Fig. 2

Data flow diagram showing the major steps in our extended supervised link prediction algorithm to predict in vivo drug sensitivity. (a) The training and test data are provided to the extended supervised link prediction algorithm. (b) A feature vector construction method is applied to the training and test data, to obtain new feature representations of the training and test data. (c) A link filtering algorithm is applied to the new feature representation of the training data, to yield subsampled data. (d) A feature selection step is applied to subsampled data, to obtain subsampled data with fewer features (i.e., genes). (e) A learning algorithm takes as input the subsampled data with fewer features, to induce the model h. (f) The features in the test data are selected using the same positions as in the training data and the model h is applied to the test data with the selected features, to yield predictions