Large Data, Wide Spaces: Designing Scalable and Flexible Graphics Infrastructure with Applications to Genomics

• Activity Number: 267
• Type: Invited
• Date/Time: Tuesday, July 31, 2012 : 8:30 AM to 10:20 AM
• Sponsor: Section on Statistical Computing
• Author(s): Michael Lawrence*+
• Companies: Genentech
• Keywords: graphics ; scalability ; pipeline ; genomics

Abstract

Creating an interactive visualization for large data, on the scale of millions or more, requires attention to both the design and implementation of the graphics. This talk is concerned with the latter, in particular software infrastructure for fast, scalable interactive graphics. The implementation of an interactive visualization consists of both drawing and computation (filtering, summarizing, transforming, etc), and a graphics system needs to scale in both respects. For drawing, we have developed a hardware-accelerated renderer, with multi-layered buffers for incremental updating and efficient spatial algorithms for mapping user actions to the data. It is low-level and applicable to a variety of graphics problems. Towards scalable computation, we have applied the model-view-controller pattern to the design of computational pipelines that operate dynamically on subsets of the data, and makes feasible the flexible and rapid development of algorithms implemented in the R language. We demonstrate a system for viewing high-throughput genomic data, which is large in both the number of records (e.g., billions of sequence alignments) and the size of the domain (the length of a genome).