SCIENTIFIC SUBSTITUTION OS
SUBSTØR
Latent space navigation for innovation
Molecular substitution intelligence for formulation teams. Map a source ingredient, apply your constraints, and receive an explainable shortlist of candidates.
REGULATORY
Anticipate the next restriction
Substitution is rarely a free choice — it’s usually triggered by something outside the lab.
What actually forces a substitution
- A regulatory restriction removes an ingredient you rely on.
- A retailer or brand standard rules out an active before the law does.
- A supplier shortage or discontinuation breaks your sourcing.
- Public controversy around an ingredient forces an early exit.
Frameworks relevant to cosmetics
Regulation (EC) No 1223/2009
The EU Cosmetic Products Regulation and its restriction annexes.
REACH
EU rules on the registration and restriction of chemical substances.
SCCS opinions
Scientific Committee on Consumer Safety assessments that often precede a restriction.
CosIng updates
The European Commission’s cosmetic ingredient database.
{{TODO: confirm which of these SUBSTØR currently covers, and whether non-EU frameworks are supported}}
SUBSTØR filters candidates on their regulatory status, not just their chemical structure — so a candidate that scores well structurally but is already flagged for review doesn’t make it onto your shortlist unnoticed.
PLATFORM
Built first for cosmetics — the same engine reaches further
Cosmetics is where SUBSTØR is most mature today: map the source ingredient, define your constraints, rank alternatives, then explain why each candidate deserves a closer look.
- Reduce discovery cycles from slow manual screening to guided candidate maps.
- Connect in-silico scoring with practical formulation and sourcing constraints.
- Keep decisions explainable so scientific, product and business teams can align.
Cosmetics
irritation-checkedFind safer, available substitutes for actives, fragrances and other formulation-sensitive ingredients — before a restriction or shortage forces the decision.
The same engine applies to polymers, materials and metallurgy.
Polymers
PEG-b-PLA→PEG-b-PCL
Materials
LiCoO2→LiFePO4
Metallurgy
316L Steel→Lean duplex
HOW IT WORKS
From ingredient to candidate map
SUBSTØR turns molecular substitution into a navigable scientific workflow, combining latent geometry, graph reasoning and constraint-aware ranking.
Source ingredient
Start from a cosmetic active ingredient, molecular structure or SMILES representation.
Latent mapping
Project molecules into a navigable latent space built from chemical descriptors and similarity graphs.
Constraint filtering
Filter candidates by bioactivity, safety, stability, formulation compatibility and availability.
Ranked substitutes
Return explainable alternatives with scores, molecular pathways and candidate rationales.
DEMO
Try the substitution workflow in three decisions
Illustrative example. Candidates and scores are pre-computed for demonstration and shift as you change the inputs below — real projects run on your own structures.
Input constraints
SOURCE INGREDIENT
PRIORITY FILTER
GENERATED CANDIDATE MAP
Source: Retinol
Bakuchiol
0.91Ranked by property alignment, feasibility and selected constraint weight.
beta-Ionone
0.78Ranked by property alignment, feasibility and selected constraint weight.
Granactive Retinoid
0.76Ranked by property alignment, feasibility and selected constraint weight.
NO BLACK BOX
Every suggestion comes with a clear reason why
We don’t just hand you a black-box AI guess. Every candidate is checked step by step against your own requirements, so you can see exactly why it made the shortlist — and trust it enough to take it to the lab.
Matched by real chemistry
Candidates are compared to the source ingredient by actual molecular structure, not just by category or name.
Your rules, built in
Anything that fails your safety, stability or sourcing requirements is filtered out automatically, before it reaches you.
Answers in minutes, not weeks
Skip weeks of manual research. Get a ranked shortlist ready for lab testing right away.
FIVE STEPS, EVERY TIME
See a full example →Read the molecule
Break it down into its chemical properties
RUNNINGFind similar matches
Scan for molecules with a similar structure
QUEUEDApply your rules
Remove anything that fails your requirements
QUEUEDRank the results
Score what’s left from best to worst fit
QUEUEDGenerate report
Explain why each candidate was chosen
QUEUEDDATA & VALIDATION
Where the answers actually come from
Before “latent space” and “similarity graph” mean anything, here’s what’s behind them.
Where does the data come from?
