ClassificationLungCancerMethylationMarker_04

ClassificationLungCancerMethylationMarker_04 - Journal of...

Info iconThis preview shows pages 1–2. Sign up to view the full content.

View Full Document Right Arrow Icon
Classification of Individual Lung Cancer Cell Lines Based on DNA Methylation Markers Use of Linear Discriminant Analysis and Artificial Neural Networks Alberto M. Marchevsky,* Jeffrey A. Tsou, and Ite A. Laird-Offringa From the Department of Pathology and Laboratory Medicine, * Cedars-Sinai Medical Center, Los Angeles, California; and Departments of Surgery and of Biochemistry and Molecular Biology, Norris Cancer Center, Keck School of Medicine, University of Southern California, Los Angeles, California The classification of small cell lung cancer (SCLC) and non-small cell lung cancer (NSCLC) can pose diagnos- tic problems due to inter-observer variability and other limitations of histopathology. There is an inter- est in developing classificatory models of lung neo- plasms based on the analysis of multivariate molecu- lar data with statistical methods and/or neural networks. DNA methylation levels at 20 loci were measured in 41 SCLC and 46 NSCLC cell lines with the quantitative real-time PCR method MethyLight. The data were analyzed with artificial neural networks (ANN) and linear discriminant analysis (LDA) to clas- sify the cell lines into SCLC or into NSCLC. Models used either data from all 20 loci, or from five signif- icant DNA methylation loci that were selected by a step-wise back-propagation procedure ( PTGS2 , CALCA , MTHFR , ESR1 , and CDKN2A ). The data were sorted randomly by cell line into 10 different data sets, each with training and testing subsets composed of 71 and 16 of the cases, respectively. Ten ANN mod- els were trained using the 10 data sets: five using 20 variables, and five using the five variables selected by step-wise back-propagation. The ANN models with 20 input variables correctly classified 100% of the cell lines, while the models with only five variables cor- rectly classified 87 to 100% of cases. For comparison, 10 different LDA models were trained and tested using the same data sets with either the original data or with logarithmically transformed data. Again, half of the models used all 20 variables while the others used only the five significant variables. LDA models pro- vided correct classifications in 62.5% to 87.5% of cases. The classifications provided by all of the differ- ent models were compared with kappa statistics, yielding kappa values ranging from 0.25 to 1.0. We conclude that ANN models based on DNA methylation profiles can objectively classify SCLC and NSCLC cells lines with substantial to perfect concordance, while LDA models based on DNA methylation profiles provide poor to substantial concordance. Our work supports the promise of ANN analysis of DNA methylation data as a powerful approach for the development of automated methods for lung cancer classification. (J Mol Diagn 2004, 6:28–36) There is an increasing interest in the use of molecular data from human neoplasms for diagnostic and prognos- tic purposes. 1 A prodigious amount of data regarding the molecular composition of a variety of neoplasms is being collected from tissue samples. To our knowledge, it is not
Background image of page 1

Info iconThis preview has intentionally blurred sections. Sign up to view the full version.

View Full DocumentRight Arrow Icon
Image of page 2
This is the end of the preview. Sign up to access the rest of the document.

This note was uploaded on 04/06/2010 for the course COMPUTER S COSC1520 taught by Professor Paul during the Spring '09 term at York University.

Page1 / 9

ClassificationLungCancerMethylationMarker_04 - Journal of...

This preview shows document pages 1 - 2. Sign up to view the full document.

View Full Document Right Arrow Icon
Ask a homework question - tutors are online