the 18-genes randomly selected signatures on the “AD” validation test set.

The Random forest algorithm was used as classifier. For each signature 10 runs with different seeds were done. We used the WEKA software implementation, and the algorithm was allowed to generate 150 trees. The best and worst signatures are highlighted in bold text. In two cases we found signatures that classify above 90%, comparable with the results of Ray et al. that report on 91% AD predictability as a result of their proposed methodology.