Seminar by Sayan Ranu
Mining Significant Subgraphs and their Applications to Drug Discovery
Sayan Ranu
IBM, Bangalore
Date: Wednesday, January 9th, 2013
Time: 12NOON
Venue: CS101.Abstract:
Given a graph database, how can you mine subgraphs that are significantly more frequent than its expected frequency? Furthermore, if each graph in the database is labeled with a binary class, how can you identify subgraphs that are highly frequent in the positive class, but infrequent in the negative class? More precisely, how can you mine statistically significant subgraphs? In this talk, I will present a technique to mine statistically significant subgraphs that have p-values below a user-provided threshold. To demonstrate the utility of mining significant subgraphs, I will highlight their immense potential in identifying over-represented substructures from molecular databases and their subsequent application in making molecular activity prediction.
Towards the later half of my talk, I will briefly go over my current research project at IBM and discuss the problem of performing similarity searches on trajectories of moving objects. While few techniques exist to compute similarities between trajectories, they fail to negotiate the noise inherent in GPS-traces. To combat this void, I have developed a distance function that employs the unique concept of "projections" and automatically adopts to noise and non-uniform sampling rates.
Finally, I will conclude by outlining my research goals spanning the areas of dynamic network analysis and bioinformatics.