Intro & scope
Use cases that accelerate discovery and engineering work with AI. Four patterns: SCI.LIFE (proteins, drugs, genomics, targets, breeding, toxicology), SCI.MATERIALS (materials, reactions, formulations), SCI.ENGINEERING (simulation, design, CAD/CAM, EDA, test data, systems and quality engineering) and SCI.METHOD (cross-cutting methods: lab automation, imaging, spectra, climate, subsurface, AI scientists, theorem proving). Most are minimal-risk R&D tools on non-personal data; the exceptions involve human genetic or patient-derived data.
Scope boundaries: - SCI-001 (protein structure/design) vs SCI-021 (nucleotide sequence design) vs SCI-009 (human variant interpretation, clinical) vs SCI-020 (breeding). - SCI-002 (small-molecule discovery) vs SCI-017 (targets and biomarkers upstream) vs SCI-015 (toxicity/regulatory safety) vs SCI-004 (synthesis routes). - SCI-003 (materials discovery) vs SCI-014 (company-specific formulation on small data) vs SCI-007 (closed-loop lab execution). - SCI-008 (images) vs SCI-018 (spectra and chromatography): same analytical-lab setting, different data modality. - SCI-005 (learned CFD/FEM surrogates) vs SCI-006 (generative design/topology optimisation) vs SCI-010 (earth-system emulators) vs SCI-019 (subsurface). - SCI-013 (MBSE/requirements) vs SCI-025 (FMEA/8D/CAPA): both document-grounded LLM assistants; 013 covers specification and model consistency, 025 failure analysis and corrective actions. - SCI-022 (test-data ML) vs SCI-025 (quality documents) vs SCI-023 (CAM) vs SCI-024 (CAD/drawings). - SCI-011 (agentic literature-to-hypothesis) vs SCI-012 (formal proof) vs SCI-007 (physical experiments).
AI Act: minimal for nearly all. SCI-009 is "depends": clinical variant reports used for diagnosis can fall under IVDR/MDR and high-risk rules. GDPR applies where human genomic or patient data is used (SCI-009, SCI-017; "depends" for SCI-008, SCI-011, SCI-013, SCI-014, SCI-018, SCI-020, SCI-025).
Use cases
| ID | Title | Pattern | Maturity | Path | HW tier | AI Act risk |
|---|---|---|---|---|---|---|
| SCI-001 | Protein structure prediction & protein design | SCI.LIFE | early-adopter | B | HW3 | minimal |
| SCI-002 | Drug discovery: virtual screening & molecule generation | SCI.LIFE | early-adopter | B | HW3 | minimal |
| SCI-003 | Materials discovery & property prediction | SCI.MATERIALS | early-adopter | C | HW3 | minimal |
| SCI-004 | Reaction prediction & retrosynthesis planning | SCI.MATERIALS | early-adopter | A | HW2 | minimal |
| SCI-005 | Simulation surrogates | SCI.ENGINEERING | early-adopter | A | HW3 | minimal |
| SCI-006 | Generative engineering design & topology optimisation | SCI.ENGINEERING | mainstream | A | HW3 | minimal |
| SCI-007 | Lab automation & self-driving laboratories | SCI.METHOD | early-adopter | A | HW1 | minimal |
| SCI-008 | Microscopy & scientific image analysis | SCI.METHOD | mainstream | C | HW2 | minimal |
| SCI-009 | Genomics & variant interpretation | SCI.LIFE | early-adopter | A | HW3 | depends |
| SCI-010 | Climate & earth-system modelling | SCI.METHOD | early-adopter | C | HW3 | minimal |
| SCI-011 | AI scientist agents & hypothesis generation | SCI.METHOD | emerging | B | HW4 | minimal |
| SCI-012 | Mathematical reasoning & formal theorem proving | SCI.METHOD | early-adopter | B | HW4 | minimal |
