Nanouniverse: Virtual Instancing of Structural Detail and Adaptive Shell Mapping

TL;DR

Nanouniverse tackles the memory bottleneck in atomistic biological visualization by introducing virtual instancing through proxy geometries, Wang tiling, and a three-level acceleration structure for Ray Tracing. The system uses an adaptive shell space and a core space to render mesostructures both on the surface and inside proxy geometries, with on-the-fly transformation matrices that avoid storing complete atomistic data. The authors demonstrate interactive, fully detailed renderings of scenes containing trillions of atoms (e.g., Red Blood Cells and SARS-CoV-2 virions) using minimal GPU memory and a multi-level traversal that reduces intersection tests versus conventional two-level schemes. This approach significantly extends the scale of interactive molecular visualization and has potential applications in educational visualization and large-scale mesoscale modeling.

Abstract

Rendering huge biological scenes with atomistic detail presents a significant challenge in molecular visualization due to the memory limitations inherent in traditional rendering approaches. In this paper, we propose a novel method for the interactive rendering of massive molecular scenes based on hardware-accelerated ray tracing. Our approach circumvents GPU memory constraints by introducing virtual instantiation of full-detail scene elements. Using instancing significantly reduces memory consumption while preserving the full atomistic detail of scenes comprising trillions of atoms, with interactive rendering performance and completely free user exploration. We utilize coarse meshes as proxy geometries to approximate the overall shape of biological compartments, and access all atomistic detail dynamically during ray tracing. We do this via a novel adaptive technique utilizing a volumetric shell layer of prisms extruded around proxy geometry triangles, and a virtual volume grid for the interior of each compartment. Our algorithm scales to enormous molecular scenes with minimal memory consumption and the potential to accommodate even larger scenes. Our method also supports advanced effects such as clipping planes and animations. We demonstrate the efficiency and scalability of our approach by rendering tens of instances of Red Blood Cell and SARS-CoV-2 models theoretically containing more than 20 trillion atoms.

Figures (12)

• Figure 1: Red Blood Cell (RBC) occluded by SARS-CoV-2 particles, with full atomistic detail. (Left) The proxy geometry comprising the overall scene structure; (Middle) The intermediate acceleration structures for ray tracing; (Right) The resulting model with on-the-fly instancing, rendered at interactive rates via ray tracing at the level of individual atoms.
• Figure 2: The Nanouniverse system constructs the scene out of proxy geometries (top left), which are then filled with Wang tiles (middle left), comprising Wang squares for their volumetric shells (shell space) corresponding to membranes, as well as Wang cubes for their 3D interior (core space). Wang tiles are populated with the atomistic detail of instances of proteins from PDB files (bottom left). Wang tiling recipes are created in a pre-processing step. In addition, a three-level acceleration structure (AS) is built to accelerate rendering using ray tracing. During rendering, the interiors of all Wang tiles are instantiated virtually and rendered with full atomistic detail.
• Figure 3: Geometric Wang tiles are either (left-most) Wang square tiles, or (right-most) Wang cube tiles. Both contain full atomistic detail. Middle part of the figure illustrates the process of mapping Wang square tiles to the corresponding uv coordinates in the Wang tiles texture.
• Figure 4: The shell space (left) is the volumetric layer between a base mesh and an offset mesh. The adaptive shell space (right) is the layer between positive and negative offset meshes. The bottom row shows both types of shell spaces filled with mesostructures with atomistic detail.
• Figure 5: Three levels of acceleration structures (AS) used in Nanouniverse. Micro Level AS ($\mu$LAS) store the main geometry in the scene. Meso Level AS ($m$LAS) contains Wang tiles and cubes. Nano Level AS ($n$LAS) are designed for protein instances data.