← Back

Computational Drug Discovery

13.2M/sec

Compound screening velocity. 10 billion candidates in 8–12 minutes.
Physics-based. No training data. No GPU.

The Problem

Traditional drug discovery:
15 years, $2.6 billion, 90% failure.

The pharmaceutical industry screens compounds through high-throughput physical assays — expensive, slow, limited to what's in the library. Virtual screening with docking tools like AutoDock or Glide processes a few thousand compounds per hour. AI approaches require millions of training samples and still can't explain why a drug works.

Method Throughput Physics-based Training Data
Physical HTS ~100K/day Yes (real) N/A
AutoDock Vina ~500/hr Yes (docking) None
Schrödinger Glide ~5K/hr Yes (docking) None
AI / ML (various) ~1M/hr No Millions of samples
22Rx / LoNC 13.2M/sec Yes (first principles) Zero
Performance

The speed changes
what questions you can ask.

13.2M
Compounds screened per second
10B
Full ZINC library in 8–12 min
0
Training samples required

At 13.2 million compounds per second, you don't pick a subset to screen. You screen everything. Every known compound. Every synthesizable molecule. Every conformer. The bottleneck moves from "can we test it" to "can we synthesize it."

Retrospective Benchmarks

15+ known FDA drug classes
screened as retrospective throughput benchmarks.

In retrospective benchmarks, the engine screens compound libraries against known target structures. Results are compared to approved drug classes to demonstrate throughput — not independent discovery.

COVID-19 — Protease Inhibitor
Paxlovid-class compound
Ranked a nirmatrelvir-like compound targeting SARS-CoV-2 protease as a retrospective benchmark. In silico affinity estimates require experimental confirmation.
COVID-19 — Polymerase Inhibitor
Remdesivir-class nucleoside analog
Retrospective benchmark: RdRp-targeting adenosine analog with prodrug scaffold surfaced in screening against the clinical compound class.
Alzheimer's — Cholinesterase
Donepezil
World's #1 prescribed Alzheimer's drug. Retrospective benchmark via first-principles binding to acetylcholinesterase.
Alzheimer's — NMDA
Memantine
World's #2 Alzheimer's drug. Non-competitive NMDA receptor antagonist surfaced in screening via channel dynamics.
Oncology — KRAS
Sotorasib-class covalent inhibitor
KRAS G12C — previously "undruggable." Retrospective benchmark: covalent binding mechanism ranked at Switch II pocket.
Psychiatry
Carbamazepine
FDA-approved since 1974 for bipolar disorder. Retrospective benchmark: sodium channel modulation surfaced in screening from voltage-gated ion dynamics.
HIV — Capsid
Lenacapavir-class
Novel mechanism of action — capsid inhibition. Retrospective benchmark against Gilead's once-yearly injectable via allosteric dynamics.
Biodefense

178,000 candidates banked
across every major threat. Experimental validation required.

Full compound libraries pre-screened against every major pandemic and bioweapon target family. When the next outbreak hits, we don't start from zero — we pull from a pre-computed arsenal.

Coronaviridae COVID / SARS / MERS
Flaviviridae Zika / Dengue / Hepatitis C
Orthopoxviridae Smallpox / Monkeypox
Filoviridae Ebola / Marburg
Orthomyxoviridae H5N1 / Pandemic Flu
Paramyxoviridae Nipah / Hendra
B. anthracis Anthrax
Y. pestis Plague
F. tularensis Tularemia
How It Works

LoNC: physics as compute.
Not statistics. Not AI.

Laws of Nature Compute models real molecular physics from first principles — using proprietary navigation to explore the energy landscape. It doesn't learn from data. It derives answers from the laws of nature. That's why it enables exhaustive library screening; novel hit quality requires scoring calibration and experimental validation.

Running on 44s.io's lock-free cloud infrastructure, the system can screen the entirety of known chemical space against any target in minutes. Not hours. Not days. Minutes.