The cellular function of kinases combined with the difficulty of designing selective small molecule kinase inhibitors (SMKIs) poses a challenge for drug development. The late-stage attrition of SMKIs could be lessened by integrating safety information of kinases into the lead optimization stage of drug development. Herein, a mathematical model to predict bone marrow toxicity (BMT) is presented which enables the rational design of SMKIs away from this safety liability. A specific example highlights how this model identifies critical structural modifications to avoid BMT. The model was built using a novel algorithm, which selects 19 representative kinases from a panel of 277 based upon their ATP-binding pocket sequences and ability to predict BMT in vivo for 48 SMKIs. A support vector machine classifier was trained on the selected kinases and accurately predicts BMT with 74% accuracy. The model provides an efficient method for understanding SMKI-induced in vivo BMT earlier in drug discovery.