Abstract 1988: In silico estimates of cell components in cancer tissue based on expression profiling data

• Published in: Cancer research (Chicago, Ill.) Vol. 70; no. 8_Supplement; p. 1988
• Main Authors: Wang, Yipeng, Xia, XiaoQin, Jia, Zhenyu, Sawyers, Anne, Yao, HuaZhen, Wang-Rodriquez, Jessica, Mercola, Dan, McClelland, Michael
• Format: Journal Article
• Language: English
• Published: 15.04.2010
• ISSN: 0008-5472; 1538-7445
• DOI: 10.1158/1538-7445.AM10-1988

Abstract Cancer gene expression profiling studies often measure samples that vary widely in the mixtures of cell types they contain. Such variation could confound efforts to correlate expression with clinical parameters. In principle, the proportion of each major tissue type can be estimated from the profiling data and used to triage samples before using the data to study correlations with disease parameters. Four large gene expression microarray data sets from prostate tissue whose cell components were estimated by pathologists were used to test the performance of in silico prediction of tissue components. Multi-variate linear regression models were developed for in silico prediction of major cell components of prostate cancer tissue. 10-fold cross-validation within each data set gave average differences between the pathologist and in silico predictions of 8∼14% for the tumor component and 13∼17% for stroma component. Across data sets that used similar platforms and fresh frozen samples, the average differences were 11∼12% for tumor and 12∼17% for stroma. Prediction models were applied to expression data on the same platform from 219 other tumor-enriched prostate cancer samples for which tissue proportions were not known. The tumor “enriched” samples were predicted to have a wide range of tumor percentages; 0 to 87%. Furthermore, there was a 10.5% difference in the average predicted tumor percentages between 37 recurrent and 42 non-recurrent cancer patients. This systematic difference in tumor content would likely cause tissue-specific gene changes to falsely appear to be correlated with recurrence unless some samples were excluded to remove this bias or unless tissue percentages were incorporated into the prediction model. Similar circumstances may arise in other sets of clinical samples. A web service, CellPred, has been designed for the in silico prediction of prostate cancer sample cell components. While this site is currently based on microarray data, it could equally well use high-throughput sequencing data. The approach presented here can be generalized to other tissue mixtures once data on both tissue content and expression profiles are obtained for a training set. CellPred is freely available at http://www.webarraydb.org/. Note: This abstract was not presented at the AACR 101st Annual Meeting 2010 because the presenter was unable to attend. Citation Format: {Authors}. {Abstract title} [abstract]. In: Proceedings of the 101st Annual Meeting of the American Association for Cancer Research; 2010 Apr 17-21; Washington, DC. Philadelphia (PA): AACR; Cancer Res 2010;70(8 Suppl):Abstract nr 1988.