Therapeutic antibody engineering and optimization, lead-to-clinical-candidate. Covers sequence humanization (germline alignment, framework retention), affinity maturation, developability (aggregation, stability, PTMs), structure modeling (AlphaFold/PDB CDR analysis), immunogenicity prediction, and manufacturing feasibility. Use for biologic-drug optimization, mAb design review, biosimilar engineering, and clinical-precedent comparison.
84%
Does it follow best practices?
Run evals on this skill
Adds up to 20 points to the overall score
Passed
No findings from the security scan
AI-guided antibody optimization pipeline from preclinical lead to clinical candidate. Covers sequence humanization, structure modeling, affinity optimization, developability assessment, immunogenicity prediction, and manufacturing feasibility.
KEY PRINCIPLES:
When uncertain about any scientific fact, SEARCH databases first (PubMed, UniProt, ChEMBL, ClinVar, etc.) rather than reasoning from memory. A database-verified answer is always more reliable than a guess.
Apply when user asks:
antibody_optimization_report.mdoptimized_sequences.fasta - All optimized variantshumanization_comparison.csv - Before/after comparisondevelopability_assessment.csv - Detailed scoresSee REPORT_TEMPLATE.md for the full report template with section formats.
Every optimization MUST include per-variant documentation with:
| Tool | Purpose | Category |
|---|---|---|
IMGT_search_genes | Germline gene identification | Humanization |
IMGT_get_sequence | Human framework sequences | Humanization |
SAbDab_search_structures | Antibody structure precedents | Structure |
TheraSAbDab_search_by_target | Clinical antibody benchmarks | Validation |
ESMFold_predict_structure | Fv structure modeling (VH-linker-VL scFv) | Structure |
alphafold_get_prediction | Retrieve a precomputed AlphaFold DB model by UniProt accession (target antigen) | Structure |
iedb_search_epitopes | Epitope identification | Immunogenicity |
iedb_search_bcell | B-cell epitope prediction | Immunogenicity |
UniProt_get_entry_by_accession | Target antigen information | Target |
STRING_get_interaction_partners | Protein interaction network | Bispecifics |
PubMed_search_articles | Literature precedents | Validation |
CRITICAL: SOAP tools (IMGT, SAbDab, TheraSAbDab) require an operation parameter. See QUICK_START.md for correct usage.
Phase 1: Input Analysis & Characterization
├── Sequence annotation (CDRs, framework)
├── Species identification
├── Target antigen identification
├── Clinical precedent search
└── OUTPUT: Input characterization
↓
Phase 2: Humanization Strategy
├── Germline gene alignment (IMGT)
├── Framework selection
├── CDR grafting design
├── Backmutation identification
└── OUTPUT: Humanization plan
↓
Phase 3: Structure Modeling & Analysis
├── AlphaFold prediction
├── CDR conformation analysis
├── Epitope mapping
├── Interface analysis
└── OUTPUT: Structural assessment
↓
Phase 4: Affinity Optimization
├── In silico mutation screening
├── CDR optimization strategies
├── Interface improvement
└── OUTPUT: Affinity variants
↓
Phase 5: Developability Assessment
├── Aggregation propensity
├── PTM site identification
├── Stability prediction
├── Expression prediction
└── OUTPUT: Developability score
↓
Phase 6: Immunogenicity Prediction
├── MHC-II epitope prediction (IEDB)
├── T-cell epitope risk
├── Aggregation-related immunogenicity
└── OUTPUT: Immunogenicity risk score
↓
Phase 7: Manufacturing Feasibility
├── Expression level prediction
├── Purification considerations
├── Formulation stability
└── OUTPUT: Manufacturing assessment
↓
Phase 8: Final Report & Recommendations
├── Ranked variant list
├── Experimental validation plan
├── Next steps
└── OUTPUT: Comprehensive reportGoal: Annotate sequences, identify species/germline, find clinical precedents.
Key steps:
IMGT_search_genesTheraSAbDab_search_by_targetUniProt_get_entry_by_accessionOutput: Sequence information table, CDR annotation, target info, clinical precedent list.
See WORKFLOW_DETAILS.md Phase 1 for code examples.
Goal: Select human framework, design CDR grafting, identify backmutations.
Key steps:
Output: Framework selection rationale, grafting design, backmutation analysis, humanized sequences.
See WORKFLOW_DETAILS.md Phase 2 for code examples.
Goal: Predict structure, analyze CDR conformations, map epitope.
Key steps:
ESMFold_predict_structure (VH-linker-VL scFv; alphafold_get_prediction only retrieves precomputed models by UniProt accession)iedb_search_epitopesSAbDab_search_structuresOutput: Structure quality table, CDR conformation analysis, epitope mapping, structural comparison.
See WORKFLOW_DETAILS.md Phase 3 for code examples.
Goal: Design affinity-improving mutations via computational screening.
Key steps:
Output: Ranked mutation hypotheses, combination strategy, computed metrics where supported, and explicit unverified affinity/conditional-binding claims. Do not invent KD ranges or fold improvements.
See WORKFLOW_DETAILS.md Phase 4 for code examples.
