Proportion of predicted number of transcripts in a simulation performed with 600 genes (synthetic data). The 600 genes used were randomly selected from the human genome with one to six transcripts. Annotated transcript structures in Brainarray version 11 CDFs were used to make the synthetic data for each gene. Probe affinities were estimated from real data (Affymetrix sample dataset for human tissues). Both additive and multiplicative noise mimics the noise present in real datasets. Transcript concentrations have been randomly assigned from a uniform distribution across samples multiplied for a proportional factor (also from a uniform distribution) between transcripts, the resulting concentrations are such that some transcripts present lower concentrations in all samples compared with the concentrations of the most significant transcript of the same gene. The area of each circle represents the proportion of times the corresponding predicted number is chosen by the algorithm for a given number of transcripts. Genes with a single transcript are wrongly predicted to have AS (several transcripts) only 1%.