Leave No Development Candidate Behind

90% of drugs fail in clinical development. Most of that failure happens because preclinical models are poor predictors of what happens in humans. We are building the models that close that gap.

Explore a Partnership See Our Validated Assets

Want to run our tools yourself? Explore the platform →

100+ scientists in biology, chemistry & physics
$50M + $32M capital raised · ARPA-H contract
2 / 40 partnership slots filled

CD73 protein with overlaid docked hit compounds — Deep Origin internal program

CD73 program · 159 compounds, 30% hit rate

9 in 10 drugs entering clinical trials never get approved

Decades of time, billions of dollars, countless patients waiting because preclinical models still don’t predict clinical behavior. We build ones that do — applied at the five decisions that change a program’s trajectory.

>100,000

compounds screened

~250

advanced through hit-to-lead

~10–20

enter clinical testing

1

approved drug

Source: Sun et al., Acta Pharm Sin B, 2022. Cost breakdown across the development pipeline.

Target ID
Kill weak targets early

Efficacy and tox risk flagged before you commit program budget.
Hit discovery
Synthesize the right molecules

Physics-based filtering before a medicinal chemist touches a flask.
Lead optimization
Pick the candidate that survives

Multi-parameter ranking instead of a gut call.
Program rescue
Redesign around a tox signal

Save the program — keep the sunk investment.
Trial design
Enrich the responder population

Predictive subpopulation selection to lift Phase II PoS.

Explore a partnership

Why Deep Origin

Three things that don’t exist together anywhere else.

Coverage, novel-target predictivity, and independent benchmarks — most platforms own one. We’re built around all three.

01 Scale + explainability

Scale and explainability no one else combines.

Physics-based docking suites are explainable but stop at Molecular. Phenotypic-imaging platforms reach Cellular — as image correlations, not physics. Generative AI platforms span similar scale to ours but as black boxes. PK/PBPK and trial-simulation tools operate only at Tissue and above. Deep Origin combines Quantum-through-Cellular coverage with explainable physics throughout — and ARPA-H extends the roadmap further.

Learn more about our ARPA-H Program

QuantumMolecularMacromol.CellularTissue+

Deep Origin
Physics-based docking suite
PK/PBPK & trial-simulation tools
AI structure prediction platform
Phenotypic-imaging platform
Generative AI platform

Deep Origin’s coverage extends through Tissue and beyond on the ARPA-H roadmap.

02 Novel targets

We work where other models can’t.

DO Dock retains predictivity on targets with 0–20% Tanimoto similarity to the training set — the space where novel programs live, and where the most valuable partnership conversations happen. Verified on the Runs N’ Poses benchmark.

View our Science

Predictive confidence on novel targets (0–20% Tanimoto similarity)

DO Dock 55%
Vina 22%
AF3 18%
Boltz-1 9%
Chai 5%

03 Public benchmarks

Top-2 against foundation models on the ADME tasks that drive early triage.

Deep Origin Togo ranks #1 on permeability, lipophilicity, solubility, half-life, hERG, and AMES, and #2 on plasma protein binding and DILI — ahead of Recursion MolE and Google TxGemma on those tasks. Tradeoffs remain on CYP-mediated inhibition; benchmarks are public via Therapeutic Data Commons.

View our Science

TDC rank · selected tasks

Permeability #1
Lipophilicity #1
Solubility #1
Half life #1
hERG #1
AMES #1
Plasma protein binding #2
DILI #2

Deploy · Partner · License

Three ways to work with us

Pick the one that fits your program.

SaaS · self-serve

Use our platform yourself

The full Deep Origin discovery suite with an AI copilot that brings complex workflows to the whole team.

View our platform →

Research partnership

Let us use it on your program

Multi-year research collaborations backed by physics-based AI, a 100-person team, and a $32M ARPA-H contract.

Explore a partnership →

Out-licensing

See our validated assets

Internally generated programs against targets including CD73 — ready for co-development or license.

View validated assets →

Our vision

The next decade of drug discovery runs on predictive preclinical models.

90% of drugs fail in clinical development. Most of that failure happens because preclinical models are poor predictors of what happens in humans. We’re building the physics-based, AI-accelerated models that close that gap — so fewer programs die in the clinic.

The platform combines molecular physics with AI acceleration. Our team of 100+ scientists across biology, chemistry, and physics work with pharma partners on real programs, not demos.

Bringing this vision to life with

ARIA program

Endometriosis target discovery

An autonomous AI Scientist evaluated 178 protein targets across 10,539 papers — filtered by druggability, mechanism, efficacy, and safety.

Explore the targets ARPA-H · PREDICTS consortium

Replacing animal testing with in-silico safety

Deep Origin leads a $31.7M ARPA-H CATALYST contract to build in-silico and ex-vivo systems that predict drug safety before first-in-human trials.

Read the announcement

Founders

Built by people who’ve done this before — and are in it for the long arc.

Michael Antonov

Co-Founder & CEO

Co-founded Oculus, sold to Meta for more than $2B. Now personally financing Deep Origin through Formic Ventures — a long-term bet on computational biology from someone who has built frontier technology at scale.

LinkedIn →

Garegin Papoian, Ph.D.

Co-Founder & CSO

Monroe Martin Professor and Director of Chemistry & Biophysics at the University of Maryland. Twenty years of peer-reviewed research in molecular physics underpins the models we apply to pharma programs.

LinkedIn →

Learn about Deep Origin