Figure 13

SVMs predict preferential expression for some tissues, such as the brain and the adipose tissue. The average preferential expression measure (PEM _avg ) expresses the degree of preferential expression of genes expressed in a given tissue. An individual PEM value for each transcript equals its expression in a given tissue divided by its average expression across all tissues. This being a continuous variable, prediction accuracy is the correlation between SVM’s predicted value of the response variable and its observed value in the half of the dataset designated for prediction (the other half was used for training). The highest prediction accuracy was achieved for the adipose tissue and the brain tissue cluster (which were also the two tissues with the highest degree of tissue-specific expression). Overall, there was a strong correlation between the overall degree of preferential expression in a given tissue (PEM _avg ) and the power to predict preferential expression in this tissue (Spearman’s correlation of 0.543). For technical details of the SVMs used see Methods.