Loading skill
Install any skill in seconds. Free to start, no credit card required.
Get Started Free →AI-powered molecular glue discovery for targeted protein degradation, enabling neo-substrate recruitment and undruggable target degradation through E3 ligase interface modulation.
.claude/skills/molecular-glue-discovery-agent/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-07 | ✗→✓ | ▲ Improved | 2% | 0% |
| case-20 | ✗→✓ | ▲ Improved | 10% | 0% |
| case-21 | ✗→✓ | ▲ Improved | 20% | 0% |
| case-22 | ✗→✓ | ▲ Improved | 40% | 0% |
| case-06 | ✓→✓ | ＝ Same ✓ | 34% | 0% |
<!--
#
#
-->
The Molecular Glue Discovery Agent enables AI-driven discovery of molecular glue degraders that induce protein-protein interactions between E3 ligases and neo-substrates for targeted protein degradation. Unlike PROTACs, molecular glues are smaller, more drug-like molecules that can access previously "undruggable" targets through induced proximity mechanisms.
| Class | E3 Ligase | Mechanism | Example | |-------|-----------|-----------|---------| | IMiDs | CRBN | Degron recognition | Lenalidomide | | CELMoDs | CRBN | Enhanced IKZF binding | Iberdomide | | DCAF15 Glues | DCAF15 | Splicing factor degradation | Indisulam | | CDK12 Glues | DDB1-CRBN | Cyclin K degradation | CR8 derivatives | | β-catenin Glues | Novel | WNT pathway targets | Emerging |
| Substrate | Glue Class | Disease Relevance | |-----------|------------|-------------------| | IKZF1/3 | IMiDs | Multiple myeloma | | CK1α | Lenalidomide | MDS del(5q) | | GSPT1 | CC-885 | AML | | RBM39 | Indisulam | Solid tumors | | Cyclin K | CR8 | Cancer | | SALL4 | Thalidomide | Teratogenicity |
User: "Discover molecular glues that degrade IKZF1 through CRBN with improved selectivity over IKZF3."
Agent Action:
bashpython3 Skills/Drug_Discovery/Molecular_Glue_Discovery_Agent/discover_glue.py \ --target_substrate IKZF1 \ --e3_ligase CRBN \ --selectivity_against IKZF3 \ --scaffold_library imid_derivatives.sdf \ --interface_model crbn_ikzf1_complex.pdb \ --n_candidates 100 \ --output glue_discovery/
| Parameter | Consideration | Optimization | |-----------|---------------|--------------| | Interface Complementarity | E3-substrate fit | Shape/electrostatics | | Degron Recognition | Substrate degron motifs | Motif compatibility | | Binding Cooperativity | Positive cooperativity | Enhanced ternary | | Selectivity | Off-target degradation | Substrate specificity | | Drug Properties | MW, solubility, permeability | Standard optimization |
| Output | Description | Format | |--------|-------------|--------| | Glue Candidates | Ranked molecules | .sdf, SMILES | | Predicted Substrates | Neo-substrate profiles | .csv | | Interface Models | Ternary complex structures | .pdb | | Selectivity Scores | On-target vs off-target | .csv | | Degradation Predictions | DC50, Dmax estimates | .csv | | SAR Analysis | Structure-activity trends | .json |
Interface Prediction:
Neo-Substrate Discovery:
Glue Optimization:
| Feature | Molecular Glue | PROTAC | |---------|----------------|--------| | Molecular Weight | <500 Da | 700-1500 Da | | Target Discovery | Serendipitous/AI | Rational | | Selectivity | Can be exquisite | Often broader | | Substrate Range | Induced neo-substrates | Direct binders | | Oral Bioavailability | Generally better | Challenging |
| Drug | Mechanism | Target | Phase | |------|-----------|--------|-------| | Iberdomide (CC-220) | CELMoD | IKZF1/3, Aiolos | Phase 3 | | Mezigdomide (CC-92480) | CELMoD | IKZF1/3 | Phase 3 | | Golcadomide (CC-99282) | CELMoD | IKZF1/3 | Phase 2 | | CFT7455 | IKZF1/3 | IKZF1/3 | Phase 1 |
| Degron Type | Sequence Features | E3 Recognition | |-------------|-------------------|----------------| | Zinc Finger | C2H2 ZF domain | CRBN-IMiD | | Phosphodegron | pSer/pThr motifs | SCF E3s | | N-degron | N-terminal residues | UBR1/2 | | Hydrophobic | Exposed hydrophobics | Quality control |
