Fig. 5

The framework of our research. The ChEMBL25 Dataset consists of 58,846 compounds, 382 kinases, and 95,462 kinase-compound bioactivity data points. The Metz Kinase profiling dataset contains 1421 compounds, 172 kinases, and 100,786 bioactivity data points [26]. Then, statistic criteria are used to select compounds that can be regarded as efficient inhibitors for kinase family. This selection process leads to inhibitors that cover eight kinase families. For each compound, we generated 238 moiety-based features for constructing DNN models for these kinase families. The SHAP approach is applied to uncover the preference moieties of each kinase family