Santa Theresa Square
6017 Snell Ave
Paste sequences (FASTA format) or protein names (one per line) below.Example:#1, #2, #3
(For protein names, HGNC symbol is recommended.)
In case you have more than 100 sequences, please consider using the high throughput workflow. Please fill out the form below.
Here you can download the precomputed data for all available kinase predictors against ENSEMBL version 59 on known phosphorylation sites from the KinomeXplorer-DB.
|Human Ensembl v59||Version 59 of Ensembl protein sequences for human||NetworKIN predictions||NetPhorest scores|
Here are the changes to the algorihtms since the 2014 publication.
- NetPhorest 2.1 (2017-05-02)
- Fix: SH2 neural network predictors were registered incorrectly and the scores were fully arbitrary (determined by the compiler).
- Fix: neural networks for all domains were registered incorrectly if they had no hidden layers, resulting in a shift of the weights. Neural network scores are calculated as an average of 5-12 neural networks with mixed architectures. Approx. 15% of all networks have no hidden layers and produced arbitrary predictions (determined by the compiler). Nearly all nn predictors are affected to various degrees, except the following which had hidden layers for all neural networks: KIN_CaMKIV, KIN_CaMKI_group, KIN_FLT3_CSF1R_Kit_PDGFR_group, KIN_IKKalpha_IKKbeta_group, KIN_KDR_FLT1_group, KIN_Met_group, KIN_PKC_group, KIN_ROCK_group, KIN_Syk_group, KIN_TTK, PTP_PTPN13, 1433_any_group. PSSM (scansite) scores are not affected by this bug.
- Fix: possible buffer overflow with very long fasta names.
- The yeast release now behaves like the human: no prior filtering, and skips sites with unsupported amino acids.
- The NetPhorest binary now stops with an error message if it finds non-ASCII characters in the protein sequence.