EGFR (ErbB-1/HER1) kinase plays an important role in cancer therapy. Two classification
models were established to predict whether a compound is an inhibitor or a decoy of human EGFR
(ErbR-1) by using Kohonen’s self-organizing map (SOM) and support vector machine (SVM). A
dataset containing 1248 ATP binding site inhibitors and 3090 decoys was collected and randomly
divided into a training set (831 inhibitors and 2064 decoys) and a test set (417 inhibitors and 1029
decoys). The descriptors that represent molecular structures were calculated by software ADRIANA.Code. Thirteen
significant descriptors including five global descriptors and eight 2D property autocorrelation descriptors were selected by
Pearson correlation analysis and stepwise analysis. The prediction accuracies on training set and test set are 98.5% and
96.3% for SOM model, 99.0% and 97.0% for SVM model, respectively. Both of these two classification models have
good performance on distinguishing EGFR inhibitors from decoys.
Keywords: Human Epidermal Growth Factor Receptor (EGFR), EGFR inhibitors, classification models, Self-organizing Map
(SOM), Support Vector Machine (SVM).
Rights & PermissionsPrintExport