Comprehensive antibody optimization skill for therapeutic development. From mouse antibody to clinical candidate in one workflow.
from tooluniverse import ToolUniverse
tu = ToolUniverse()
tu.load_tools()
# Example 1: Humanize a mouse antibody
result = tu.chat("""
Humanize this mouse anti-PD-L1 antibody:
VH: EVQLVESGGGLVQPGGSLRLSCAASGYTFTSYYMHWVRQAPGKGLEWVSGIIPIFGTANYAQKFQGRVTISADTSKNTAYLQMNSLRAEDTAVYYCARDDGSYSPFDYWGQGTLVTVSS
VL: DIQMTQSPSSLSASVGDRVTITCRASQSISSYLNWYQQKPGKAPKLLIYAASSLQSGVPSRFSGSGSGTDFTLTISSLQPEDFATYYCQQSYSTPLTFGQGTKVEIK
Target: PD-L1 (Q9NZQ7)
""")Output: Humanized sequences with 85-90% human framework, CDR preservation, developability assessment, and recommendations.
-
Humanization (Example 1)
- Identify germline genes (IMGT)
- Design CDR grafting strategy
- Identify backmutations for CDR support
- Predict affinity retention
-
Affinity Maturation (Example 2)
- Analyze binding interface
- Design affinity-improving mutations
- In silico screening for optimal variants
- Balance affinity vs. developability
-
Developability Assessment (Example 3)
- Identify aggregation-prone regions
- Detect PTM liability sites
- Predict stability and expression
- Calculate comprehensive developability score (0-100)
-
Immunogenicity Prediction (Example 6)
- Predict T-cell epitopes (IEDB)
- Calculate immunogenicity risk score
- Design deimmunization strategy
- Compare to clinical precedents
-
Manufacturing Assessment (Example 7)
- Expression optimization (CHO cells)
- Purification strategy design
- Formulation recommendations
- CMC timeline and cost estimates
-
Advanced Engineering
- Bispecific antibodies (Example 4)
- pH-dependent binding (Example 5)
- Aggregation mitigation (Example 3)
- CDR optimization strategies
Creates antibody_optimization_report.md with progressive updates throughout analysis.
- Humanization Score: % framework identity to human germline
- Developability Score: 0-100 combining aggregation, PTM, stability, expression
- Immunogenicity Risk: Low/Medium/High based on T-cell epitopes
- Evidence Grading: T1-T4 tiers for variant ranking
- Uses IMGT germline database for humanization
- Benchmarks against approved antibodies (TheraSAbDab)
- Validates with clinical precedents
- Structural modeling (AlphaFold)
- Antibody sequence (VH and/or VL)
- Target antigen (UniProt ID or name)
- Optimization goal (humanization, affinity, developability)
- Current affinity (KD)
- Known issues (aggregation, immunogenicity)
- Species (if not mouse)
- Epitope information
- Clinical development stage
File: antibody_optimization_report.md
Contains:
- Input characterization
- Humanization strategy (if applicable)
- Structure modeling & analysis
- Affinity optimization (if requested)
- Developability assessment
- Immunogenicity prediction
- Manufacturing feasibility
- Final recommendations
File: optimized_sequences.fasta
All designed variants in FASTA format with annotations:
>VH_Humanized_v2 | 85% human | KD=2.1nM | Dev=79/100
EVQLVQSGAEVKKPGASVKVSCKASGYAFTSYYMHWVRQAPGQGLEWMV...
File: humanization_comparison.csv or developability_assessment.csv
Detailed metrics for all variants.
Query: "Humanize this mouse antibody"
Process:
- CDR annotation (IMGT numbering)
- Germline gene search (IMGT)
- Framework selection (identity + clinical use)
- CDR grafting design
- Backmutation analysis
- Structure validation (AlphaFold)
- Developability scoring
Output: 2-3 humanized variants (85-90% human) with recommendations
Timeline: ~30 minutes for full analysis
Query: "Improve affinity from 15 nM to <5 nM"
Process:
- Interface analysis
- Hotspot identification
- In silico mutation screening
- Combination variant design
- Developability impact assessment
- Testing strategy
Output: 5-10 affinity variants ranked by predicted improvement
Timeline: ~20 minutes for full analysis
Query: "This antibody aggregates at >50 mg/mL. Fix it."
