Skip to main content
    Ivy Biosciences logo
    Ivy Biosciences.bloom* docs
    login.bloom*
    Ivy Biosciences Docs/Tools/Docking & MD/Diffdock/DiffDock quickstart

    DiffDock quickstart

    Run DiffDock-L in six steps — pick a protein, add SMILES or an SDF, keep 10 poses and 20 diffusion steps, and download ranked pose files in 5-15 minutes.

    Open in Claude(opens in a new tab)Open in ChatGPT(opens in a new tab)4 min read

    You want to predict how a ligand sits in a protein's pocket. DiffDock does it in a few guided steps — set your protein target, add your ligand(s), and submit — no scoring function to tune, and no binding site to define (the platform always docks against the whole protein).

    Quick start with .bloom* platform

    DiffDock is a diffusion-based molecular docking method. The .bloom* platform uses DiffDock-L, the latest version with state-of-the-art success rates on standard benchmarks. Unlike traditional docking methods that rely on scoring function optimization, DiffDock generates multiple binding poses per ligand through a learned diffusion process over the space of ligand poses. The model is released under the MIT license.

    Step-by-step docking workflow

    Access the DiffDock tool

    Navigate to the Molecular Docking tool in .bloom*'s interactive workspace. The tool provides a guided interface for all input methods and validation.

    Set up your protein target

    Option A: PDB ID (fastest)

    • Click PDB ID in the protein input section
    • Type a PDB code (e.g., "6LU7" for SARS-CoV-2 main protease)
    • .bloom* automatically loads and previews the structure
    • Review the 3D preview to confirm the correct protein

    Option B: Upload custom structure

    • Click Upload PDB File in the protein input section
    • Select your PDB or mmCIF file (supports .gz compression)
    • .bloom* validates the format and loads a 3D preview
    • Use the viewer to inspect the structure before proceeding

    Option C: From previous job

    • Click From Previous Job in the protein input section
    • Select a completed job that produced protein structures (e.g., ESMFold results)
    • If the job output contains multiple files, choose the specific PDB file
    • .bloom* extracts and previews the selected structure

    Specify your ligands

    Option A: Direct SMILES entry

    • Click SMILES String in the ligand input section
    • Enter your SMILES string (e.g., "CC(=O)Oc1ccccc1C(=O)O" for aspirin)
    • For multiple ligands, enter each SMILES on a new line
    • .bloom* validates syntax and shows molecular previews

    Option B: Upload compound file

    • Click Upload SDF File or Upload SMILES/CSV File
    • Select your compound file from your computer
    • For CSV files, ensure SMILES are in the first column
    • .bloom* parses the file and shows the number of compounds detected

    Option C: From chemistry tools

    • Click From Previous Job in the ligand input section
    • Select outputs from AiZynthFinder, DeepPurpose, or other generative chemistry tools
    • Choose specific compounds or use all generated molecules
    • .bloom* handles format conversion automatically

    Configure job settings

    Basic settings (recommended for first runs):

    • Leave Job Name blank for automatic naming, or enter a descriptive name
    • Keep default parameters (10 poses, 20 diffusion steps)

    Advanced settings (via gear icon):

    • Increase Number of Poses (up to 40) for virtual screening
    • Adjust Diffusion Steps higher (up to 40) for better quality (slower)

    Submit and monitor the job

    • Click Run Molecular Docking to start the job
    • .bloom* shows real-time progress through diffusion phases
    • Typical jobs complete in 5-15 minutes depending on protein size
    • Receive notifications when complete (if enabled)

    Review and download results

    View results in .bloom:*

    • Poses load automatically in the 3D viewer
    • Compare different poses side-by-side
    • Check confidence scores and rankings
    • Export images or share results

    Download data:

    • Click Download to get results as ZIP archive
    • Contains: results.json, individual pose PDB/SDF files
    • Use for further analysis in external tools

    Common quick-start scenarios

    Virtual screening campaign

    1. Use PDB ID for protein target
    2. Upload CSV file with 100+ SMILES strings
    3. Set poses to 20-30 per compound
    4. Submit and let .bloom* process the library

    Hit optimization

    1. Upload protein structure from previous work
    2. Enter SMILES for compound series
    3. Set higher pose count (30) for diversity
    4. Compare poses across analogs

    Target validation

    1. Use UniProt ID to fetch protein sequence
    2. Enter known ligand SMILES
    3. Compare predicted vs. experimental binding modes

    Quality control checklist

    Protein preparation

    • Structure is complete (no missing backbone atoms)
    • Appropriate protonation state
    • Relevant chains only (remove unnecessary domains)

    Ligand preparation

    • Valid SMILES syntax (use external validators if unsure)
    • Reasonable molecular weight (<1000 Da for typical screening)
    • No unusual elements or query atoms
    • Correct formal charges and stereochemistry

    Job configuration

    • Appropriate number of poses for your use case
    • Realistic runtime expectations
    • Sufficient computational resources allocated

    Troubleshooting common issues

    Job fails immediately

    • Check protein file format (valid PDB/mmCIF)
    • Verify ligand SMILES syntax

    No poses generated

    • Check for protein structure issues (missing atoms, invalid chains)
    • Verify the ligand SMILES parses to a reasonable 3D conformer

    Poor pose quality

    • Increase diffusion steps (inference_steps, up to 40) in advanced settings
    • Consider protein preparation (protonation, cleaning)

    Slow performance

    • Reduce number of poses requested
    • Split large batches into smaller jobs

    Next steps after docking

    Pose validation:

    • Run molecular dynamics simulations
    • Calculate binding free energies (MM/GBSA)
    • Analyze protein-ligand interactions

    Compound optimization:

    • Use poses as input for generative chemistry (DeepPurpose)
    • Design focused libraries around best poses
    • Optimize synthetic accessibility (AiZynthFinder)

    Further analysis:

    • Export to visualization tools (PyMOL, Chimera)
    • Share results with collaborators
    • Integrate into larger drug discovery workflows

    DiffDock inputs and parameters

    Protein and ligand inputs, sampling limits, and no binding-site parameter.

    DiffDock on the platform

    Confidence-ranked poses from blind, diffusion-based docking with DiffDock-L.

    Was this page helpful?

    DiffDock inputs and parameters

    Protein by PDB ID, UniProt ID or upload; ligands as SMILES, SDF or CSV; samples_per_complex and inference_steps up to 40; no binding-site parameter at all.

    DiffDock outputs and ranking

    Ranked pose SDF and PDB files plus results.json with poseId, rank and score — why that score is pose accuracy within 2 Å, not affinity, and which to keep.

    On this page

    On this page

    Loading...