Gene expression–based prediction of copy number aberrations (CNAs) in glioblastoma. (A) The maximum (red) and mean (blue) amplification probabilities on chromosome 12q14–12q15 were estimated from glioblastoma gene expression data by a hidden Markov model (HMM). The interrogated region flanked by CDK4 and MDM2 encompasses a ∼11 Mb window. (B) The respective mean DNA copy number of this chromosomal region was determined by array comparative genomic hybridization (aCGH, Humarray 3.0 and 3.1 of the University of California at San Francisco; manuscript in preparation). The bacterial artificial chromosome (BAC) probes are ordered by their genomic positions. The BACs corresponding to CDK4 and MDM2 BACs are shown in red. (C) The heat map visualizes the structure of the aCGH data shown in (B) for the genomic region encompassing 12q13 to 12q15 from 68 glioblastomas. The BACs are ordered by their genomic position, while the glioblastomas on the x-axis are ordered by similarity using Sorting Points into Neighborhood software. Blue depicts deletion; red, amplification; and white, missing data. The color scale is truncated to [−1, 1] for presentation. The BAC corresponding to CDK4 is GS-561N1; for MDM2, CTB-136O14 (red); and for WIF1, RP11-18B8 (blue). (D) WIF1 expression (Affymetrix probe set 204712_at) in glioblastoma is significantly lower than in nonneoplastic brain tissues (P = .001), as determined in our gene expression data set. (E) Low WIF1 expression was confirmed in 5 independent glioblastoma data sets (Freije et al., red; Rich et al., violet; Phillips et al., orange; Sun et al., dark blue; Horvath et al., green; our data set, Murat et al., light blue).– WIF1 expression values are median centered within each data set independently.