|
Activity Number:
|
374
|
|
Type:
|
Contributed
|
|
Date/Time:
|
Wednesday, August 6, 2003 : 10:30 AM to 12:20 PM
|
|
Sponsor:
|
Biopharmaceutical Section
|
| Abstract - #300405 |
|
Title:
|
Kernel Methods for Biological Classification
|
|
Author(s):
|
Robert Bress*+
|
|
Companies:
|
Rensselaer Polytechnic Institute
|
|
Address:
|
587 Broadway, Menands, NY, 12204-2841,
|
|
Keywords:
|
kernel ; classification ; partial least squares regression ; support vector machine ; splice junction
|
|
Abstract:
|
Trying to fully understand the structure and function of genetic material provides us with many complex statistical challenges. Though not everything is known about how protein and DNA structure comes about, there are some flags within DNA and proteins that alert us to the presence of useful genetic information. Since the appropriate classification of this useful information is such a complex endeavor, modeling it successfully requires powerful nonlinear models. The incorporation of kernels into methods such as partial least squares regression and support vector machines allows us to model complex systems similar to the human genome. These methods are considered as alternatives to more typical biological classification techniques such as Hidden Markov Models. It is shown that kernel methods can match the performance of Hidden Markov Models in many cases. Kernel methods also provide a statistical alternative to machine learning techniques that are often considered as black box methods. This allows for an easier understanding and greater acceptance of these techniques. This study will address the recognition of splice junctions with partial least squares in particular.
|
