Genotype-aware ranking, 3D tissue physics, and a governed evidence trail — so your team can
see which compounds fit the biology, where delivery breaks, and what to test next.
Drug discovery does not lack data. It lacks evidence a team can act on — before the
wet lab spends a year proving the wrong shortlist.
Compute behind the loop
Proudly accepted into the NVIDIA Inception Program.
LURA is built for accelerated, parallel biology: the compute foundation to move from broad
screening to deeper tissue simulation and governed evidence without waiting through another
wet-lab cycle.
The bottleneck is not intelligence. It is time, false confidence, and the distance between a
promising assay and a working therapy. LURA gives oncology teams a faster filter: genotype-aware
ranking, mechanism, 3D tissue physics, and a governed evidence trail. The result is not another
score — it is a shortlist your team can defend and test.
— Elyris Labs · LURA
Not another scoreboard. A decision system.
01
Pharmacogenomic ranking
Screen
Genotype-conditional selectivity at campaign scale — which compounds differentiate for this genetics, not a generic scoreboard across everyone.
→02
Targeting + spatial physics
Confirm
Therapeutic-target and pathway context on the shortlist, then 3D tissue physics: penetration, hypoxia, stromal shielding — kill pressure, not vibes.
→03
Repurposing + export
Atlas · Receipt
Structural analogs and indication hypotheses for portfolio rescue — then a governed Pilot Package with provenance, confidence, and degradation flags.
→
Workflow — 03
Screen → Shortlist → Confirm → Receipt.
Scroll ⟶
01Explore
Screen
Pharmacogenomic campaign: 2.6M drug–cell pairs in ~1.5s — rank what is selective for this genetics first.
2.6M / 1.5sA100 fleet
02Gate
Shortlist
~40 high-conviction candidates — therapeutic leads and repurposing options that cleared genotype ranking.
~40 survivorsHigh conviction
03Causal
Confirm
Mechanism + targeting context, then 3D penetration and hypoxia — fail the bad physics before wet-lab spend.
Spatial TMEHypoxia resolved
04Export
Receipt
Deliver the cryptographically anchored Preclinical Proof Pack — audit-ready for reviewers and partners.
Proof PackTraceable
05Product
Pilot Package
The artifact BD, translational, and portfolio teams share — from screen noise to a market-facing evidence deck.
Reviewer-gradeAct on it
One motion. Four phases.
Receipt-native
Fig. 023D tissue lattice — where delivery fails
Science — 04
Where the drug dies — and why.
Flat assays hide the failure modes that kill programs later: hypoxic cores, stromal shells,
and penetration collapse. LURA models genotype-conditional selectivity and 3D tissue physics,
then packages the answer as a governed Pilot Package your reviewers can act on —
so the wet lab validates the right shortlist, not the lucky one.
PGx Pharmacogenomic ranking for this genotype — not a population average
Target Drug → target → pathway context when the question is “why this molecule”
Spatial Kill-pressure: potency, hypoxia, stroma, penetration — where physics kills the bet
Atlas Repurposing and analog families for portfolio rescue and indication expansion
0Drug–cell pairs screened on A100 — 1699 cells × 1545 compounds in a single pass
0Selectivity throughput — 2,624,955 pairs scored in 1.5 seconds
0Stable shortlist survivors — genotype-ranked, ready for mechanism and physics
What we measure before we claim
Receipts, not adjectives.
Every number below is bound to an artifact and a split. Research-grade results are labeled
exactly that — and are never promoted into a governed product claim until bound assets back them.
0.92
Joint posterior coverage ≥ 0.90 gate
Pre-registered 20,000-sim / 100-trial SNPE design with a 72-hour multi-dose + oxygen series. Joint parameter coverage passed.
Research calibration · B1-v2 · not promoted
8/8
Oracle panel suite — asset-bound
Engine evidence is fail-closed and bound to assets. No governed claim ships without bound evidence.
Track A · fail-closed
6/6
Tahoe panel suite — live + audited
Mechanism pathway context is live. It is cell-aware and drug-invariant — we do not use Tahoe to rank drugs.
Track A · mechanism context
0.83
Held-out scaffold AUROC
Structural generalization on the Oracle panel — held out by scaffold, not by random split.
Science · generalization
~104×
PDE summary surrogate
A research accelerator for summary field statistics under shadow deferral — p95 relative L2 well under 1%. The full-field PDE remains the authority.
C3 · research-only · shadow-gated
Identity before science · partial truth over fake green · fail-closed promotion.
The slipstream to a better shortlist.
Most portfolios stall between “interesting in vitro” and “ready for real capital.”
LURA is the filter in that gap — rank by genetics, stress with mechanism and physics,
export evidence BD and translational teams can carry forward.
01
Stale inventory
Shelved assets, broad libraries, and half-tested indications with no genotype-aware ranking.
02
Pharmacogenomic screen
Campaign-scale selectivity: what differentiates for this context — including repurposing candidates.
03
Targeting + physics
Shortlist only: pathway/target context and 3D failure modes that flat plates never show.
04
Governed package
Pilot Package ready for internal review, partner decks, and the next go/no-go — not a heatmap dump.
Library → shortlist → evidence → decision. Less theater. More motion.
Oncology first. Same loop, wider tissue.
The beachhead is cancer drug development — where 3D resistance and genotype mismatch burn the most capital.
The architecture is tissue-general: genotype anchor → mechanism map → spatial physics → governed export.
Active research tracks extend the same slipstream.
Now
Oncology
Virtual tumors, TME physics, PGx ranking, repurposing atlases — production path for pilots and CROs.
Next
Brain / CNS
Pressurized, hypoxic lesions and barrier-aware penetration — already in the spatial research lane.
Next
Longevity
Multi-pathway combinatorial screens before multi-year aging studies burn calendar and capital.
Next
Cardiovascular
Complex tissue contexts where targeting without physics still ships the wrong lead.
Elyris Labs.
Builders across engineering, translational science, and clinical systems —
gathered for one purpose: to close the distance between a promising assay
and a therapy a patient can actually receive.
We believe the industry does not need another scoreboard. It needs a decision
system — genotype-aware ranking, 3D tissue physics, and a governed evidence trail —
so capital and wet-lab cycles go to the shortlist that survives, not the one that
merely looks interesting in a flat plate. Oncology is the beachhead. The architecture
is built for the next tissue.
Aanu Oshakuade
B.S. Computer Engineering · AI/ML
Founder & CEO
Elisha Sanoussi
M.Eng. Translational Medicine · UC Berkeley/UCSF
Co-Founder · Translational Medicine
Samuel Tandeka
M.S. Engineering Management · B.S. Electrical Eng.
Co-Founder · Operations & GTM
Colton Kirsten
B.S. Computer Engineering · cloud & multi-tenant infra
Founding Engineer · Infrastructure
Ricardo Silva, PhD
WHO consultant · clinical engineering & regulatory
Principal Scientific Advisor
What serious teams actually need.
001
Pharmacogenomic ranking — selective for this genotype
PGx→
002
Therapeutic targeting + pathway context without fake ranks
Target→
003
3D physics — hypoxia, stroma, penetration failure modes
Spatial→
004
Repurposing atlas + governed Pilot Package for portfolio motion
Export→
Conviction
The next decade of medicine will be computed before it is tested.
The wet lab confirms the answer. It no longer searches for one.
Contact — 10
Move the portfolio upstream.
Bring a compound set, a biological context, or a stalled program. We’ll show how LURA turns it
into a ranked, explainable next step before more wet-lab spend.
