User Information

Virtual Kinome Profiler (VKP) is a web application that enables users to virtually profile their compound of interest across a set of 248 molecular kinases. Being a comprehensive statistical and classification, VKP has not been designed for batch studies. VKP has been checked for compatibility with standard modern browsers (Mozilla Firefox, Google Chrome, and Safari). Since VKP utilizes a session management system, it is advisable to make sure that browser privacy settings allow cookies to be stored.

I/O formats

Input: Virtual Kinome Profiler only requires the SMILES description of the compound to perform the profiling process. The users are requested to submit the SMILES notation of the compound with a chosen compound ID, delimited by a single space. Please see the example input.

Output: The output from Virtual Kinome Profiler is a comprehensive enumeration of likelihoods describing the association of the compound with a list of 248 kinases. These kinase targets and their corresponding HGNC, gene nomenclatures are hyperlinked to the UniProt Databases. Each column enumerates the statistical measure of similarity obtained from different molecular representations summarized using the prediction score and a normalized SVM score that can be utilized to classify the compounds activity classes as 1-binders and 0-non-binders. The output can be downloaded as a .csv or .pdf file for further analysis.

Data repository

The curated data resources and the accompanying files utilized in implementing the computational framework can be downloaded as a compressed repository for local use. These files include:

  • interaction.csv file documenting the kinase-specific interaction data, including information of compounds (ChEMBL-ID’s), targets (UniProt-ID’s), the activity measures in nM.
  • kinase_ligand_sets.pkl file listing the set of active and inactive compound sets associated with each kinase targets. Wherein the kinases Uniport ID’s are used as primary keys to retrieve their respective compound sets.
  • compound_structures.pkl compound structure file, where SMILES information for each compound is documented in an RDKit molecule format. This information can be retrieved by using the compounds ChEMBL-ID’s as primary keys.
  • HDF5 files enumerating the optimal thresholds (actives and inactives) optimized during the model building process and are necessary to estimate the statistical measures associations.
  • A model_eSVM.pkl listing the set of 23 ensemble SVM classifier models trained and utilized in the classification process. Wherein each classifier can be accessed using the key ‘ENS_1’ etc.
All files that are not necessary for the implementation but gives additional information regarding the data repository can be downloaded from the Supplementary information.

Batch Profiling

Owning to their run time complexities, the current version of Virtual Kinome Profiler is not adept for batch profiling process. In cases where such formulation is required, the users are advised to download the accompanying data repository and customize the source codes provided in GitHub to enable their implementation. Further, a local implementation of VKP requires stringent dependency requisites listed below.

Dependencies

The backend computational framework of Virtual Kinome Profiler was programmed using Python (version 2.7.12) and largely depends on the following prerequisites for standalone implementation.

  • RDKit (version 2016.09.1)
  • Pandas (version 0.20.1)
  • NumPy (version 1.12.1)
  • Scipy (version 0.19.0)
  • Scikit-learn (version 0.18.1)
  • cPickle & sklearn
For a comprehensive list of requirements, please visit the GitHub Virtual Kinome Profiler repository.