SpliceRover is a prediction tool which can be used for donor and acceptor splice site prediction, in both human and arabidopsis. By making use of convolutional neural networks, we achieve state-of-the-art accuracy.

All implementation details can be found in our publication at .

The input for this tool is required to be in a fasta format. A minimum sequence length of 398 is needed for the human models, and one of 402 for the arabidopsis models. The maximum length is limited at 30000 .

Data

All four available models are trained on datasets described in the aforementioned publication.

The human donors & acceptors models are trained on the GWH dataset as described in Sonnenburg et al (2007), with a pos:neg sub sampling of 1:10. The sequence length is 398.

The arabidopsis donors & acceptors models are trained on the arabidopsis dataset as described in Degroeve et al (2005), retaining the original pos:neg ratio from the publication. The sequence length is 402.