670 likes | 813 Views
A Random Walk Based Approach for Improving Interaction Network and Increasing Prediction Accuracy. Chengwei LEI, Ph.D. Assistant Professor of Computer Science Department of Electrical Engineering and Computer Science McNeese State University. What is Interaction Network.
E N D
A Random Walk Based Approachfor Improving Interaction Network and Increasing Prediction Accuracy Chengwei LEI, Ph.D. Assistant Professor of Computer Science Department of Electrical Engineering and Computer Science McNeese State University
What is Interaction Network • Interaction network is a network of nodes that are connected by features.
First Introduced in Biology • If the feature is a physical and molecular, the interaction network is molecular interactions usually found in cells.
Real-world Classification Good Bad • Noisy data • Overfitting problem • Few true “driver” changes / vast number of “passenger” changes.
Current Methods Classifier Prediction
Current Methods Statistical test Pick the most significant ones Classifier Prediction
Problem? • Ignore the relationships between nodes/features/sensors
Our approach • Improve prognosis by combining • Node readout data • Node-node interaction networks
Classifier Prediction
Network Transformation Matrix
Network Transformation Matrix
Network Classifier Transformation Matrix Prediction
Transformation Matrix • Transformation matrix is generated by apply the Random Walk with Restart (RWR) algorithm on the Interaction network.
Random Walk • A random walk is a mathematical formalization of a path that consists of a succession of random steps.
Random Walk • A random walk is a mathematical formalization of a path that consists of a succession of random steps. • Random walk for one node on a graph G is a walk on G where the next node is chosen uniformly at random from the set of neighbors of the current node • when the walk is at node v, the probability to move in the next step to the neighbor u is Pvu = 1/d(v) for (v, u) is connected and 0 otherwise.
Random Walk Step 1 Step 1
Random Walk Step 2 Step 2
Random Walk Step 3
Random Walk Step 1 Step 2 Step 3 …… Step N
Random Walk with Restart • A random walker start from a node (v) with • uniform probability to visit its neighbors • fixed probability c to revisit the start node (v) • The probability for a random walker to be on node j after k times is • fijk(v) is the probability for a random walker to take path i to j at time k • Fj(v) at equilibrium is the probability for a random walker starting from node v to reach node j => Similarity between patient v and j
Experiments • Biology Data • Cancer prediction
Wang’s Dataset Network Transformation Matrix
Wang’s Dataset 10144 286 10144 286 10144 7885 7885 7885 2259
286 286 Good Bad 7885 7885 2259 T-test 1247
286 Good Bad 286 7885 7885 2259 1678 T-test