PROTAC Ternary Pose Prediction

A PROTAC is a molecule that works by bringing two proteins together: a target protein you want to degrade and an E3 ligase that tags it for destruction. It does this with two binding ends, a warhead that grips the target and a recruiter that grips the E3 ligase, joined by a flexible linker. The three way arrangement of target, PROTAC, and E3 ligase is called the ternary complex.

This tool predicts the 3D pose of that ternary complex. You supply the target protein with its warhead bound, the E3 ligase with its recruiter bound, and the linker that joins the two ends. The tool searches over many ways the two proteins can be oriented relative to each other, minimizes the promising ones on a GPU, and scores them with a trained ranker to return the most plausible poses.

When to use it

Use it when you are designing or evaluating a PROTAC and want to know whether a given linker can actually hold the target and the E3 ligase in a productive arrangement. Comparing poses across different linkers helps you reason about which designs are geometrically feasible before synthesis. If you already have an experimental ternary structure, you can supply it and the tool reports how close its predictions come to the real arrangement.

Inputs

Input	Required	What it is
`target_pdb_content`	yes	The target protein with its warhead bound, uploaded as a PDB file.
`e3_pdb_content`	yes	The E3 ligase with its recruiter bound, uploaded as a PDB file.
`linker_smiles`	yes	The linker as a SMILES string with two attachment points, written `[:1]` and `[:2]`, marking where it joins the warhead and the recruiter.
`warhead_attach_atom_idx`	yes	Which atom of the warhead the linker attaches to, given as a number counting from 0.
`recruiter_attach_atom_idx`	yes	Which atom of the recruiter the linker attaches to, given as a number counting from 0.
`target_chain`	no	The chain identifier of the target protein in its PDB file, for example `A`, or several separated by commas. Defaults to `A`.
`e3_chain_label`	no	The chain identifier of the E3 ligase in its PDB file. Defaults to `B`.
`preset`	no	How fine the orientation search is. `quick` is a coarse, fast, lower accuracy pass for demos, `medium` (the default) is the production setting the scoring model was trained on, and `long` is the finest and slowest.
ground truth (optional)	no	If you have an experimental ternary structure, upload it as `gt_complex_pdb_content` and name its PROTAC residue with `gt_ligand_resname` (plus optional `gt_target_chain` and `gt_e3_chain`) to get an accuracy comparison.

There are a few further optional fields, such as warhead_smiles and e3_anchor_smiles for supplying clean reference structures of the two binding ends, and keep_dirs to control how much output is saved. Leave them at their defaults for a first run.

How to run it

Submit your two proteins and linker from Azulene Studio, the Python SDK, or the CLI. Local file paths are uploaded for you automatically. New here? The Get started page walks through installing, logging in, and running a ready made example first.

In Azulene Studio

Open PROTAC Ternary Pose Prediction from the tools list, then on the Inputs and Parameters step upload the target PDB and the E3 ligase PDB, enter the linker SMILES with its two attachment points, give the warhead and recruiter attachment atom numbers, leave the preset at medium, then Review and Submit.

From the Python SDK

from opal import jobs

result = jobs.submit(
    job_type="protac_pose_prediction",
    input_data={
        "target_pdb_content": "target_with_warhead.pdb",
        "e3_pdb_content": "e3_with_recruiter.pdb",
        "linker_smiles": "[*:1]CCOCCOCC[*:2]",
        "warhead_attach_atom_idx": 12,
        "recruiter_attach_atom_idx": 7,
        "preset": "medium",
    },
)

From the CLI

Pass the inputs as a JSON string. File paths are uploaded automatically.

opal jobs submit --job-type protac_pose_prediction \
  --input-data '{"target_pdb_content": "target_with_warhead.pdb", "e3_pdb_content": "e3_with_recruiter.pdb", "linker_smiles": "[*:1]CCOCCOCC[*:2]", "warhead_attach_atom_idx": 12, "recruiter_attach_atom_idx": 7, "preset": "medium"}'

Reading the result

You get back the predicted ternary complex poses, ranked best first, each a 3D arrangement of the target protein, the PROTAC, and the E3 ligase. The full set of poses, overlays, and a summary are saved as a downloadable ZIP you can retrieve from the result with opal jobs download.

Each pose carries a score from the trained ranker so you can tell the more plausible arrangements from the less plausible ones. When you supply an experimental ternary structure, the result also reports how far each prediction sits from the real one, measured as an RMSD (a single number for the average distance between matching atoms, where smaller means closer), reported both for the PROTAC alone and for the PROTAC together with its surrounding pocket.

Notes

The three required structural inputs are the target with its warhead, the E3 ligase with its recruiter, and the linker SMILES with its two attachment points, plus the two attachment atom numbers. Make sure the chain identifiers you give match the chains in your uploaded PDB files. Start with the medium preset, since it is the setting the scoring model was trained on; long recovers harder, more extended poses at a higher cost. The search and minimization run on a GPU.