Steam Frame hit Steam's top 5 in a month. The interesting part isn't the headset.

Hey —

Quick note before the main piece. Valve's Steam Frame has been out for a little over a month, is still heavily backordered, and according to Road to VR's read of the latest Steam Hardware Survey, it's already the fifth most-used headset on Steam. It's the first survey to include it at all.

I want to be precise about what that does and doesn't mean. It's a usage survey, not a sales chart. It doesn't tell us the install base. What it does tell us is that people who own the thing are spending time in it — enough time, across enough owners, to register in a ranking dominated by established hardware.

That's the signal. Sustained use is the precondition for anything ambitious in spatial computing. A demo economy produces demo software.

Why I keep writing about proteins and depth

Here's the thing I can't stop noticing: structural biology is a three-dimensional discipline that mostly gets taught and practiced in two dimensions.

A binding pocket is a cavity. A helix is a spiral. An allosteric transition is a motion through space. Flatten all three onto a monitor and you get a set of conventions — good conventions, useful conventions, but conventions. Students learn to read them instead of seeing the object. Researchers summarize the geometry with numbers and then go back to inspecting a still image.

Numbers are summaries. They are not inspections. When you want to know why two structures disagree, the answer is usually geometric, and geometry is exactly what a screen is worst at.

The tooling, concretely

This is the part where I tell you about the thing I work on, so: NexoPeek is a VR molecular laboratory for Meta Quest, made by Seven Sportz. Two halves.

The guided half is 12 narrated tours across six pillars — Biotechnology Basics, Foundations of Life, Pathways & Interactions, Medicines in Action, Genetic Technologies, Green Biotech. You step inside hemoglobin and watch oxygen bind. You follow glucose through glycolysis. You watch aspirin block COX enzymes. You stand inside a CRISPR-Cas9 edit while someone explains what you're looking at.

The professional half is a PDB viewer, and this is where it gets interesting as a systems problem. A gene name resolves across RCSB PDB, UniProtKB, AlphaFold DB, SWISS-MODEL, and PDBe, and the structure loads at human scale. One tap cycles Mol* presets: cartoon, ball-and-stick, spacefill CPK, molecular surface, B-factor heat, tube trace. Your own .pdb, .cif, and .mol2 files load with electrostatic field-line maps and electron-density volumes.

Then the analysis bench. You grab and rotate the molecule with bare hands. AI pocket prediction — PrankWeb / P2Rank cloud or the offline engine — lights up a candidate druggable site. Foldseek superimposes a homolog. The Ligand Inspector traces hydrogen bonds and pulls a 3D ligand out of PubChem. Trigger-pick two atoms, get the distance in angstroms. Tap a residue pill in the sequence, watch a 3D halo land on the structure.

Engineering notes from the bench

A few things I'd want to know if I were reading this on HN rather than writing it.

The input model is the hard part. Desktop viewers assume a pointer with sub-pixel precision and a keyboard for search. In a headset you have two tracked controllers, a gaze vector, and no reliable text entry. We leaned on that constraint rather than fighting it: residue selection became a tap on a pill, distance measurement became a two-handed trigger gesture, camera framing became a physical step rather than a scroll. The bet is that spatial gestures map onto spatial questions better than menus do — not that gestures are universally faster.

Rendering load is the second constraint. Molecular surfaces and B-factor heat maps are cheap on a discrete GPU and expensive on a standalone headset. Mol* presets let you trade fidelity for frame rate deliberately, and the electron-density volumes are the heaviest thing in the app — expect to reach for surface representation when you're inspecting a volume, and cartoon when you're inspecting a fold.

The third is where compute actually runs. Pocket prediction has a cloud path and an offline path. Cloud gets you the current PrankWeb / P2Rank models; offline gets you a result with no network dependency and no upload. Those are different products for different people, and the honest answer is that neither one replaces a validated pipeline.

The honest caveat

None of this replaces your validated pipeline. Pocket prediction and alignment are aids — they help you build the spatial intuition that decides which hypothesis is worth running at all. If you're expecting a citable result straight out of the headset, you'll be disappointed. If you're expecting to understand a binding site faster than you would on a screen, that's the bet.

Where it fits

Room View floats the molecule into your actual room while the wall menu stays operable, which turns out to matter in classrooms and shared lab spaces — you can gesture at a structure without a monitor between you and the person you're explaining it to. A student exploring ATP, an educator teaching CRISPR, a structural biologist checking a pocket — same headset, same afternoon.

If the Steam Frame numbers tell us anything, it's that the "is VR ready?" question is closed. The open question is which scientific tools are worth building on top of it — and what the interaction model for those tools should actually be.

NexoPeek is on the Meta Quest store: https://www.meta.com/experiences/nexopeek/1453797551140347/

— Educational and research-visualization software. Not medical advice, diagnosis, or treatment. Analysis aids support, never replace, validated scientific pipelines and professional judgment.

If you try it, reply and tell me what broke. That's the feedback I actually want.