Support Vector Machine Classification of Microarray Data

Abstract

The Problem: Use the learning from examples paradigm to make class predictions and infer genes involved in these predictions from DNA microarray expression data. Specifically, we use a Support Vector Machine (SVM) classifier [6] to predict cancer morphologies and treatment success and determine the relevant genes in the inference. Motivation: Previous Work: Ageneric approach to classifying two types of acute leukemias was introduced in Golub et. al. [3]. SVM’s have been applied to this problem [5] and also to the problem of predicting functional roles of uncharacterized yeast ORF’s [1]. Approach: We used a SVM classifier to discriminate between two types of leukemia. The output of classical SVM’s isaclassdesignation ±1. Inthisparticularapplication it is important to be able to reject points for which the classifier is not confident enough. We introduced a confidence interval on the output of the SVM that allows us to reject points with low confidence values. It is also important in this application to infer which genes are important for the classification. We have preliminary results for a feature selection algorithm for SVM classifiers. The SVM was trained on the 38 points in the training set and tested on the 34 points in the test set. Our results (see table 2 and figure (1)) are the best reported so far for this dataset. genes rejects errors confidence level |d|

Keywords

Related Publications