| Strategy | Approach | Success Examples | |----------|----------|------------------| | Phenotypic Screening | Degradation readout | IMiDs, indisulam | | Target-Based | E3-substrate docking | Rational glues | | Chemoproteomics | Pull-down identification | Neo-substrate discovery | | AI-Guided | Computational prediction | Emerging |
| Metric | Purpose | Threshold | |--------|---------|-----------| | Interface Score | Complex stability | >0.6 | | Cooperativity | Enhanced binding | >1.5 | | Selectivity Index | On/off-target ratio | >10 | | Drug-likeness | Developability | Lipinski compliant |
| Direction | Status | Potential | |-----------|--------|-----------| | New E3 Ligases | Active research | Expanded target space | | Protein-Protein Glues | Emerging | Beyond degradation | | AI-First Discovery | Advancing | Reduced serendipity | | Combination Glues | Conceptual | Multi-target degradation |
AI Group - Biomedical AI Platform
<!-- AUTHOR_SIGNATURE: 9a7f3c2e-MD-BABU-MIA-2026-MSSM-SECURE -->
| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-06 | pass→pass | 14,573 | 7,488 | -49% | 1 | 1 | 0% | 2,403 | 3,230 | +34% | 0 | 0 | — |
case-07 | fail→pass | 18,379 | 7,401 | -60% | 1 | 1 | 0% | 3,257 | 3,307 | +2% | 0 | 0 | — |
case-08 | pass→pass | 4,563 | 3,529 | -23% | 1 | 1 | 0% | 876 | 2,540 | +190% | 0 | 0 | — |
case-09 | fail→fail | 5,553 | 5,404 | -3% | 1 | 1 | 0% | 911 | 2,918 | +220% | 0 | 0 | — |
case-01 | fail→fail | 25,663 | 6,034 | -76% | 1 | 1 | 0% | 5,241 | 2,349 | -55% | 0 | 0 | — |
case-02 | fail→fail | 14,781 | 29,785 | +102% | 1 | 1 | 0% | 2,567 | 8,125 | +217% | 0 | 0 | — |
case-03 | fail→fail | 21,755 | 24,651 | +13% | 1 | 1 | 0% | 3,877 | 6,878 | +77% | 0 | 0 | — |
case-04 | pass→pass | 20,974 | 14,785 | -30% | 1 | 1 | 0% | 3,570 | 4,326 | +21% | 0 | 0 | — |
case-05 | pass→pass | 13,825 | 13,659 | -1% | 1 | 1 | 0% | 2,422 | 4,231 | +75% | 0 | 0 | — |
case-10 | pass→pass | 10,269 | 5,642 | -45% | 1 | 1 | 0% | 2,087 | 2,986 | +43% | 0 | 0 | — |
case-11 | pass→pass | 18,575 | 12,488 | -33% | 1 | 1 | 0% | 3,069 | 4,014 | +31% | 0 | 0 | — |
case-12 | pass→pass | 4,528 | 3,912 | -14% | 1 | 1 | 0% | 653 | 2,529 | +287% | 0 | 0 | — |
case-13 | pass→pass | 13,928 | 9,197 | -34% | 1 | 1 | 0% | 2,499 | 3,452 | +38% | 0 | 0 | — |
case-14 | pass→pass | 20,263 | 18,833 | -7% | 1 | 1 | 0% | 3,469 | 5,017 | +45% | 0 | 0 | — |
case-15 | pass→pass | 6,892 | 4,766 | -31% | 1 | 1 | 0% | 1,377 | 2,798 | +103% | 0 | 0 | — |
case-16 | pass→pass | 3,596 | 7,942 | +121% | 1 | 1 | 0% | 616 | 3,284 | +433% | 0 | 0 | — |
case-17 | pass→pass | 12,054 | 2,307 | -81% | 1 | 1 | 0% | 1,979 | 2,312 | +17% | 0 | 0 | — |
case-18 | fail→fail | 14,263 | 10,305 | -28% | 1 | 1 | 0% | 2,511 | 3,750 | +49% | 0 | 0 | — |
case-19 | pass→pass | 12,231 | 8,595 | -30% | 1 | 1 | 0% | 1,974 | 3,355 | +70% | 0 | 0 | — |
case-20 | fail→pass | 10,340 | 1,978 | -81% | 1 | 1 | 0% | 2,069 | 2,273 | +10% | 0 | 0 | — |
case-21 | fail→pass | 10,245 | 1,854 | -82% | 1 | 1 | 0% | 1,805 | 2,171 | +20% | 0 | 0 | — |
case-22 | fail→pass | 10,823 | 3,081 | -72% | 1 | 1 | 0% | 1,734 | 2,422 | +40% | 0 | 0 | — |
DecimalAI ran this skill against gemini-3.6-flash twice over the same eval suite — once with the skill loaded and once without — and compared the two runs case by case. 22 cases were attempted, and 21 counted toward the lift figure. The other 1 produced results that are not comparable between the two arms, so they are excluded from the headline rather than averaged into it. The headline lift of +18 percentage points is the difference between those two pass rates over the 21 comparable cases.
Without the skill loaded, the model failed this case. With it loaded, the same prompt on the same model passed. This is one improved case from the latest verified run; every case, including any that regressed, is in the table above.
| Model | Method | Date | Lift |
|---|---|---|---|
| gemini-3.6-flash | verified | 7/26/2026 | +32% |
Other measured skills in the registry, with their headline benchmark lift.