Abstract
Maturation of mRNA in eukaryotes is a very complex process that includes exon recognition through specific elements such as branch point motifs , 5' and 3' splice sites and splicing regulators. Mutations may affect these signals either directly by disrupting constitutive splice sites or indirectly by creating cryptic splice sites. We thus evaluated the prediction efficiency of nine software programs based on Position Weight Matrices, Markov Models, Maximal Dependence Decomposition, Neural Networks or Feature Generation Algorithms using 623 mutations for which the consequence on mRNA has been demonstrated in vitro. Position Weight Matrix-based tools were the most efficient in predicting the impact of a given mutation. Overall, at least one system correctly predicted 100% of mutations affecting invariant positions as well as-1, +3 and +5 positions of the 5'ss. Deep intronic mutations resulting in the activation of cryptic exons were almost all correctly predicted (92.31%), while other intronic mutations were less efficiently (70-80%). Exonic mutations that create cryptic splice sites were also efficiently detected (70%). Because of the prediction heterogeneity and specificity, a single tool could not be used for all predictions. Moreover, since these systems are all based on text analysis an in vitro validation step is still required.