Entry for:The Bioinformatics Peer Prize
Biowheel (https://biowheel.dibsvis.com/) is a web-based award-winning data visualization tool, for exploring high-dimensional and heterogeneous biomedical data. Through interactive sorting and filtering of data, Biowheel enables researchers to quickly detect data outliers, evaluate data consistency, and discover mixed trends. Its interactive data presentation, visually-engaging design, and friendly user interface opens the door to easier, faster and better high-dimensional data interpretation for biomedical professionals with and without programming training.
Biowheel rapidly displays any tabular data in a circular heatmap, where columns are represented as rings, rows as spokes, and values as colors. The video at the top of this page illustrates examples of visualizing high-dimensional molecular time-course data from the HPN-DREAM Breast Cancer challenge.
Leveraging its interactive capabilities, Biowheel facilitates data vetting and pattern discovery. Data values are displayed in real-time via tooltips when users mouse over any data-associated graphical element, offering an additional layer of information. With a simple click on a ring’s name, Biowheel will automatically sort (or unsort) samples based on values of the corresponding variable and will reorder spokes instantaneously. This interactive sorting enables fast visual comparison and pattern discovery.
Fully harnessing the pattern recognition power in human eyes, Biowheel integrates its interactive visualization platform with a semi-supervised clustering algorithm (MPCK-Means) to enable visually guided clustering. Clustering solutions are iteratively refined based on user-defined and visually inspired cluster constraints in the form of must-links and cannot-links between sample pairs.
Biowheel shows a broad ability to visualize time-course data from experiments and drug trials (Figure 1, bioRxiv manuscript), as well as clinical covariates and proteomics data from cancer patients (Figure 2, bioRxiv manuscript). To illustrate the utility of visually guided clustering in Biowheel, we have successfully applied it to increase the accuracy of recognizing Iris species groups using the publicly available Iris dataset (https://archive.ics.uci.edu/ml/datasets.html), and to generate patient clusters with more homogeneous protein expression patterns using the DREAM Acute Myeloid Leukemia Outcome Prediction (https://qutublab.org/apps-code-tools/#DREAM) challenge data.
Biowheel serves as a visual interface for both unsupervised and semi-supervised clustering tasks. Differentiating its design from other bioinformatic tools and data visualization programs, Biowheel is fully interactive, programming-free, and drives data interpretation through interactive display, filtering and sorting of the raw data. These features make Biowheel an ideal tool for biological reseachers working with high-dimensional time-course, and/or expression level data.
5. Future ideas/collaborators needed to further research?
Biowheel has been applied broadly by researchers across domains: bioengineering, clinical research, fitness, data science, space research, oil & gas and smart-building design. We welcome new applications of the tool, and feedback for new features and refinement of the software.
Biowheel is free academically at academic.dibsvis.com. Standalone interactive visualizations derived from Biowheel for data from the HPN-DREAM Breast Cancer Challenge are available online http://dream8.dibsbiotech.com/ (Hill et al., 2016 Nature Methods). Enterprise software is available through https://biowheel.dibsvis.com/ and directly contacting us.