Protein Design · Structure prediction

PLACER

(Work in Progress)

Generate an ensemble of protein-ligand poses and side-chain conformations. Given a protein pocket and ligand, estimates how the ligand and sidechains will arrange themselves.

Runs PLACER (Protein-Ligand Atomistic Conformational Ensemble Resolver), a graph network that denoises corrupted atomic coordinates back to plausible ones. Given a structure and approximate knowledge of the binding site, it samples an ensemble of ligand poses and side-chain conformations rather than a single answer, with a predicted uncertainty (prmsd) per sample.

Documentation ↗ Paper ↗ Source code ↗ License ↗

No jobs before this; it will run immediately.

Not logged in. Register an account to keep track of your jobs and results, and add presets and bookmarks. Email results to this address when complete:

Ready

Job name

Structure (PDB or mmCIF) Only mmCIF files from RCSB are correctly parsed; PDB is the safer choice otherwise.

Samples 50-100 is a good number for docking; fewer is fine for side-chain conformations alone.

Ligands to predict (optional) Comma-separated ligand IDs, as name3, name3-resno, or chain-name3-resno. Empty predicts every ligand present.

Ligands to hold fixed (optional) Comma-separated ligand IDs kept at their input coordinates instead of being predicted.

Crop-center residue (optional) A protein residue, as chain-resno. Required when the input has no ligand at all.

Exclude common crystallography additives Drops solvents and crystallography additives (AlphaFold 3's exclusion list) before predicting.

Predict every allowed ligand Scores every eligible ligand rather than only the ones named in predict_ligand.

Rank output by

Include the small molecule Off predicts side chains only (apo), ignoring any ligand in the input.

Ready