Goal: Comprehensive developability scoring (0-100) across five dimensions.
Key steps:
scripts/developability.py --seq <VH/VL>. It returns AGGRESCAN aggregation-prone regions (real per-residue propensity), isoelectric point, and Kyte-Doolittle hydrophobic patches.Scoring: Weighted average (aggregation 0.30, PTM 0.25, stability 0.20, expression 0.15, solubility 0.10). Tiers: T1 (>75), T2 (60-75), T3 (<60).
Output: Component scores, overall score, tier classification, mitigation recommendations.
See WORKFLOW_DETAILS.md Phase 5 and CHECKLISTS.md for scoring details.
Goal: Predict immunogenicity risk and design deimmunization strategy.
Key steps:
Output: T-cell epitope list, risk score breakdown, deimmunization strategy, clinical comparison.
See WORKFLOW_DETAILS.md Phase 6 for code examples.
Goal: Assess expression, purification, formulation, and CMC feasibility.
Key steps:
CodonUsage_get_optimal_codons(taxid=<Cricetulus griseus NCBI taxid>) gives the CHO-preferred codon per amino acid to compare against; resolve the taxid first if not already known rather than assuming one. CodonUsage_get_table (same taxid) additionally gives per-codon usage frequency, needed to flag which codons in the actual sequence are rare in CHO specifically -- human codon-usage tables (e.g. from tooluniverse-sequence-analysis) are the wrong reference here, rare-in-human is not the same set as rare-in-CHOOutput: Expression assessment, purification strategy, formulation recommendation, CMC timeline.
See MANUFACTURING.md for detailed manufacturing content and WORKFLOW_DETAILS.md Phase 7 for code.
Goal: Compile all findings into a ranked recommendation with validation plan.
Key outputs:
See REPORT_TEMPLATE.md for the full report template.
IMGT_search_genes: Search germline genes (IGHV, IGKV, etc.)IMGT_get_sequence: Get germline sequencesIMGT_get_gene_info: Database informationSAbDab_search_structures: Search antibody structuresSAbDab_get_structure: Get structure detailsTheraSAbDab_search_therapeutics: Search by nameTheraSAbDab_search_by_target: Search by target antigeniedb_search_epitopes: Search epitopesiedb_search_bcell: B-cell epitopesiedb_search_mhc: MHC-II epitopesiedb_get_epitope_references: CitationsESMFold_predict_structure: Fv structure prediction from a sequencealphafold_get_prediction: Retrieve a precomputed AlphaFold DB model by UniProt accessionUniProt_get_entry_by_accession: Target infoRCSBData_get_entry: Experimental structuresAntibodyRegistry_search: Find validated commercial antibodies for a target (by target gene/protein name, antibody name, or vendor keyword). Returns each hit's RRID, target (with UniProt/Entrez id), clonality, host organism, applications, vendor/catalog number, and defining citation. Useful during Phase 1 (target/precedent research) to see what antibody reagents already exist against a target antigen before designing a new one, or during Phase 8 (validation planning) to identify a benchmark reagent.AntibodyRegistry_get_by_rrid: Resolve a specific RRID (e.g. AB_2298772, accepts the bare numeric id or RRID:AB_... form too) cited in a methods section to its full record — verify which exact antibody a paper used rather than assuming from the name alone. Live-verified: both tools return real records (e.g. AntibodyRegistry_search(query="GFAP") -> a real GFAP rabbit mAb with UniProt target P14136).SciCrunch_resolve_rrid: General RRID resolver for everything that ISN'T an antibody — cell lines (CVCL_ prefix), model organisms, software/tools (SCR_ prefix), and plasmids. Use it when a clinical-precedent or validation paper's methods section also cites the cell line or software used alongside an antibody. It will still resolve an AB_ RRID, but only returns generic fields (name, description, categories, proper_citation) — no target/clonality/host — so prefer AntibodyRegistry_get_by_rrid for antibody RRIDs specifically. Live-verified: SciCrunch_resolve_rrid(rrid="CVCL_0045") returns the real HEK293 cell-line record (12 synonyms, proper_citation: "NCBI_Iran Cat# C497, RRID:CVCL_0045"); on the same AB_2298772 antibody RRID above it returns only name/description/categories with no target info, confirming the richer-antibody-fields distinction.STRING_get_interaction_partners: Protein interactionsSTRING_get_enrichment: Pathway analysis| File | Contents |
|---|---|
QUICK_START.md | Getting started guide, SOAP tool parameters, Python SDK and MCP usage |
WORKFLOW_DETAILS.md | Code examples for all 8 phases |
REPORT_TEMPLATE.md | Full report template with section formats and example tables |
MANUFACTURING.md | Detailed manufacturing content (expression, purification, formulation, CMC) |
EXAMPLES.md | Complete clinical scenario examples (humanization, affinity, bispecific) |
CHECKLISTS.md | Evidence grading, completeness checklists, scoring details, special considerations |
mims-harvard/ToolUniverse
37bb81a
If you maintain this skill, you can claim it as your own. Once claimed, you can manage eval scenarios, bundle related skills, attach documentation or rules, and ensure cross-agent compatibility.