Process:
- Aggregation-prone region (APR) identification
- Hydrophobic patch analysis
- Charge distribution assessment
- Mutation design (disrupt APRs)
- Formulation recommendations
- Validation plan
Output: 3-5 aggregation-mitigated variants with predicted max concentration
Timeline: ~25 minutes for full analysis
Query: "Take this mouse antibody to clinical candidate"
Process (all phases):
- Humanization (Phase 2)
- Structure modeling (Phase 3)
- Affinity optimization (Phase 4)
- Developability assessment (Phase 5)
- Immunogenicity prediction (Phase 6)
- Manufacturing assessment (Phase 7)
Output: Comprehensive 1000+ line report with top candidate recommendation
Timeline: ~45-60 minutes for full pipeline
| Tool Category | Tools Used | Purpose |
|---|---|---|
| Immunogenetics | IMGT_search_genes, IMGT_get_sequence | Germline identification |
| Antibody Databases | SAbDab, TheraSAbDab | Structural & clinical precedents |
| Structure | AlphaFold_get_prediction | Structure modeling |
| Immunogenicity | iedb_search_epitopes, iedb_search_bcell | Epitope prediction |
| Target Info | UniProt_get_protein_by_accession | Target characterization |
| Literature | PubMed_search | Clinical precedents |
- STRING (for bispecifics - protein interaction networks)
- EMDB (if target has cryo-EM structures)
- PDB (for experimental structures)
- CDR Preservation: Warns if humanization changes CDR sequences
- Vernier Zone: Identifies residues affecting CDR conformation
- PTM Sites: Flags deamidation (NG, NS), isomerization (DG, DS), oxidation (M, W)
- Aggregation: TANGO scores, hydrophobic patches
- pI Range: Checks for extreme values (<4 or >10)
- Canonical Classes: Validates CDR conformations
| Tier | Criteria | Recommendation |
|---|---|---|
| T1 | Humanness >85%, Dev score >75, Low immunogenicity | Advance to development |
| T2 | Humanness 70-85%, Dev score 60-75, Medium immunogenicity | Acceptable, monitoring needed |
| T3 | Humanness <70%, Dev score <60, or High immunogenicity | Requires optimization |
| T4 | Failed validation or major liabilities | Do not advance |
Provide as much context as possible:
- Current affinity (if known)
- Known issues (aggregation, immunogenicity, stability)
- Development stage (discovery, preclinical, clinical)
- Target indication and therapeutic modality
The skill provides computational predictions. Always:
- Validate top candidates experimentally
- Test multiple variants (don't rely on single prediction)
- Consider backup options
- Monitor for unexpected issues
Use results to guide next steps:
- If humanization reduces affinity → Test backmutations
- If aggregation persists → Try alternative formulations
- If immunogenicity high → Apply deimmunization strategy
Optimal candidate balances multiple factors:
- Affinity (target: <10 nM for most applications)
- Humanization (target: >85% for low immunogenicity)
- Developability (target: >75 for clinical success)
- Manufacturing (target: Expression >1 g/L, formulation >100 mg/mL)
- Affinity predictions: ±2-3x accuracy
- Structure predictions: CDR-H3 may have lower confidence
- Aggregation: In silico scores are estimates (require validation)
Always validate computationally designed variants:
- Binding affinity (SPR, BLI)
- Expression and stability
- Functional activity
- In vivo PK (for final candidate)
- Primarily optimized for human therapeutics
- Mouse → Human humanization most common
- Other species may require custom germline databases
- Bispecifics require additional manufacturing development
- Non-IgG formats (scFv, Fab) have different considerations
- Antibody-drug conjugates (ADCs) need separate conjugation site engineering
Solutions:
- Try alternative germline frameworks
- Accept lower score if affinity critical (clinical monitoring)
- Consider fully human antibody (phage display, transgenic mice)
Solutions:
- Introduce backmutations at Vernier zone positions
- Test multiple humanization versions
- Combine humanization with affinity maturation
Solutions:
- Apply aggregation mitigation mutations (disrupt APRs)
- Optimize formulation (pH, excipients, concentration)
- Consider alternative framework (lower pI)
Solutions:
- Apply deimmunization (remove T-cell epitopes)
- Increase humanization level
- Consider fully human framework
- IMGT: http://www.imgt.org/ (germline genes, numbering)
- SAbDab: http://opig.stats.ox.ac.uk/webapps/sabdab-sabpred/ (antibody structures)
- TheraSAbDab: http://opig.stats.ox.ac.uk/webapps/newsabdab/ (therapeutic antibodies)
- IEDB: http://www.iedb.org/ (epitope database)
- Antibody humanization: Almagro & Fransson, Front. Biosci. 2008
- CDR grafting: Queen et al., PNAS 1989
- Developability: Jain et al., mAbs 2017
- pH-dependent binding: Igawa et al., Nat. Biotechnol. 2010
- Use TheraSAbDab to search approved antibodies
-
100 approved therapeutic antibodies as of 2026
- Most use IGHV1-69, IGHV3-23, IGKV1-39 germlines
"Humanize this mouse anti-PD-L1 antibody"
"Improve affinity of this antibody to <5 nM"
"Assess developability of this sequence"
"Reduce aggregation in this antibody"
"Predict immunogenicity risk"
"Humanize this mouse antibody, optimize affinity, and assess developability"
"Design a bispecific antibody targeting PD-L1 and TIM-3"
"Engineer pH-dependent binding into this anti-HER2 antibody for improved PK"
"Take this mouse antibody through complete optimization to clinical candidate"
"Humanize to >85% while maintaining affinity within 2x"
"Reduce aggregation to enable >150 mg/mL formulation"
"Improve affinity from 15 nM to <5 nM without compromising developability"
"Deimmunize this humanized antibody to reduce T-cell epitopes"
For issues, questions, or feature requests:
- Check EXAMPLES.md for detailed use cases
- Review SKILL.md for complete workflow documentation
- Consult ToolUniverse documentation for tool-specific questions
- v1.0 (2026-02-09): Initial release
- Humanization pipeline
- Affinity maturation
- Developability assessment
- Immunogenicity prediction
- Manufacturing assessment
- Bispecific antibody design
- pH-dependent binding engineering
This skill is part of the ToolUniverse project. See main repository for license information.