PRESTO, a new tool for integrating large-scale -omics data and discovering disease-specific signatures

Abstract

BackgroundCohesive visualization and interpretation of hyperdimensional, large-scale -omics data is an ongoing challenge, particularly for biologists and clinicians involved in current highly complex sequencing studies. Multivariate studies are often better suited towards non-linear network analysis than differential expression testing. Here, we present PRESTO, a PREdictive Stochastic neighbor embedding Tool for Omics, which allows unsupervised dimensionality reduction of multivariate data matrices with thousands of subjects or conditions. PRESTO is intuitively integrated into an interactive user interface that helps to visualize the multidimensional patterns in genome-wide transcriptomic data from basic science and clinical studies.\n\nResultsPRESTO was tested with multiple input omics platforms, including microarray and proteomics from both mouse and human clinical datasets. PRESTO can analyze up to tens of thousands of genes and shows no increase in processing time with a large number of samples or patients. In complex datasets, such as those with multiple time points, several patient groups, or diverse mouse strains, PRESTO outperformed conventional methods. Core co-expressed gene networks were intuitively grouped in clusters, or gates, after dimensionality reduction and remained consistent across users. Networks were identified and assigned to physiological and pathological functions that cannot be gleaned from conventional bioinformatics analyses. PRESTO detected gene networks from the natural variations among mouse macrophages and human blood leukocytes. We applied PRESTO to clinical transcriptomic and proteomic data from large patient cohorts and detected disease-defining signatures in antibody-mediated kidney transplant rejection, renal cell carcinoma, and relapsing acute myeloid leukemia (AML). In AML, PRESTO confirmed a previously described gene signature and found a new signature of 10 genes that is highly predictive of patient outcome.\n\nConclusionsPRESTO offers an important integration of powerful bioinformatics tools with an interactive user interface that increases data analysis accessibility beyond bioinformaticians and coders. Here, we show that PRESTO out performs conventional methods, such as DE analysis, in multi-dimensional datasets and can identify biologically relevant co-expression gene networks. In paired samples or time points, co-expression networks could be compared for insight into longitudinal regulatory mechanisms. Additionally, PRESTO identified disease-specific signatures in clinical datasets with highly significant diagnostic and prognostic potential.