Figure 3

ROC curves of SPACE structure prediction (Affymetrix sample dataset of human tissues). ROC curves that test SPACE performance for structure prediction. They measure the concordance of the predictions with the transcript structures that appear in Ensembl release 51. The y-axis shows the sensitivity (the proportion of probes that are said to hybridize against a transcript and in fact do) and in the x-axis 1 - specificity is shown (the proportion of probes that are said not to hybridize against a transcript and in fact they do not). A total of 15 genes that present alternative splicing between tissues and were validated by RT-PCR (figures 3 and 4 of de la Grange et al. [8]) were chosen to make the ROC curves. Two genes SNX13 and IDE (also validated in the same figures) were not included because alternative splicing events shown in the PCRs do not appear in Ensembl release 51. Another two genes ABLIM1 and MICAL2 were validated in two different splicing events. The number of transcripts of each gene in the tissue dataset was estimated from the PCR figures. The first ROC curve corresponds to the prediction without taking into account the improvements described in this paper. The second ROC curve was obtained after applying Adapt algorithm and structure coherence correction (probes within the same exon or exon part must have the same hybridization pattern). The ROC curves of each of the 15 genes are shown in grey as background.