01

Virtual Cell

Biological intelligence,
rendered at atomic fidelity.

PROTEOMICS GENOMICS SIMULATION
SCROLL
12 M
Atoms per simulation All-atom MD
4.2 PF
Peak compute FP64 petaFLOPS
0.91 TM
Mean TM-score vs. PDB experimental
8 s
Structure prediction ML-assisted inference
02 — PLATFORM

Three pillars of
biological intelligence

01

Proteomics

Full-atom protein structure prediction and interaction mapping across 200M+ annotated sequences.

Folding Docking Binding Sites
02

Genomics

Sequence analysis, CRISPR target prediction, and epigenetic mapping at single-base resolution.

CRISPR Epigenetics Variant Calling
03

Simulation

Molecular dynamics, Monte Carlo sampling, and quantum chemistry computations in real time.

MD QM/MM Force Fields
03 — ARCHITECTURE

How the engine
renders reality

Each simulation passes through four precision-engineered stages, from raw molecular input to interactive real-time output.

01

Molecular Input

PDB structures, FASTA sequences, or CIF files parsed and validated before any computation begins.

PDB · FASTA · CIF
02

Force Field Assembly

AMBER/CHARMM parameters applied. Charge distributions calculated at quantum precision.

AMBER99 · CHARMM36
03

Parallel Integration

4.2 PF of FP64 compute across distributed tensor cores with adaptive time-stepping.

GPU · TPU · Distributed
04

Fidelity Render

Real-time WebGL visualization of the running trajectory. Export completed simulations to CIF, XYZ, or PDB.

WebGL · XYZ · Export
04 — ATLAS

Explore the
cell atlas

Interact with 847 catalogued cell types
across 32 tissue families.

Neuron 86B instances
Epithelial 600B instances
Stem Cell Variable
Erythrocyte 25T instances
Macrophage 200B instances
Myocyte 3T instances
05 — LIVE DATA

Simulation
in motion

Every computation rendered as it runs. Full trajectory access with sub-picosecond resolution and real-time energy tracking.

Time step 2 fs
Output frequency every 500 steps
Trajectory format DCD / XTC / TRR
Energy precision ±0.1 kJ/mol
vcell — simulation output
$ vcell init --cell neuron --organism h.sapiens
Loading PDB:4ZQK (MAP2 microtubule-associated protein 2)
Force field AMBER99SB-ILDN
Atoms 124,832 heavy  /  389,210 total (incl. H)
Solvent TIP3P water box  12Å padding  
$ vcell run --steps 10000000 --dt 2fs --temp 310K
[     0 / 10000000 ]  E = −1,284,332.4 kJ/mol  T = 309.8 K
[ 142000 / 10000000 ]  E = −1,284,118.9 kJ/mol  T = 310.2 K
[ 318500 / 10000000 ]  E = −1,284,007.1 kJ/mol  T = 310.1 K  
06 — ACCESS

Begin your Simulation

Virtual Cell is currently in early access. Apply now to join research institutions, biopharma teams, and independent investigators already on the platform.

47 research teams on waitlist 3 major pharma partners 12 peer-reviewed validations
07 — BLUEPRINT

How something
like this gets built

None of the core engines need to be invented. The hard part is assembling them, scaling the infrastructure, and having the scientific expertise to know when the output is meaningful.

01
Visualization Three.js · WebGL · GLSL

The 3D frontend — rendering the cell, shaders, scroll and mouse interaction. This is what you're looking at right now.

Already built
02
Structure Prediction AlphaFold2 · ESMFold · RoseTTAFold

ML models that predict a protein's 3D shape from its amino acid sequence alone. Open source and deployable on cloud GPUs.

Open source
03
Simulation Engine GROMACS · OpenMM · NAMD · AMBER

Physics engines that simulate how atoms move and interact over time. Decades of academic research, freely available.

Open source
04
Biological Databases PDB · UniProt · Ensembl · ChEMBL

200M+ protein sequences, 200K+ experimentally solved structures, and annotated genomes. All public, regularly updated.

Free & public
05
Compute Infrastructure NVIDIA A100 · AWS HPC · Google Cloud

GPU clusters that do the heavy lifting. The primary cost driver — a meaningful cluster runs $2–5M per year at cloud rates.

Cloud available
06
Scientific Expertise Biophysicists · Computational Chemists

The hardest dependency. Domain experts who know whether the simulation results are physically meaningful — or just plausible-looking noise.

The real barrier

Real-world equivalents: Schrödinger (founded 1990), D.E. Shaw Research (2002), Relay Therapeutics (2016). All took 5–10 years to reach product maturity.