Rayca Precision Molecular intelligence for therapeutic discovery
Rayca helps pharma and biotech teams decide which molecules to make next, using everything their programs have already learned.
The attrition problem 01
Commonly cited estimates put the cost of an approved medicine near 2.6 billion dollars, much of it spent on candidates that did not make it. We believe carrying lessons forward is one of the most direct ways to change that.
Fragmented program evidence
Typical oncology programProject team query
Has a scaffold like RC-0480 failed before, and on which liability?
Assay results
Plate exports and LIMS
18,240 values
Chemistry
Registry and notebooks
3,120 compounds
Structures
Crystal and modelled
46 files
ADME and safety
Contract lab reports
212 reports
Decisions
Slides and meeting notes
390 decks
Expertise
In people’s heads
14 scientists
15 years
Typical path from target to approval
$2.6B
Average cost of one approved medicine
90%
Of clinical candidates not approved
Commonly cited industry estimates. Each cell is one clinical candidate.
The intelligence layer 02
It runs within your institution, so what the program learns stays yours and each new cycle can start from it.
It runs within your institution, so what the program learns stays yours and each new cycle can start from it.
The molecular intelligence stack
PARP1 program · cycle 4Program record
Every measurement, normalised, with provenance
4,612 records
- Assays
- Chemistry
- Structures
- ADME
- Safety
- Notes
Specialized models
Potency, selectivity and ADMET models, trained on this program and retrained every cycle
6 models
Potency
pIC50 8.9
Selectivity
> 400×
Solubility
85 µM
Permeability
ER 1.8
hERG
> 30 µM
Clearance
9 mL/min/kg
Physics-based simulation
Docking, dynamics and free energy on real structures
3 methods
- Docking
- Molecular dynamics
- Free-energy perturbation
Autonomous compute governance
Simulation on HPC and model inference on GPU, planned, allocated and supervised by the system
96 jobs governed
Simulation on HPCModel inference on GPU2 failed jobs recovered automatically
Our approach 03
Trade-offs between potency, exposure and safety become visible early, while they are still inexpensive to change.
One molecule, every discipline
RC-0512- Medicinal chemistrySAR, synthetic tractability
- Structural biologyBinding mode, pocket states
- BiochemistryPARP1 and PARP2 potency
- DMPKSolubility, permeability, clearance
- SafetyhERG, blood counts
- TranslationalExposure at projected dose
3 vectors open
H-bond to Gly863
PARP1 1.2 nM
85 µM soluble
hERG > 30 µM
50 mg once daily
Evidence integration 01
Every SAR table, assay and structure, resolved into one program record.
Sources this program owns
- Compound registry3,120 compounds
- PARP1 and PARP2 assays9,804 values
- Crystal structures12 files
- ADME panel418 results
- Rat safety study1 report
- Meeting notes64 documents
Attrition analysis 02
Each discontinued compound, traced to its molecular cause.
| PARP1 IC50 | PARP2 IC50 | Kinetic solubility | Caco-2 efflux | hERG | Rat blood counts | |
|---|---|---|---|---|---|---|
| RC-0412 | ||||||
| RC-0433 | ||||||
| RC-0451 | ||||||
| RC-0467 | ||||||
| RC-0480 | ||||||
| RC-0486 | ||||||
| RC-0490 | ||||||
| RC-0494 | ||||||
| RC-0497 | ||||||
| RC-0503 |
Each cell is one measured outcome. Red cells failed their target product profile threshold.
Failure mode
PARP2 activity
Tracks with the platelet and neutrophil drop in the 14-day rat study.
Design constraint, now enforced
PARP2 / PARP1 IC50 ratio of 100 or more
Seen in 5 compounds, backed by 22 measurements
Constrained generation 03
New chemical matter, generated inside the program’s constraints.
Olaparib, the reference chemotype
0 proposed, 0 pass every rule
Each cell is one generated molecule. Ink passed every constraint, red broke a learned constraint, grey scored below the potency model.
RC-0512
Cyanopyridine replaces the amide cap
- PARP2 window–
- cLogP below 3–
- No basic amine–
RC-0515
A nitrogen added to the core ring
- PARP2 window–
- cLogP below 3–
- No basic amine–
RC-0518
A spiro ring replaces the piperazine
- PARP2 window–
- cLogP below 3–
- No basic amine–
RC-0521
A basic amine on the piperazine
- PARP2 window–
- cLogP below 3–
- No basic amine–
Free-energy triage 04
Selectivity estimated with physics before compounds are made.