| SCI-013 | Systems engineering & requirements management | SCI.ENGINEERING | early-adopter | A | HW3 | minimal |
| SCI-014 | AI-guided formulation & recipe development | SCI.MATERIALS | early-adopter | A | HW1 | minimal |
| SCI-015 | In-silico toxicology & chemical safety assessment | SCI.LIFE | early-adopter | A | HW1 | minimal |
| SCI-016 | AI-driven chip & PCB design automation | SCI.ENGINEERING | early-adopter | A | HW3 | minimal |
| SCI-017 | Drug target identification & biomarker discovery from multi-omics | SCI.LIFE | early-adopter | A | HW3 | minimal |
| SCI-018 | Spectral & analytical data interpretation | SCI.METHOD | early-adopter | C | HW2 | minimal |
| SCI-019 | Subsurface intelligence: mineral exploration, seismic interpretation, geothermal & CO2 storage siting | SCI.METHOD | early-adopter | B | HW3 | minimal |
| SCI-020 | Genomic selection & AI-assisted plant and animal breeding | SCI.LIFE | mainstream | C | HW2 | minimal |
| SCI-021 | DNA/RNA sequence design: mRNA, regulatory elements & strains | SCI.LIFE | early-adopter | B | HW3 | minimal |
| SCI-022 | Engineering test-data analytics & physical test reduction | SCI.ENGINEERING | early-adopter | A | HW1 | minimal |
| SCI-023 | CNC/CAM programming automation | SCI.ENGINEERING | early-adopter | A | HW1 | minimal |
| SCI-024 | Text/sketch-to-CAD modelling & manufacturing drawing automation copilot | SCI.ENGINEERING | early-adopter | A | HW3 | minimal |
| SCI-025 | FMEA & 8D/CAPA quality-engineering assistant | SCI.ENGINEERING | early-adopter | A | HW3 | minimal |
Shared stack
Common building blocks: scientific foundation models (AlphaFold-class, MLIPs, ESM, nucleotide LMs), tabular/graph ML, Bayesian optimisation and solvers, simulation, document RAG (embedding + reranker + vector DB + knowledge graph), agent frameworks, workflow engines.
- HW1 (CPU/small GPU, workstation or VM): SCI-007 control layer, SCI-014, SCI-015, SCI-022, SCI-023. Stack: Python + scikit-learn/XGBoost/BoTorch/Ax, RDKit, OR-Tools, PostgreSQL + object storage, MLflow, Airflow or Prefect. Reuse for PRED/OPT families.
- HW2 (single 24-48 GB GPU): SCI-004, SCI-008, SCI-018, SCI-020. Stack: PyTorch, Cellpose/SAM-class segmenters, retrosynthesis and spectra models, ONNX/Triton serving, DVC for datasets, JupyterHub.
- HW3 (multi-GPU node, 80 GB class, or EU cloud GPU): SCI-001, SCI-002, SCI-003, SCI-005, SCI-006, SCI-009, SCI-010, SCI-013, SCI-016, SCI-017, SCI-019, SCI-021, SCI-024, SCI-025. Stack: Slurm or Kubernetes with GPU operator, vLLM serving an open-weight 30-70B LLM, Qdrant/pgvector + bge-m3 embeddings and a reranker, Neo4j for knowledge graphs, docling/OCR for documents, Nextflow/Snakemake for pipelines, a shared model/weights registry. Same LLM+RAG layer serves SCI-013/025 and DOC/KNOW families.
- HW4 (multi-node GPU cluster or frontier API under EU DPA): SCI-011, SCI-012. Stack: reasoning LLM (open-weight 100B+ MoE or EU-hosted API), agent framework with sandboxed code execution, Lean/Isabelle toolchain, same retrieval layer as HW3.
Cross-cutting: keep personal genomic/patient data (SCI-009, SCI-017) in a separate EU-hosted, access-controlled enclave with pseudonymisation and DPIA; everything else may share the HW3 platform. Log provenance of data, model versions and prompts for reproducibility and human review.