Research article

Neural network analysis of lymphoma microarray data: prognosis and diagnosis near-perfect

Michael C O'Neill and Li Song

Department of Biological Sciences, University of Maryland, Baltimore County Baltimore, Maryland 21250, USA

author email corresponding author email

BMC Bioinformatics 2003, 4:13doi:10.1186/1471-2105-4-13

Published:	10 April 2003

Abstract

Background

Microarray chips are being rapidly deployed as a major tool in genomic research. To date most of the analysis of the enormous amount of information provided on these chips has relied on clustering techniques and other standard statistical procedures. These methods, particularly with regard to cancer patient prognosis, have generally been inadequate in providing the reduced gene subsets required for perfect classification.

Results

Networks trained on microarray data from DLBCL lymphoma patients have, for the first time, been able to predict the long-term survival of individual patients with 100% accuracy. Other networks were able to distinguish DLBCL lymphoma donors from other donors, including donors with other lymphomas, with 99% accuracy. Differentiating the trained network can narrow the gene profile to less than three dozen genes for each classification.

Conclusions

Here we show that artificial neural networks are a superior tool for digesting microarray data both with regard to making distinctions based on the data and with regard to providing very specific reference as to which genes were most important in making the correct distinction in each case.