- Ensemble docking9 compounds · CPU
- Molecular dynamics, 100 ns6 compounds · GPU
- RBFE, PARP13 compounds · GPU
- RBFE, PARP23 compounds · GPU
GPU fleet. Blue cells run this campaign.
Predicted PARP2 / PARP1 window
- RC-0512–
- RC-0518–
- RC-0515–
Line marks the 100-fold constraint.
Candidate nomination 05
A decision package for the project team to review.
Cycle 5 synthesis plan
- 01SynthesiseRC-0512 and RC-0518Pass every rule and are predicted to favour PARP1 by 140 to 180 times.
- 02Run firstPARP2 counter-screen and 7-day rat CBCConfirms the window in vitro and in vivo before scale-up.
- 03DeprioritiseRC-0515 and RC-0521One is not selective enough. One breaks the efflux rule.
Evidence behind this plan
4,612
records
3
constraints
9
simulations
1
partner assay
RC-0512 against the target product profile
- PARP1 IC50, predicted–
- PARP2 / PARP1 window–
- cLogP–
- Basic amine–
- Caco-2 efflux ratio, predicted–
Program record, by cycle
C1
C2
C3
C4
C5
Lab results return to the program record and cycle 5 starts from there.
Example program. Values are illustrative.
Lab in the loop 05
Less time between a question and the data that answers it.
Experiment through a partner lab
PARP2 cellular counter-screen- Open question
- Catalogue match
- Your approval
- Partner run
- Back in the record
- Question
- Does the PARP2 window hold in cells?
- Matched assay
- Cellular PARP2 counter-screen, partner catalogue
- Approval
- Approved by the project lead
- PARP2 IC50, RC-0512
- > 3 µM
- Program record
- Selectivity model updated for cycle 5
Darker wells show stronger PARP2 inhibition. Red marks a compound losing its window. Illustrative.
Where we differ 06
We want fewer compounds made that the program’s own data could have ruled out, and recommendations a team can check.
We want fewer compounds made that the program’s own data could have ruled out, and recommendations a team can check.
Attrition becomes institutional knowledge
Every discontinued compound leaves a constraint behind, linked to its evidence and applied to every future design.
- PARP2 window
- cLogP below 3
- No basic amine
Compute is governed autonomously
An autonomous system plans, allocates and supervises molecular simulation on HPC and AI model inference on GPU, within the budgets and policies your team sets.
RBFE campaign · 0 of 48 GPUs allocated
Every recommendation is auditable
Each call traces back to the runs, files and rows behind it, so a project team can check and challenge it.
Scientists keep the decision rights
Rayca works alongside the project team. Your scientists own every go and no-go and every wet-lab result.
Partnering 07
A contained way to see how Rayca works on your own science before deciding on anything larger.
A contained way to see how Rayca works on your own science before deciding on anything larger.
A first engagement
One programFrame the question
One program, one decision that matters
Deploy in your environment
Your data stays with you
Run targeted computation
Only what the question needs
Return prioritised actions
Designs and experiments, with lineage
Program intelligence review
Hidden constraints, recurring liabilities and open decisions across a running program.
Molecular second opinion
An independent computational assessment before the next synthesis or IND-enabling spend.
Attrition analysis
Why earlier compounds failed, rebuilt as design constraints your chemists can apply.
Design cycle
New chemical matter generated, simulated and ranked, with every run recorded.
What a first engagement returns
Integrated program record
Built and kept within your environment.
Failure-mode map
Recurring liabilities, each with the evidence behind it.
Prioritised designs and experiments
What to make and test next, with the reasoning shown.
Lineage for every call
Runs, versions and files behind each recommendation.
Co-developing novel therapeutics with leading research centres and biotechnology companies.
With our academic and industry partners we are co-developing novel therapeutics and the platforms behind them. Rayca is part of the DayOne Accelerator in Basel and Health Incubator Helsinki.
Therapeutic co-development partners
Accelerator and ecosystem partners
Vision 08
We see this as a new foundation for pharma R&D: medicines designed from everything already known, so more of them reach the patients who need them.
We see this as a new foundation for pharma R&D: medicines designed from everything already known, so more of them reach the patients who need them.
Levels of molecular intelligence
Our roadmap- L1IntegratedOne record of everything a program has measured, with provenance.Today
- L2LearningEach failure becomes a constraint the next design has to respect.Today
- L3DesigningNew molecules generated and tested with physics before synthesis.Today
- L4ExperimentingProposes the next experiment, runs it through partner labs with your approval, and learns from the result.In development
- L5Molecular superintelligenceReasons from a biological goal to a development candidate, across targets and modalities.Our direction