Platform
The operating intelligence for the molecular world. Across discovery, development, formulation and delivery.
Molecular evidence, AI models, physical simulation and experiments share a continuous learning system. Program outcomes guide the next design and experiment across modalities.
Link program sources through compound, assay and decision identifiers
Whatever a program already holds, registries, assays, structures and minutes, resolves into one set of linked entities used by later design, simulation and decision steps. The same for any target or modality.
Sources this program owns
Resolved and cross-linked in one pass, so a design agent can trace any compound to the assay, structure or decision that explains it.
Rayca selects and runs methods across design, structure, properties and simulation. Every execution keeps its version, inputs, compute environment and outputs.
800+
scientific tools,
one call away
280
of 576 GPUs busy,
scheduled by agents
A designed binder on its target, PDB 3OGO
Protein design
Generate new protein backbones and sequences, then check they fold and bind as designed.
Method families in the catalog
- Diffusion backbone generationin use
- Inverse folding
- Protein language models
- Motif scaffolding
- Binder hallucination
- Interface design
- Fold validation
- Stability scoring
Why it matters
The molecular question determines the method, depth and execution path.
Evidence
Interpret program evidence in molecular context.
Rayca connects experimental results, molecular structures, design history and development evidence. Measurements retain the biological and experimental context needed for interpretation.
Compound dossier
RC-0440 · leadThe lead of the switch-II series. Nine sources resolve to one record, in common units, with values traceable to their source rows.
Provenance KRAS G12C nucleotide exchange
Why it matters
Scientists and models work from the same evidence, context and provenance.
Design principles
Learn from why therapeutic candidates fail. Carry supported lessons into future design.
Rayca connects failed candidates to the evidence behind them, tests whether the pattern is supported and preserves the result as a design principle for future programs.
Series memory
KRAS G12C · cycles 1 to 4Compounds, assays, liabilities, mechanisms and design rules, linked. Read here on a representative series: ask it what a chemist would ask.
AnswerA basic amine binds the hERG pore, and rule C-07 excludes it from subsequent designs.
- Attrition6 compoundshERG IC50 1.4 to 4.2 µM
- Evidence2.1 to 38 µMamine to amide, pIC50 8.3 to 8.1
- MechanismTyr652 · Phe656aromatic cage, inner cavity
- Design ruleC-0761 designs removed, cycle 4
hERG pore, PDB 5VA1
Why it matters
Past failures help future teams avoid repeating the same design path.
The loop
Program knowledge deepens with each design cycle.
Design, simulation and experiment remain connected. New results update the program’s knowledge, models and design principles before the next cycle begins.
Closed-loop design
What the loop measures, read on a representative KRAS G12C series
Error by cycle
The endpoint models are calibrated on the program’s chemical series. Every assay result retrains them before the next library is scored, so prediction error enters the next model-calibration set.
Why it matters
The next cycle starts with a stronger understanding of the therapeutic program.
Scientific agency
Coordinate models, simulations and experiments. Keep the scientific reasoning available for review.
Rayca plans the analyses required by a program question, runs the relevant models and simulations and checks the results before they reach the project team.
Agent run 4-117
RunningGoal: nominate 8 analogues for cycle 5 that clear C-07 to C-12 with predicted ΔΔG at or below −1.0 kcal/mol. Read on a representative KRAS G12C series.
Events
- 3:56ComputeJob 71 preempted, resubmitted on gpu-h100-07
- 3:51SimulationRBFE edge RC-0504 to RC-0440 converged, −1.62 ± 0.21 kcal/mol
- 2:26PlannerPlan revised: R3 re-enumerated under C-09, RC-0523 replaces RC-0521
- 2:12CriticFlag: RC-0521 bound-state strain 2.4 kcal/mol, above C-09
- 0:44MemoryModels retrained on cycle 4; rule C-12 written
Why it matters
Project scientists retain control of design, experiment and development decisions.
Design
Invent therapeutic candidates within what the program has learned.
Generative models propose candidates across modalities within biological and program constraints. Predictive models and physical simulation evaluate activity, selectivity, developability and feasibility before experimental commitment.
Design workbench
Each candidate is reviewed in the pocket, against the lead and the full target profile, before a chemist approves it. Read on a representative KRAS G12C series.
Acylated distal nitrogen removes the basic centre; fluorine holds the switch-II groove contact.
Profile against the TPP
RC-0440 lead RC-0501- Predicted pIC508.4 → 8.9
- ΔΔG vs RC-0440 kcal/mol0.0 → -1.84
- hERG IC50, predicted µM34 → 52
- cLogP3.8 → 2.9
- TPSA Ų78 → 92
- Bound-state strain kcal/mol0.9 → 0.8
- Synthetic steps6 → 6
Predicted pIC50 8.9 ± 0.18 · probability of meeting the TPP 81%
Why it matters
Scientists review a focused set of candidates with evidence and uncertainty attached.
Rayca considers the properties required for a candidate to be made, formulated and delivered from the beginning of design. The relevant profile changes with each modality and intended product.
Developability
Which form of the molecule should go forward?
The same molecule can be made as different salts and crystal forms, and they behave very differently. Rayca weighs each form's solubility against its physical stability and recommends one, with the reasons attached.
Sample runlead RC-0440 of a KRAS G12C series
- HCl salt, Form ASelected: soluble and stable
- HCl salt, Form BChanges form in humidity
- Mesylate saltAbsorbs water from the air
- Free baseStable, but too insoluble
Why it matters
Development risks are visible before the therapeutic candidate is fixed.
