PROTAC troubleshooting

PROTAC troubleshooting

This page helps you diagnose failures across the PROTAC design pipeline — invalid warhead or E3 ligand SMILES, linker enumeration that returns too few candidates, ternary complex modeling errors, low feasibility scores, weak degradation predictions and hook-effect risk, and GPU job timeouts.

Invalid warhead or E3 ligand SMILES

Symptom: "Invalid SMILES" or "Failed to parse structure" error during linker enumeration.

Fixes:

• Validate your SMILES using a structure editor or RDKit before submission
• Check for unmatched parentheses, brackets, or ring closure digits
• Ensure stereochemistry notation is correct (especially for VHL ligands with hydroxyproline)
• Remove salts or counterions from the SMILES -- provide only the parent structure
• Maximum fragment size is 200 heavy atoms. If your warhead or E3 ligand exceeds this, simplify the structure

Linker enumeration returns zero or very few linkers

Symptom: Fewer linkers generated than requested.

Fixes:

• Widen the linker length range. If min_linker_length and max_linker_length are too close (e.g., both 6), there may not be enough chemical space to fill
• Increase sa_threshold to allow less synthetically accessible linkers (default 5.0, try 7.0)
• Increase max_rotatable_bonds to allow more flexible linkers
• Switch scaffold_type from a specific type to "diverse" for more options
• Check that the warhead and E3 ligand SMILES have accessible attachment points. If the exit vectors are buried in a ring system, linker attachment may fail

Ternary complex modeling fails

Symptom: "Complex modeling failed" or timeout error.

Fixes:

• Verify the PDB ID is correct and contains the expected protein chain
• If using an uploaded structure, ensure it is a valid PDB or mmCIF file with proper chain and residue numbering
• The E3 ligase structure should include the full binding domain. For CRBN, include DDB1; for VHL, include ElonginB and ElonginC
• Reduce num_conformers if the job times out (try 5 instead of 10)
• Increase clash_tolerance if the modeler reports excessive clashes (try 3.0 instead of 2.0)
• Check that the PROTAC SMILES matches the warhead and E3 ligand used in the structure files

All ternary conformers have low feasibility scores

Symptom: Feasibility scores below 0.4 for all conformers.

Fixes:

• The linker may be too short or too long for the distance between binding sites. Try a different linker length range
• The exit vectors of the warhead and E3 ligand may be geometrically incompatible. Try rotating the attachment point or using a different warhead binding pose
• Some POI-E3 ligase pairs have unfavorable protein-protein interfaces regardless of linker. This is a genuine biological result, not a tool error
• Try a different E3 ligase system (switch from CRBN to VHL or vice versa)

Degradation score shows high hook effect risk

Symptom: Hook effect risk reported as "high."

Fixes:

• The hook effect is driven by high binary binding affinities relative to ternary complex stability. This is a molecular property, not a tool error
• Reduce warhead affinity slightly (if possible) to favor ternary complex formation
• Increase linker rigidity to stabilize the ternary complex (use triazole or piperazine scaffolds)
• Increase cooperativity by optimizing the POI-E3 protein-protein interface (use ternary modeling to guide this)
• In practice, the hook effect primarily matters at high concentrations (> 1 uM). If your target DC50 is < 100 nM, the hook effect may not be clinically relevant

Degradation score reports DC50 > 10 uM

Symptom: Predicted DC50 in the micromolar range, grade D or F.

Fixes:

• Check the ternary complex feasibility first. Poor ternary complex geometry is the most common cause of weak degradation
• Ensure the warhead has sufficient binding affinity to the POI. A weak binder (Kd > 1 uM) will produce a weak degrader
• Try a different E3 ligase system. Some targets are better degraded via CRBN than VHL, or vice versa
• Optimize the linker: length, flexibility, and attachment point all affect degradation efficiency
• If ternary_job_id was not provided, supply it for more accurate predictions incorporating 3D geometry

Job timeout or GPU error

Symptom: "Job failed: timeout" or "GPU allocation failed."

Fixes:

• Linker enumeration with num_linkers > 100 may take over 2 minutes. Reduce the count or split into multiple runs
• Ternary modeling with num_conformers > 20 is computationally expensive. Use 10 conformers for initial screening
• If GPU allocation fails, retry after a few minutes. Peak usage can cause temporary queuing
• Very large protein structures (> 1000 residues) increase modeling time. Trim the PDB to include only the relevant domain

Getting help

If issues persist:

1. Check the tool error message for specific parameter suggestions
2. Validate SMILES strings independently using RDKit or a structure editor
3. Verify PDB structures are complete and contain the expected chains
4. Describe the issue to Ivy

Read next