They predict structure.
We simulate dynamics.
AlphaFold 2 / 3
Static structure prediction
Predicts one frozen conformation from sequence.
Requires GPU cluster (TPU v4 for AF3).
Cannot screen compounds.
Cannot simulate protein motion.
Minutes to hours per protein.
Requires GPU cluster (TPU v4 for AF3).
Cannot screen compounds.
Cannot simulate protein motion.
Minutes to hours per protein.
vs
22Rx / LoNC
Full molecular dynamics
Simulates the actual physics — folding, binding, motion.
Runs on commodity CPU. No GPU.
Screens 13.2M compounds/sec against targets.
370,000 atoms in 30 seconds.
Enables high-throughput compound screening against simulated targets.
Runs on commodity CPU. No GPU.
Screens 13.2M compounds/sec against targets.
370,000 atoms in 30 seconds.
Enables high-throughput compound screening against simulated targets.
| Capability | AlphaFold | Traditional MD | 22Rx / LoNC |
|---|---|---|---|
| Structure prediction | Yes | Yes (slow) | Yes |
| Dynamics simulation | No | Yes | Yes |
| Drug screening | No | ~1K compounds/day | 13.2M/sec |
| Hardware required | GPU / TPU cluster | Supercomputer | Commodity CPU |
| Time (large protein) | Minutes–hours | Weeks–months | 30 sec |
| Retrospective FDA drug-class benchmarks | Not applicable | Not demonstrated | 15+ known drug classes screened in retrospective demos |
Molecular dynamics at a scale
that wasn't possible before.
370K
Atoms simulated in 30 seconds
13.2M
Compounds screened per second
10B
Compounds screened in 8–12 minutes
Traditional molecular dynamics takes weeks to months on supercomputers for a single protein. We simulate 370,000 atoms in 30 seconds on a commodity CPU — then screen billions of drug candidates against the result. AlphaFold gives you a static structure. We simulate dynamics and rank compound libraries computationally.
Retrospective Benchmarks15+ FDA-approved drug classes used as
retrospective screening speed benchmarks.
Physics-based screening at 13.2M compounds/sec. Retrospective comparisons to known approved drugs demonstrate throughput. Scoring accuracy requires domain-expert calibration.
Paxlovid-class COVID-19 benchmark
Remdesivir-class COVID-19 benchmark
Carbamazepine-class Bipolar benchmark
Donepezil-class Alzheimer's benchmark
Memantine-class Alzheimer's benchmark
Sotorasib-class KRAS G12C benchmark
JAK inhibitor class Autoimmune benchmark
Lenacapavir-class HIV Capsid benchmark
178,000 candidates banked (experimental validation required).
Every major pandemic family.
Coronaviridae COVID, SARS, MERS
Flaviviridae Zika, Dengue
Orthopoxviridae Smallpox
Filoviridae Ebola, Marburg
Orthomyxoviridae H5N1 Pandemic Flu
Bacillus anthracis Anthrax
Yersinia pestis Plague