Panos G Georgopoulos

US Patent:

20060271300, Nov 30, 2006

Filed:

Jul 29, 2004

Appl. No.:

10/565417

Inventors:

William Welsh - Princeton NJ, US
Ming Ouyang - Holmdel NJ, US
Paul Lioy - Canford NJ, US
Panos Georgopoulos - Princeton NJ, US

International Classification:

G06F 19/00

US Classification:

702019000, 702020000

Abstract:

Clustering is routinely applied in the exploratory analysis of microarray data. Missing entries arise from blemishes on the microarrays. The present invention provides a new method, and computer program and/or computer product thereof to impute missing values. The method involves the steps of clustering microarray data by partitioning the data into a select number of clusters, wherein each data point is iteratively moved from one cluster to another, until two consecutive iterations have resulted in the same partition pattern; obtaining a select number of estimates of the data in the clusters by probabilistic interference; and averaging the select number of estimates to obtain missing values in the microarray data. The method is superior to other imputation models as measured by root mean squared errors.