CS&E Profile: Rui Kuang

Computational Biology, Machine Learning

Ph.D. 2006, Computer Science, Columbia University
M.S. 2002, Computer Science, Temple University
B.S. 1999, Computer Science, Nankai University

About

Assistant Professor Kuang joined the department in the fall of 2006 and specializes in computational biology (bioinformatics) and machine learning. He has co-authored eight refereed publications for various conferences and journals.

Kuang’s research applies specifically to advanced learning techniques, including kernel methods and network diffusion algorithms, to address remote homology detection problems and related fundamental learning obstacles in protein structure prediction and analysis.

Research

As the complete genomes of more and more species become available, the large-scale study of protein structure is becoming increasingly important. Given the difficulty of experimentally determining protein structure with X-Ray crystallography or nuclear magnetic resonance (NMR), much effort over the past decade has been devoted to the development of new machine learning approaches for tackling the challenging problem of protein structure inference. One particular problem is to predict the structural class of a protein sequence, relative to a fixed taxonomy of protein structural classes, using a database of proteins with known structures as training data. A more general problem is to retrieve a ranked list of protein sequences from a large sequence database that are related to a query sequence, with the hope that any annotations for the retrieved sequences might help identify the structural or functional properties of the query. A difficult special case of both these problems is the remote homology detection problem, where the query sequence is only remotely related to the database sequences and one wants to detect this remote evolutionary relationship from subtle sequence similarity. Moreover, a protein sequence consists of one or more domains --- subsequences representing separate functional evolutionary units that are thought to fold autonomically --- and segmenting a protein sequence into domains is also a challenging unsolved problem.

We apply several advanced learning techniques, including kernel methods and network diffusion algorithms, to address the remote homology detection problem and several related fundamental learning problems in protein structure inference.