In silico prediction of spleen tyrosine kinase inhibitors using machine learning approaches and an optimized molecular descriptor subset generated by recursive feature elimination method

Comput Biol Med. 2013 May;43(4):395-404. doi: 10.1016/j.compbiomed.2013.01.015. Epub 2013 Feb 10.


We tested four machine learning methods, support vector machine (SVM), k-nearest neighbor, back-propagation neural network and C4.5 decision tree for their capability in predicting spleen tyrosine kinase (Syk) inhibitors by using 2592 compounds which are more diverse than those in other studies. The recursive feature elimination method was used for improving prediction performance and selecting molecular descriptors responsible for distinguishing Syk inhibitors and non-inhibitors. Among four machine learning models, SVM produces the best performance at 99.18% for inhibitors and 98.82% for non-inhibitors, respectively, indicating that the SVM is potentially useful for facilitating the discovery of Syk inhibitors.

Publication types

  • Research Support, Non-U.S. Gov't

MeSH terms

  • Algorithms
  • Arthritis, Rheumatoid / metabolism
  • Computer Simulation
  • Decision Trees
  • Humans
  • Intracellular Signaling Peptides and Proteins / antagonists & inhibitors*
  • Models, Statistical
  • Protein Kinase Inhibitors / pharmacology*
  • Protein-Tyrosine Kinases / antagonists & inhibitors*
  • Reproducibility of Results
  • Spleen / drug effects
  • Spleen / enzymology*
  • Support Vector Machine*
  • Syk Kinase


  • Intracellular Signaling Peptides and Proteins
  • Protein Kinase Inhibitors
  • Protein-Tyrosine Kinases
  • SYK protein, human
  • Syk Kinase