Fig. 1

Contribution of features to SVM performance. a TargetExpress score prediction distribution, with positive numbers representing predicted True targets for a particular expression context. Features consist of: b Expression, c TargetScan Total Context+, d microT-CDS miTG, e TargetScan Probability of Conserved Targeting (PCT), f MIRZA, g MIRZA-F and h MIRZA-N. Scaled score ranges for each feature ([−1. -0.5], [−0.5, 0], [0, 0.5], [0.5, 1]) used to build the SVM models contribute differently to predict True targets. When there are transcripts with no information for a particular feature (C-H), we assign them the scaled min-max centre, in this case 0