We noticed that the ensemble method consistently predicted some individuals as relatively older or younger than their chronological age across many bootstrap re-samplings and by almost all of the individual classifiers within the ensemble (data not shown). It is unclear if these individuals were incorrectly predicted due to poor performance of the classifier, or because these individuals had a markedly different biological age compared to their chronological age. Future studies should attempt to tease apart these differences by collecting additional health-related markers.
BART allows a scientist with no bioinformatics background to extract knowledge from their own microarray data or microarray experiments available from GEO. BART is functional on more microarray experiments and provides more comprehensive analyses than extant microarray analysis tools. BART is hosted on bart.salk.edu, includes a user tutorial, and is available for download from _amaral/bart.
BART contains six modules that enable users to process raw microarray data from GEO or locally into a list of differential genes and associated pathways, enabling everyone to interpret microarray data in terms of underlying biological processes. Figure 1 summarizes the workflow from data import from CEL, GEO accession, or data matrix, through grouping by variable or feature, batch effect correction, normalization, visualization with heatmaps/PCA, differential expression testing, and finally functional enrichment. Users can access BART from our dedicated server (bart.salk.edu), or by downloading the associated R  Shiny code and running it locally.
A thorough and reliable microarray analysis tool is essential to scientists who are interested in extracting knowledge from the more than 50,000 microarray experiments on GEO, or from their experiments. BART is a free and powerful online microarray analysis tools that allows users without bioinformatics knowledge to analyze microarray data starting from GSE accession ids from GEO, raw CEL files, or expression tables. In addition to flexible input, users can specify custom sample groupings, specify batch ids for downstream correction, generate full lists of all differentially expressed genes between any pairwise comparison using the LIMMA modeling package, and check for enriched pathways among differentially expressed genes. All data tables, heatmaps, PCA plots, and volcano plots are available for download (Fig. 1). We designed BART to be more powerful and flexible than current microarray tools to facilitate meaningful interpretation of array data. BART is uniquely capable of processing raw CEL files and performing RMA normalization instead of relying on preprocessed expression tables. In addition, BART is the only comprehensive tool that offers batch effect correction. BART provides a simple interface and comprehensive analysis for any scientist interested in analyzing microarray data (Table 1). In addition, the flexibility and wealth of features allows users to improve analyses of datasets with batch effects (Fig. 2), or when custom comparisons are required (Fig. 3). BART code is available from _amaral/bart and is hosted at bart.salk.edu. 2b1af7f3a8