LEARNING DAGS USING MULTICLASS SUPPORT VECTOR MACHINES

In this paper we consider the problem of learning the genetic-interaction-network that is underlying the measured double knockout (DK) data. Based on the biological system model of [3], we propose a multiclass-SVM approach that yields a high prediction accuracy of the genetic-interaction-network underlying the DK data while being able to estimate the network topology for large sets of genes. We demonstrate the performance of our proposed multiclass-SVM approach by synthetic data simulations where we use the recently proposed GENIE method of [3] as a benchmark.