One learning loop for molecular design.
Rayca links what your team already knows about a molecule to the models, simulations and compute needed to decide what to make next.
A public structure, shown to illustrate a cryptic pocket. The pocket only opens as the protein moves.
Most programs already hold the reason their next molecule will fail.
It sits in an assay report, an old SAR table or a developability review that never reached the chemist designing the next compound. Rayca connects those records to design.
- Evidence is spread out
- SAR, structures, assays, omics and CMC data live in different systems and rarely meet before a design decision.
- Generating molecules is the easy part
- Programs stall on potency, selectivity, exposure or manufacturability problems that earlier data could have flagged.
- Compute is still run by hand
- Simulations and cluster jobs are set up, queued and tracked manually, so many useful questions never get asked.
- Not linked
SAR tables
Medicinal chemistry
- Not linked
Crystal structures
Structural biology
- Partly linked
ADME and safety assays
DMPK
- Not linked
Omics and target biology
Biology
- Not linked
Formulation notes
CMC
- Not recorded
Why past compounds failed
Program reviews
Every result changes the next round.
The six stages run as one system. When lab data comes back, the constraints, the designs and the simulations that follow are updated with it.
Evidence
Structures, assays, SAR and the record of past failures are brought together.
Intelligence
The system works out which constraints matter for this target and this series.
Design
Candidates are proposed across small molecules, peptides, antibodies and degraders.
Simulation
Each design is scored against the moving protein with physics-based methods.
Experiment
Your lab tests the shortlist.
New evidence
Results come back in and shape what the next round looks for.
Every failed molecule leaves a reason behind. Rayca keeps it.
When a compound misses, the cause is usually recorded somewhere: a liability in an assay, a clash in a structure, a solubility problem in formulation.
Rayca links each cause to the molecular features behind it, then checks new designs against it before anything is made.
Candidate
Lead series, compound C
Reasons, linked to their source
- hERG signal in the safety panelAssay
- Low solubility at physiological pHFormulation
- Strained conformation in the bound poseStructure
Constraint added to the failure graph
Avoid a basic amine at the solvent-facing position in this series.
Next round, checked before synthesis
Design A
ClearDesign B
FlaggedDesign C
Clear
Rayca also runs the computation.
Agents set up simulations, send jobs to GPU and HPC clusters, watch them and bring the results back into design. You can open any result and see what produced it.
BRD4 and VHL, held by a degrader
PDB 5T35
- Running
Conformational ensemble
Molecular dynamics · GPU
- Done
Linker variants
Structure-aware design · GPU
- Queued
Complex stability
Free energy · HPC
Lineage of the finished run
- Input structure
- Model version
- Settings
- Result
- Models and simulation together
- Rayca picks and sequences the models and physics methods each question needs.
- GPU and HPC execution
- Jobs are planned, submitted and monitored across distributed clusters.
- Governance
- Every run is scoped, logged and reviewable, which suits regulated work.
- Lineage
- Each result records the inputs, models and tool calls behind it.
How the system is put together.
Seven layers, from the evidence at the bottom to the decision your scientists see at the top.
- 1
Molecular knowledge
Structures, SAR, assays, omics and development constraints, kept as a persistent record.
- 2
The failure graph
Why molecules failed, stored as constraints that new designs are checked against.
- 3
Specialized AI models
Design and scoring models, chosen for the question being asked.
- 4
Physics-based simulation
Free energy and molecular dynamics methods that assess binding before synthesis.
- 5
Compute orchestration
Agents that plan, submit and monitor jobs on GPU and HPC clusters.
- 6
Governance and lineage
A record linking every result to its inputs, models and settings.
- 7
Decision view
What the evidence implies for the next design, in front of your scientists.
Your scientists make the calls.
Rayca does the computational work and shows its reasoning. Go and no-go decisions stay with your team.
It plugs into the data, models and clusters you already use.
- Design AReasoning and lineage attached
- Design CReasoning and lineage attached
- Design DReasoning and lineage attached
Questions
Anything else? .
What is the Rayca platform?
Rayca is a molecular intelligence engine for drug discovery. It connects molecular knowledge, specialized AI models, physics-based simulation and GPU and HPC execution in one learning system for molecule design. Your team sets the direction and makes every decision.
Is Rayca an agentic AI drug discovery platform?
Agents are part of how it works. They plan, sequence and run the computation behind each molecular question. The product is the loop around them, which connects evidence, design, simulation and experiment, with people making every call.
How is this different from a single AI model or a molecule generator?
A generator proposes molecules. Rayca also reasons over prior evidence, scores designs against binding physics, runs the computation and feeds each result back into the next round.
What does a learning loop mean here?
Evidence informs design, designs are tested in simulation, the best go to the lab, and lab results come back as new evidence for the next round. It is a closed loop between computation and the bench.
What is the failure graph?
It is a structured record of why molecules failed. It links molecular features, assay results, liabilities and outcomes, so recurring failure modes can be checked for in new designs.
How does Rayca use physics-based simulation?
Designs are scored against how the target actually moves. Free energy methods and molecular simulation assess binding before anything is made, so lab time goes to the strongest candidates.
How does Rayca handle GPU and HPC compute?
Agents plan and submit the jobs, run models and simulations across distributed GPU and HPC resources, and bring results back into design. Governance and data lineage cover every run.
Which modalities and targets does it support?
Rayca is disease-agnostic and designs small molecules, peptides, antibodies, nanobodies and degraders. It is built for targets that resist conventional approaches, including those with cryptic pockets.
Is the system autonomous? Who decides?
No. Rayca runs the computation and shows what the evidence implies. People make every go and no-go decision, and every result can be traced to the inputs and models behind it.