NetPhorest 2.1 Update
We are happy to announce the release of NetPhorest 2.1, which fixes two major bugs that resulted in arbitrary scores from neural network predictors on SH2 domain, and partially arbitrary scores from neural networks on other domains. PSSM/scansite predictors remain unchanged.
Both web and command line versions were affected. We recommend repeating any analysis performed before May XX, 2017.
Learn more and download the updated binaries on the download page.

Both algorithms work on the same input data, and you can switch at any point of the workflow between the different algorithms.

The input page is designed for the analysis of low-throughput experiments and accepts two different types of input. Possible input formats are protein sequences (fasta format) or identifiers of interest, e.g. Uniprot, ENSEMBL or RefSeq. These identifiers are mapped against the reference set of the STRING database (v. 9.05), from which the sequences are taken. In case of ambiguous identifiers, the potential matches are suggested and the user has to select the appropriate. As identifiers and associated sequences can change with the release of each database version, the matched sequence might not provide a 100% identity. In these cases, we recommend to use the actual sequences as input. Sequences will be mapped to the STRING reference set using BLAST. These assignments will be only used to derive the context score, the motif prediction will use the supplied sequences.

Using KinomeXplorer on mouse

KinomeXplorer was trained on human and yeast data, and it is possible to use KinomeXplorer on their neighbour species, such as mouse.

To use mouse proteins, you need to go through the low-throughput workflow of KinomeXplorer ( You can just put the sequences of your mouse substrate proteins. As a result KinomeXplorer will gives you predicted human kinases on the phosphorylation sites you have chosen. You can easily map human kinases to mouse orthologs using either blast or ortholog prediction servers.

There are many studies that employed NetworKIN in species other than human and yeast, from which you can get some idea.
  • Zanivan, S. et al. In vivo SILAC-based proteomics reveals phosphoproteome changes during mouse skin carcinogenesis.Cell reports 3, 552-66 (2013).
  • Bakal, C. et al. Phosphorylation networks regulating JNK activity in diverse genetic backgrounds. Science (New York, N.Y.) 322,453-6 (2008).
  • Tan, C. et al. Comparative analysis reveals conserved protein phosphorylation networks implicated in multiple diseases.Science signaling 2, ra39 (2009).

High-thoughput experiment data

For the analysis of high-throughput experiments (more than 100 sites), please refer to the high-throughput input page. This workflow also allows for the selection of the exact database version, avoiding the issue of wrong identifier mapping.
In this page, the sites for which the algorithm should predict have to be selected.
First, clicking on any residue marks it for prediction.
For a more streamlined selection, known sites from high-/low-throughput experiments can be pre-selected. In addition, a tab delimited file with the sites of interest can be uploaded.
Finally, a predictor for phosphorylation probability can be used to 1) pre-select site with a certain probability or 2) deselect sites with a low probability.
The result page allows for filtering on mulitple different variables.
The first slider can be used to set a minimum score cut-off for predictions to show.
The second slider sets the maximum distance to the best prediction for that each site. It is most useful to get the highest scoring results for each site, as a fixed score cutoff can be too strict.
The selection boxes allow to en-and disable categorie trees we provide predictions for.
The maximum number of predictions is automaticaly set to improve the efficiency of the initial display. Increasing this number will show more prediction for each site and tree.
The real-time filter option set if the filtering should happen while moving the sliders of on release. Please keep in mind, that for bigger datasets the performance can be too slow to operate the handles precisely.
The save button opens the save-dialog, in which two options can be selected: downloading the whole set of predictions or download the set as it is currently shown.
NetworKIN Each box shows the result for one protein. The first four columns are Site, Tree, Kinase name and the final score. The sequence window is coloured based on the found motif. The more red shaded a residue appears, the higher the influence for the specified kinase. The string path is visualized by showing the number of proteins involved, while the links show the strength of each association. Clicking on the path opens a pop-up showing the nodes involved into the connection.
NetPhorest Each box shows the result of for one protein. The first four columns are Site, Tree, Group name and the final score. The sequence window is coloured based on the found motif. The more red shaded a residue appears, the higher the influence for the specified kinase. Clicking on the group name opens a pop-up showing the tree of the group an thereby all its members.