Fig. 3.

Gene weight layer. Gene feature was combined as the sum of normalized gene expression value and a weight adjusted genetic mutation value. Two ways of generating the mutation component were tested. a Single weight layer: a common weight vector $w^{\prime }$ was applied to all genetic mutations without considering the heterogeneity of gene-drug relationship. b Multiple weight layers: a weight matrix $W^{\prime }$ was applied to account for the heterogeneity of gene-drug relationship so that each drug would have its own weight vector for genetic mutation