ABSTRACT
Patient disease subtypes have the potential to transform personalized medicine. However, many patient subtypes derived from unsupervised clustering analyses on high-dimensional datasets are not replicable across multiple datasets, limiting their clinical utility. We present CoINcIDE, a novel methodological framework for the discovery of patient subtypes across multiple datasets that requires no between-dataset transformations. We also present a high-quality database collection, curatedBreastData, with over 2,500 breast cancer gene expression samples. We use CoINcIDE to discover novel breast and ovarian cancer subtypes with prognostic significance and novel hypothesized ovarian therapeutic targets across multiple datasets. CoINcIDE and curatedBreastData are available as R packages.
Subject(s)
Computational Biology/methods , Datasets as Topic , Software , Algorithms , Breast Neoplasms/diagnosis , Breast Neoplasms/genetics , Cluster Analysis , Computer Simulation , Female , Gene Expression Profiling , Humans , Ovarian Neoplasms/diagnosis , Ovarian Neoplasms/genetics , Prognosis , ROC CurveABSTRACT
We outline a paradigm for meta-microarray database creation and integration with clinical variables. We use as our implementation example a breast cancer database linking RNA expression measurements (by microarray) and clinical variables, such as survival metrics and tumor size. Such an endeavor involves integrating across different microarray datasets as well as clinical parameters. To this end, we created a data curation and processing pipeline, formal database ontology, and SQL schema to optimally query, analyze and visualize data from over 30 publicly available breast cancer microarray studies listed in the Gene Expression Omnibus (GEO). We demonstrate several pilot examples using this database. This methodology serves as a model for future meta-analyses of complex public clinical datasets, in particular those in the field of cancer.
ABSTRACT
PURPOSE: To assess the temporal sampling requirements needed for quantitative analysis of dynamic contrast-enhanced MRI (DCE-MRI) data with a reference region (RR) model in human breast cancer. MATERIALS AND METHODS: Simulations were used to study errors in pharmacokinetic parameters (K(trans) and v(e)) estimated by the RR model using six DCE-MRI acquisitions over a range of pharmacokinetic parameter values, arterial input functions, and temporal samplings. DCE-MRI data were acquired on 12 breast cancer patients and parameters were estimated using the native resolution data (16.4 seconds) and compared to downsampled 32.8-second and 65.6-second data. RESULTS: Simulations show that, in the majority of parameter combinations, the RR model results in an error less than 20% in the extracted parameters with temporal sampling as poor as 35.6 seconds. The experimental results show a high correlation between K(trans) and v(e) estimates from data acquired at 16.4-second temporal resolution compared to the downsampled 32.8-second data: the slope of the regression line was 1.025 (95% confidence interval [CI]: 1.021, 1.029), Pearson's correlation r = 0.943 (95% CI: 0.940, 0.945) for K(trans), and 1.023 (95% CI: 1.021. 1.025), r = 0.979 (95% CI: 0.978, 0.980) for v(e). For the 64-second temporal resolution data the results were: 0.890 (95% CI: 0.894, 0.905), r = 0.8645, (95% CI: 0.858, 0.871) for K(trans), and 1.041 (95% CI: 1.039, 1.043), r = 0.970 (95% CI: 0.968, 0.971) for v(e). CONCLUSION: RR analysis allows for a significant reduction in temporal sampling requirements and this lends itself to analyze DCE-MRI data acquired in practical situations.
