Output format

The prediction output for server consist of the prediction for both direct (+) and complentary (-) strand. The output lists the predictions for donor and acceptor sites in the submitted sequence.

Position: The position of the splice site in your sequence given as first (donor), or last (acceptor) nucleotide in the intron. The numbering of the direct (+) strand proceeds from the 5' end to the 3' end. For the complement (-) strand the numbering is given in both directions.
Frame: The predicted frame offset (1,2 or 3) of the acceptor/donor site.
Strand: The sequence strand (direct or complement).
Confidence: The level of confidence for the sites (relative to the cutoff used to find nearly all true sites). Sites found by using cutoff values for highly confident sites are marked by the symbol H.
exon^intron: Gives 20 bases of sequence around the predicted site.

Please observe that the lists contain predictions made by TWO detection levels for true sites, one level where around 50% of the true sites are detected with very few false positive, and another level where nearly all true sites are found, but with more false predictions as well. Sites indicated by (H) are highly confident, and represent very seldom a false positive prediction, while those comprising nearly all sites are not marked. The confindence values for the predictions can be compared within each type only. This means that confidence values not marked by (H) in some cases can be larger than those for the (H) marked sites.


The output from the prediction is displayed in the output page of the prediction server. The postscript files can be retrieved directly by Netscape by selecting one of the two references in the bottom of the prediction output. If your viewer is set up to handle postscript, it will display the graphs. Otherwise you can retrieve the compressed postscript files directly to your computer using Netscape.

The top part of the figure designated "Coding" is the activity of an ensemble of coding predicting networks, values close to 0.0 indicate intron region, while values close to 1.0 indicates exon. In the "Donor" panel the activity of the ensemble of the donor site predicting networks is shown as impulses. An impulse with a hight close to 2.0 indicates a strong donor site. A cyan impulse is a prediction that has been discarded during the refinement, and a magenta colored impulse is a prediction that has been changed by the rule based system. The variable threshold computed from the coding predicting ensemble output, is used to select donor and acceptor site predictions.


The predictions in a numerical form may be downloaded from the output page of the prediction server. They are useful for detailed analysis of a sequence. The file produced contains a line starting with the symbol `>' followed by the name and the length of the sequence. This is followed by twelve columns, with the following information given by column number below.
  1. Position in the sequence numbered from 1 to the length of the sequence.
  2. Nucleotides of the sequence.
  3. Neural network donor site score.
  4. Neural network acceptor site score.
  5. Neural network coding score.
  6. Neural network frame score.
  7. High sensitivity level cutoff value for donor site predictions.
  8. High sensitivity level cutoff value for acceptor site predictions.
  9. Confidence of the donor site prediction.
  10. Confidence of the acceptor site prediction.
  11. HMM acceptor site branchpoint score.
  12. Branchpoint position.