The platform indexes public and licensed chemical databases covering the domains it supports — {{TODO: confirm sources, e.g. PubChem, ChEMBL, CosIng/INCI for cosmetics}}. Structures a customer submits are added to a private layer that is never shared with other customers or used to train shared models.
How much is actually covered?
{{TODO: confirm current molecule count and which domains are production-ready today}}. Cosmetics has the deepest coverage right now; the same engine is being extended to the other domains shown on this site.
How is a molecule represented?
Each molecule is turned into a numerical representation — {{TODO: confirm method: molecular fingerprints, learned embeddings, similarity graphs, or a combination}} — so candidates can be compared by structural and physicochemical similarity, not just by name or category.
How do you know it works?
We validate with a leave-one-out approach: known, documented substitutions are removed from the reference set, and we check whether the model recovers them from the remaining data alone. {{TODO: publish current results — recovery rate, sample size, domains tested}}.
CONFIDENTIALITY & IP
Your formulas never leave your control
R&D teams evaluate proprietary molecular structures with us. Here is exactly what happens to that data.
Your structures stay yours
Molecules and structures you submit are never used to train our models, and are never shared with other customers.
Where your data lives
Hosted in {{TODO: confirm hosting region, e.g. EU (Paris)}}, encrypted in transit and at rest.
Access control
Access to submitted structures is limited to {{TODO: confirm which roles/team members can view customer data}}. Data is retained for {{TODO: confirm retention period}} and deleted on request.
Deployment options
Available today as a SaaS platform. {{TODO: confirm whether dedicated VPC or on-premise deployment is actually offered}}
NDA available before any technical discussion — get in touch.
BEFORE / AFTER
Manual screening vs. SUBSTØR
| Manual screening | SUBSTØR | |
|---|---|---|
| Time per search | Days to weeks of literature and database search per candidate. | Minutes to generate a ranked, constraint-checked shortlist. |
| Candidates evaluated | Limited by how much one researcher can hold in their head. | A broad, systematic sweep across the indexed latent space. |
| Decision traceability | Notes and spreadsheets that are hard to reconstruct later. | Every ranking is tied to the structure and constraints it was scored against. |
| Cost per project | Scales roughly linearly with researcher hours. | Marginal cost falls as the shared latent space grows — see Efficiency. |
EFFICIENCY
Lower your costs. Screen more molecules.
The latent space gets richer with every project, so each new search costs less than the last. Teams end up evaluating far more candidates before committing anything to the lab.
Reused, not repeated
Each project adds to the shared latent space, so later searches start from more context instead of a blank slate.
Fewer dead ends
Constraint filtering rules out weak candidates early, before they reach a chemist’s desk.
COST PER CANDIDATE
ConceptualIllustration of the expected effect, not measured results — the underlying latent space compounds with each completed project.
FAQ
Questions we get asked early
What data do I need to provide to start?
A source ingredient — name, CAS number, molecular structure or SMILES — and the constraints that matter to you: safety, sourcing, cost or regulatory status. No proprietary formula is required to get a first candidate map.
Do you support SMILES / InChI / structure files? Which formats?
{{TODO: confirm exact supported input formats — SMILES, InChI, SDF, MOL?}}
Does SUBSTØR replace lab validation?
No. It narrows a long list of possible substitutes down to a short, ranked, explainable one. Every candidate still needs lab validation before it’s used in a product.
How is this different from a similarity search in a public database?
A plain similarity search ranks candidates by structure alone. SUBSTØR combines structural similarity with your own constraints — safety, stability, sourcing, regulatory status — so the ranking reflects what’s actually usable, not just what’s chemically close.
Who owns the results?
You do. Candidate maps and reports generated from your projects belong to you — see our Privacy Policy and Terms of Service for details.
How long does onboarding take?
{{TODO: confirm typical onboarding timeline}}
Which domains beyond cosmetics are production-ready today?
Cosmetics has the deepest coverage today. Polymers, materials and metallurgy run on the same engine and are being extended — {{TODO: confirm current maturity per domain}}.
TALK TO US
Book a scientific demo
Tell us about the ingredient or material you’re looking to substitute. A member of our scientific team will follow up within one business day.
We only use the details below to respond to your request. See our Privacy Policy for details.