Welcome to the Github repository for the KinPred v1.0 project, which seeks to create sustainable and usable formats of whole human proteome predictions of kinases with site-specific substrates. Details of code are available within the Code and Data directory readme files. For orientation, the likely workflows you may be interested in, include:

Users of KinPred-formatted Prediction Algorithm Resources

Researchers wishing to understand how to use the Final Data Matrices of kinase-substrate predictions, should consider the following possible workflows:

1. Get the Final Data of KinPred v1.0 resource. This matrices, in .CSV formats, include the controlled ontology for substrates rows (site-specific phosphorylation sites in human proteins) and kinases predicted by that resource columns, with matrix(i,j) values where the entry is the edge weight between substrate i and kinase j. These results are not filtered by stringency.
2. See example types of analysis code that produced figures and analysis for our KinPred publication. These ipython notebooks can be found under Code/Comparison.
3. When developers update these resources, this documentation will be updated to reflect the latest version of the datasets.

Developers of kinase-substrate prediction algorithms

KinPred specifically recommends that to create sustainable and maximally usable results, you perform the following steps:

1. Run the algorithm for your set of kinases on the KinPred reference human proteome. Publish this proteome file and ALL predictions in a permanent location, using the same Substrate ID and Kinase ID unified ontology. For KinPred v1.0 publication, these whole-proteome formats are published for NetworKIN, GPS, and PhosphoPICK here. See example code for these projects in Code/PreProcessingPredictionData. The vignette.md shows how to use KinPred code to:
   • replicate the preprocessed prediction data for PhosphoPICK, NetworKIN, and GPS
   • incorporate the provided python modules to process the prediction data of other kinase-substrate predictor to the standarlized format.
2. Use ProteomeScout's current reference phosphoproteome to filter raw predictions for known phosphorylation sites. See Code/CrossReferenceWithProteomeScout
3. Provide an easy-to-use matrix format of these outputs, include all edges (as we noted in our publication, removing edges based on a stringency cutoff will cause serious limitations in the flexible use of your predictions). Definitions of matrix files can be seen for Final Data, here

KinPred and Prediction Algorithm Developers: Updating prediction data

There are two types of updates that a kinase-substrate prediction resource will undergo.

Phosphoproteome Update

When the phosphoproteome has been updated, take the full list of whole-proteome predictions and filter on the new phosphoproteome using Code/CrossReferenceWithProteomeScout/XRefProteomeScout.py. See Code/UpdatingResourceData/UpdateNewPhosphosites.ipynb for instruction.

Underlying Reference Proteome Update

The underlying Uniprot database of the human reference proteome often undergoes changes, such that the substrate and kinase relationships of an old reference are now invalid. Since running predictions anew on the entire reference proteome is costly, we have provided scripts to identify changes in the reference proteome that have an effect on predictions (i.e. the sequence or position of a phosphorylation site has changed), and that site position in the ontology is updated, or a subset of sequences must be run and these replace the existing predictions for that protein in the Formatted, list output of prediction resources. Code/UpdatingResourceData/UpdateHumanProteomeVersion.ipynb shows the steps to update prediction data with the upgrade from 2019-12-12 to 2020-2-26 reference humam proteome.