The JSI splice site prediction tool predicts changes in the quality of splice sites at and in close proximity to a site of genetic variation. For this, several scores are calculated, among them scores taking into account not only the overall quality of the known splice motif but also the probability of the respective sequence being present as a known splice site throughout the whole genome. The JSI splice site prediction tool has been trained with approximately 200 000 splice sites, i.e. the known splice sites throughout the whole genome (GRCh37).

Additionally, the MaxEntScan scoring tool for human splice sites has been integrated and the respective scores are displayed alongside the JSI scores for comparison. We would like to thank Gene Yeo for his friendly approval to integrate the MaxEntScan scoring algorithm into our software.

Splice Sites were predicted for the wild type genomic reference sequence of several genes.

For every GT and AG in the sequence, the scores for a 5' or 3' splice site were calculated, respectively. A score above zero indicates that the site is predicted to be a splice site, below zero means it is not predicted to be a splice site.

As the authentic and alternative splice sites are known for the gene, the predicted results can then be compared with the actual splice sites.

Additionally, the MaxENT algorithm was tested in the same manner to compare the predicted results. For MaxENT, a score above 3.5 was interpreted as a predicted splice site.

For authentic (including alternative) splice sites, the JSI tool correctly predicted more (usually all) splice sites than MaxENT. In cases where MaxENT also predicted all authentic splice sites of a gene correctly, the JSI tool's results were the same.
Hence, the JSI splice site prediction tool has a lower false negative rate.

Inversely, the JSI splice site tool is configured to be more sensitive, resulting in the prediction of more cryptic splice sites with positive scores than MaxENT. Due to the higher detection rate, the possibility to miss authentic splice sites is reduced. The higher number of false positive calls is adjusted afterwards. The scores of each called splice site are compared between the reference and the altered sequence. Thus, calls of non-authentic splice sites with positive scores can easily be filtered out in this step if their score is the same for reference and altered sequence.

Please refer to appendix 1 for the detailed result data of five example genes.

For the second test, the JSI Splice Site Prediction Tool was used to predict the effects of known splicing variants. All obvious splicing variants with alterations of the GT (5') or AG (3') of authentic splice sites are always displayed.

It was then intensely recherched for thoroughly validated splicing variants that are not caused by alterations of the highly conserved GT or AG bases. Compared to the GT or AG-affecting splice sites, fewer validated variants were found. This may be due to the fact that variants not directly affecting the GT or AG of a splice site were often not suspected to be splice-altering and therefore not examined for such an effect, along with the general expense of a thorough evaluation.

The JSI Splice Site Prediction Tool prove to be very accurate when being used to predict effects of variants not altering the consensus GT or AG.
Especially for 3' splice sites, the prediction was more accurate than the one of MaxENT. This is due to the fact that the JSI Splice Site Prediction Tool analyses a longer sequence than MaxENT. Whilst MaxENT only analyses 20 bases upstream of the AG, the JSI Splice Site Prediction Tool includes a wider range of upstream bases for the additional features of a 3' splice site. Thereby, it is able to detect changes in the polypyrimidine tract, branch point and AG exception zone.

Moreover, the JSI Splice Site Prediction Tool compares the scores of all possible splice sites surrounding the variant. Thus, it can predict when a new or cryptic splice site becomes a competitor of an authentic splice site.

Furthermore, the tool prove capable of detecting alterations in splice site strength (example 8 in appendix 2). There, the score was decreased due to the variant, but without becoming lower than the scores of surrounding cryptic splice sites. The prediction of this "medium decrease" in score matches the actual effect of a weakened splice site that results in lowered amounts of mRNA transcript.

Appendix 2 contains examples from the list of validated variants that were used to compare predicted and actual effect. To view the detailed scores, also of MaxENT, please click the given link on the gene name, which will open another browser tab with the splice site prediction query for the respective variant.

Splice Sites were predicted for the wild type genomic reference sequence of several genes.

For every GT and AG in the sequence, the scores for a 5' or 3' splice site were calculated, respectively. A score above zero indicates that the site is predicted to be a splice site, below zero means it is not predicted to be a splice site.

As the authentic and alternative splice sites are known for the gene, the predicted results can then be compared with the actual splice sites.

Additionally, the MaxENT algorithm was tested in the same manner to compare the predicted results. For MaxENT, a score above 3.5 was interpreted as a predicted splice site.