Average results of classification performed on the Test Set with NDC and samples that developed AD.

When performing on the independent test set with NDC and AD samples, Gómez Ravetti and Moscato [3] still hold the best results, obtaining an average of 92.3% of accuracy. Th e signature obtained by just selecting features to complement Gómez Ravetti and Moscato's [3] signature, S1, almost matched Ray et al.'s [1] sensitivity, even though it did not perform so well in terms of specificity. It includes EGF, IL-1, IL-3, TNF-, G-CSF, “BLC-RANTES”, “MIP-1d-IL-11”, “TNF- - ANG-2”, “TNF- - FAS” and “IL-11-I-TAC”. The second signature, obtained in the same manner after discarding samples, performed significantly better, almost matching Ray et al.'s in accuracy. It includes EGF, IL-1, TNF-, G-CSF, “EGF-IGFBP-2”, “GM-CSF-IL-1”, “IL-1-IL-11”, “MIP-1d-NT-3”, “PDGF-BB-VEGF-B” and “TNF--ANG-2”. Interestingly, the last signature, S3, obtained by just discarding the single features of S2, yields very similar results. That supports the theory that the (single) features were not playing a key role in distinguishing between AD and NDC on S2, and that the meta-features indeed hold useful information for that purpose.