SBC logo Authors: Amine Heddad, Andrea Krings, Markus Brameier and Bob MacCallum, Stockholm Bioinformatics Center, Stockholm University, Sweden.

NucPred

Fetching P62753 from www.uniprot.org...

The NucPred score for your sequence is 0.87 (see score help below)

   1  MKLNISFPATGCQKLIEVDDERKLRTFYEKRMATEVAADALGEEWKGYVV    50
  51  RISGGNDKQGFPMKQGVLTHGRVRLLLSKGHSCYRPRRTGERKRKSVRGC   100
 101  IVDANLSVLNLVIVKKGEKDIPGLTDTTVPRRLGPKRASRIRKLFNLSKE   150
 151  DDVRQYVVRKPLNKEGKKPRTKAPKIQRLVTPRVLQHKRRRIALKKQRTK   200
 201  KNKEEAAEYAKLLAKRMKEAKEKRQEQIAKRRRLSSLRASTSKSESSQK    249

Positively and negatively influencing subsequences are coloured according to the following scale:

(non-nuclear) negative ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||| positive (nuclear)

with NucPred



If you find NucPred useful, please cite this paper:
NucPred - Predicting Nuclear Localization of Proteins. Brameier M, Krings A, Maccallum RM. Bioinformatics, 2007. PubMed id: 17332022
The authors also look forward to your comments and suggestions.

What does the NucPred score mean?

You have to decide on a NucPred score threshold. Sequences which score greater than or equal to this threshold are predicted to spend some time in the nucleus. Higher thresholds yield fewer predicted nuclear proteins, but these predictions are more accurate (you can have higher confidence in them). The table below gives more details of the performance of NucPred estimated using the sequences it was trained on (by cross-validation). Another benchmark is available in the Bioinformatics 2007 paper.

NucPred score threshold Specificity Sensitivity
see above fraction of proteins predicted to be nuclear that actually are nuclear fraction of true nuclear proteins that are predicted (coverage)
0.10 0.45 0.88
0.20 0.52 0.83
0.30 0.57 0.77
0.40 0.63 0.69
0.50 0.70 0.62
0.60 0.71 0.53
0.70 0.81 0.44
0.80 0.84 0.32
0.90 0.88 0.21
1.00 1.00 0.02

Sequences which score >= 0.8 with NucPred and which are predicted by PredictNLS to contain an NLS have been shown to be 93% correct with a coverage of 16%. (PredictNLS by itself is 87% correct with 26% coverage on the same data.)

Go back to the NucPred Home Page.