Building a Molecular Laboratory for Meta Quest: Notes on Spatial Structure Visualization

The news that prompted this post

EXOSHOCK, the Halo-inspired co-op shooter for up to four players, just launched in early access on Quest and PC VR (Road to VR). Developer POLARITYONE was reportedly a two-person team targeting a Fall 2025 window. Great ship for a small studio, and a useful reminder of how far the standalone VR stack has come in a few years.

It also got me thinking about the non-gaming half of the Quest ecosystem, which is where I spend my working hours. I build NexoPeek, a VR molecular laboratory. Here are the engineering and design problems that turned out to matter most.

Problem 1: Scale is the whole point, and scale is hard

A protein is roughly 5–100 nm across. Humans reason about objects in the 0.01–2 m range. Everything about spatial cognition — parallax, occlusion, proprioception — is tuned to that band.

The naive approach is to scale the molecule up until it fills the room and call it done. That fails fast: at that scale, a single alpha helix is a wall you can't see past, and you lose the global fold. So NexoPeek loads structures at human scale — you stand roughly at the size of a small domain, with the full chain visible around you — and lets you grab and rescale in place. Bare hands or controllers.

The interesting consequence: occlusion becomes a navigation tool instead of a problem. Buried residues are simply behind you. You don't toggle a surface representation off; you walk around it.

Problem 2: Representation switching has to be instantaneous

Structural biologists don't think in one representation. Cartoon for fold, ball-and-stick for chemistry, spacefill CPK for packing, surface for interface geometry, B-factor heat for flexibility, tube trace for topology.

On desktop you have hotkeys and a command line. In VR you have a controller and about 300 ms of patience. We wired Mol* presets to one-tap cycling. The engineering cost was less about rendering and more about not blowing the frame budget when swapping between a 200k-atom surface mesh and a cartoon trace in a single frame. Culling and progressive LOD do most of the work.

Problem 3: Identity resolution is the actual friction

The most underrated bottleneck in structural work isn't visualization — it's getting from "I care about this gene" to "here is the structure." Users arrive with a gene name, not a PDB ID.

NexoPeek's smart-search bar resolves across RCSB PDB, UniProtKB, AlphaFold DB, SWISS-MODEL, and PDBe from one input. That's a fan-out-and-reconcile problem: heterogeneous schemas, differing identifier namespaces, inconsistent coverage, and the occasional expired cross-reference. The design goal was a single ranked result list rather than five tabs.

Problem 4: Making analysis usable in a headset

The analysis bench is where the project stopped being a demo. Features:

AI pocket prediction — PrankWeb / P2Rank cloud AI, with an offline engine fallback, highlighting a druggable site with a glow volume.
Foldseek superposition — structural homolog alignment rendered in place.
Ligand Inspector — hydrogen bond tracing plus 3D ligand retrieval from PubChem.
Distance measurement — trigger-pick two atoms, read the angstrom value.
Sequence-linked selection — tap a residue "pill" in the 2D sequence strip, get a 3D halo on the structure.

Each of these exists in mature desktop tools. The VR question is always: does the spatial affordance beat the keyboard? For pocket inspection and interface reasoning, our experience says yes. For anything quantitative at scale, no — and we don't pretend otherwise.

Problem 5: Local file loading and volume data

.pdb, .cif, and .mol2 load from local storage. Electrostatic field-line maps and electron-density volumes render as separate layers. Density volumes are the heaviest thing in the app; isosurface extraction on a standalone headset requires aggressive downsampling and a mesh cache, and it's the feature most likely to need iteration after launch.

Problem 6: Room View and context switching

The Room View toggle floats the molecule into the user's physical space while keeping the wall menu operable. This matters more than it sounds. Research and teaching sessions are hybrid — you're in the headset but you're also at a desk with a laptop, a paper, a colleague. Full immersion is not always the goal.

The educational layer

Twelve narrated tours across six pillars: Biotechnology Basics, Foundations of Life, Pathways & Interactions, Medicines in Action, Genetic Technologies, Green Biotech. These are scripted spatial sequences — stepping inside hemoglobin to watch oxygen bind, following glucose through glycolysis, watching aspirin block a COX enzyme, watching CRISPR-Cas9 edit DNA. Narration timing and camera choreography turned out to be a content problem disguised as an engineering one.

What this is not

NexoPeek is educational and research-visualization software. It is not medical advice, diagnosis, or treatment. Pocket prediction and structural alignment are aids for orientation — they support, never replace, validated scientific pipelines and professional judgment.