Rayca connects formulation and delivery choices to distribution, exposure and projected dose, allowing the intended product to influence candidate design early.
Delivery
Does the drug stay high enough, for long enough?
To work, the level in the blood has to stay above what is needed to block the target, right up to the next dose. Rayca predicts the full day's profile for each formulation and shows where it drops below.
Sample runlead RC-0440 of a KRAS G12C series
Level in the blood over one day, 100 mg once daily
Dashed line: the level needed to block the target
The dispersion covers the whole day. The tablet dips below the line before the next dose, which lets the target recover.
Why it matters
Product requirements can influence candidate and formulation design before nomination.
Small molecules, peptides, proteins, antibodies, degraders and delivery systems require different scientific methods. Rayca keeps their evidence, models and decisions within the same learning architecture.
01The match
Small molecule
A deep pocket inside the cell
Taken by mouth and able to reach targets inside the cell. A mutant cysteine in the pocket lets the molecule anchor covalently.
What the engine scores for it
Shared with every modality
- Typical size
- Under 500 Da
- Reaches
- Inside the cell
- Route
- Oral
- Shown
- KRAS G12C with a covalent inhibitor, PDB 6OIM
Why it matters
Programs can explore or change modality without losing accumulated knowledge.
Rayca applies learned models and docking first. Decision-critical compounds advance to molecular dynamics, free energy or quantum chemistry when the project requires greater resolution.
Physics, in layers
Each layer is slower and more exact than the one above it, and only sees what that layer lets through.
Slower, more exact
The question it answers
How much tighter does it bind than the lead, and how sure are we?
Relative binding free energy between close analogues, with replicas and cycle closure for a stated error.
Used in
Every run is scheduled across GPU and HPC, retried on failure and kept with its inputs, method and version, so any result can be reproduced.
Why it matters
Rigorous computation is spent where it changes molecular action.
Program questions answered with inspectable reasoning.
Each representative case exposes the evidence, modelling, physics, uncertainty and lineage behind the proposed action.
Kinase inhibitor series, representative
The question
Can the series lose its heart-safety risk without losing potency?
01Recall
Memory finds every compound that failed on hERG and the one structural feature they share.
Memory02Rule
The basic amine is encoded as a design constraint; subsequent analogues preserve potency using alternative solvent-facing groups.
Design rules03Check
Docking and free energy confirm the potency holds before anything is made.
Physics
The decision
Make the amide analogues; skip the amines.
hERG 2.1 to 38 µM
Potency held, pIC50 8.3 to 8.1
hERG channel pore, PDB 5VA1
A shared learning architecture for therapeutic design. Molecular evidence, models, simulation and experiment.
Each layer contributes to the same program understanding. Experimental outcomes update the models and design principles used in the next cycle.
- 07Program decisionsPrioritized designs and experiments with their reasoning attached.
- 06Governance and lineageResults retain inputs, method versions, uncertainty and provenance.
- 05Governed executionRayca plans and supervises models, simulations and compute resources.
- 04Models and simulationPredictions and physical methods matched to the scientific question.
- 03Multimodal designCandidates proposed using methods appropriate to each therapeutic modality.
- 02Learned design principlesSupported patterns of success and failure available to future design.
- 01Persistent molecular knowledgeCandidates, biological evidence, experiments and decisions held in context.
Scientists remain accountable. Rayca develops the reasoning behind each option.
Rayca runs the computation and presents the evidence, assumptions and uncertainty behind each recommendation. Program strategy and all go or no-go decisions remain with the scientific team, who can inspect the molecular reasoning before acting.
Questions and answers.
What is the Rayca platform?
Rayca is the operating intelligence for molecular R&D. It connects molecular evidence, models, simulation and experiments across discovery, development, formulation and delivery. Scientists set direction and make every go or no-go decision.
How does Rayca fit into an existing drug discovery organization?
Rayca works with the scientific data, models and compute environments an organization already uses. Inputs, methods and outputs remain available for review.
How does Rayca differ from a standalone molecule generator?
Rayca connects candidate generation with program evidence, failure patterns, simulation, experimental results and development requirements across modalities.
How does the learning loop operate?
Program evidence guides design and simulation. Experimental results update the program knowledge, models and design principles used in the next cycle.
What are learned molecular failure modes?
They are supported relationships between molecular features and observed liabilities. Rayca applies them as constraints during molecular generation and triage.
How does Rayca represent uncertainty in physics-based calculations?
Calculations retain their ensembles, uncertainty, method version and inputs. Scientists can inspect and reproduce them before they inform a decision.
Which stages of discovery and development does Rayca cover?
Rayca connects therapeutic design with developability, formulation, delivery, exposure and projected dose. The relevant methods depend on modality and intended product.
How does Rayca manage GPU and HPC computation?
Agents plan, place and supervise work across available GPU and HPC resources. Every run retains its inputs, versions, resources and outputs.
Which modalities does the platform support?
Rayca supports small molecules, peptides, proteins, antibodies, nanobodies, degraders and delivery systems using modality-appropriate models and simulations.
Who is accountable for program decisions?
Rayca runs computation and presents the molecular reasoning. Scientific teams remain accountable for strategy and every go or no-go decision.
See the platform on your own program. Engaged program by program, within your institution.