Development of a Pharmacogenomics Model based on Support Vector Regression with Optimal Features Selection Approach to Determine the Initial Therapeutic Dose of Warfarin Anticoagulant Drug

Introduction: Using artificial intelligence tools in pharmacogenomics is one of the latest bioinformatics research fields. One of the most important drugs that determining its initial therapeutic dose is difficult is the anticoagulant warfarin. Warfarin is an oral anticoagulant that, due to its narrow therapeutic window and complex interrelationships of individual factors, the selection of its optimal dose is challenging. Method: Inaccuracy in determining the initial dose of warfarin will simply lead to thrombosis or severe bleeding and ultimately, patient death. Among the relatively successful methods of kernel-based estimation, comparison and identification of suitable kernels have not been researched. In the present research, while carefully examining this approach, different features of selection algorithms were analyzed based on expert opinions, and an appropriate subset of efficient predictor variables was identified for dose estimation. Results: In the current study, a dataset collected by the International Warfarin Consortium was used. The results showed that the support vector machine with a suitable kernel and a subset of the proposed features can successfully predict the ideal dose of warfarin for a significant percentage of patients with an error of approximately 0.7 mg per week. Conclusion: The estimation was conducted using the least squares version of the support vector regression based on a suitable kernel and feature selection strategy. In this way, a better approach for predicting the optimal therapeutic dose of warfarin was presented, which can significantly reduce the wrong dose error and its consequences.