SDLabs has been used across pharma, chemical, and analytical domains. These use cases show how Bayesian optimization accelerates real workflows.
Each use case below is built on a synthetic landscape — a mathematical function that captures the key trade-offs, interactions, and deceptive optima found in the real problem. The synthetic landscape lets you run the full optimization workflow in SDLabs without lab access, and observe how the AI model learns the response surface, navigates trade-offs, and converges on optimal conditions.
Summary
Use Case | Partner / Domain | Parameters | Objectives | Key Result | Link |
Hydroformylation | Fragrance industry / Catalysis | 5 (3 numerical + 2 categorical) | Conversion, Linear Selectivity | 10-30x Rh reduction | |
HPLC Method Dev | Analytical Chemistry | 6 (4 numerical + 2 categorical) | Peak Resolution, Total Runtime | Optimal separation in <30 experiments | |
Takeda Deprotection | Takeda / Pharma | 5 (4 numerical + 1 categorical) | Yield | 50% → 90%+ yield | |
Security Ink Formulation | SICPA / Printing | 8 (6 numerical + 2 categorical) | Color Shift, Adhesion, Stability, Viscosity | Optimal formulation in ~18 experiments | |
T Cell Circuit Engineering | Stanford / Immunotherapy | 6 (4 numerical + 2 categorical) | Cytotoxicity, Sparing, Persistence | Optimal circuit in ~20 experiments | |
CO2-to-Methanol Catalyst | ETH Zurich (SwissCAT+) / Catalysis | 8 (all numerical, metal loadings) | MeOH Selectivity, CO2 Conversion, CH4 Selectivity, Metal Cost | 100 years of catalyst R&D in 6 weeks |
Common SDLabs Features Demonstrated
Multi-objective (Chimera hierarchy) — rank objectives by importance with tolerances
Categorical variables — discrete choices like solvents, ligands, and column types
Mixture constraints — enforce that component fractions sum to 100%
Target objectives — optimize toward a specific value (e.g., viscosity = 25 Pa.s)
Expert context — inject domain knowledge to guide the optimizer
Automated iteration via API — run optimization loops programmatically