Curated compendium of human transcriptional biomarker data

One important use of genome-wide transcriptional profiles is to identify relationships between transcription levels and patient outcomes. These translational insights can guide the development of biomarkers for clinical application. Data from thousands of translational-biomarker studies have been deposited in public repositories, enabling reuse. However, data-reuse efforts require considerable time and expertise because transcriptional data are generated using heterogeneous profiling technologies, preprocessed using diverse normalization procedures, and annotated in non-standard ways. To address this problem, we curated 45 publicly available, translational-biomarker datasets from a variety of human diseases. To increase the data's utility, we reprocessed the raw expression data using a uniform computational pipeline, addressed quality-control problems, mapped the clinical annotations to a controlled vocabulary, and prepared consistently structured, analysis-ready data files. These data, along with scripts we used to prepare the data, are available in a public repository. We believe these data will be particularly useful to researchers seeking to perform benchmarking studies—for example, to compare and optimize machine-learning algorithms' ability to predict biomedical outcomes.