Skip to content

RFdiffusion2 on the cluster ​

RFdiffusion2 is a protein generation and design tool from the Baker laboratory for tasks including backbone design, binding-site design, and small-molecule binder design. The CHESS application accepts JSON inputs and schedules resources automatically.

Prepare inputs ​

Sign in to CHESS:

Interface illustration

Suggested JSON and output paths (用户名 means your username):

bash
/hpcfs/fhome/用户名/rfdiffusion2/input/open_source_demo.json
bash
/hpcfs/fhome/用户名/rfdiffusion2/output/

Example JSON fragment; adapt the task, input PDB, ligands, and contig settings:

json
{
  "active_site_indexed_atomic": "inference.input_pdb=mcsa_41/M0584_1ldm.pdb inference.ligand='NAD,OXM' contigmap.contigs=['46,A106-106,59,A166-166,2,A169-169,23,A193-193,46'] inference.contig_as_guidepost=False",
  "small_molecule_binder_rasa_buried": "inference.conditions.relative_sasa_v2.active=True inference.conditions.relative_sasa_v2.rasa=0 contigmap.contigs=['150'] inference.input_pdb=trimmed_ec2_M0151_NO_ORI_zero_com0.pdb contigmap.length=150-150 inference.ligand=PH2"
}

Submit through CHESS ​

Find RFdiffusion2 in Application Center (应用中心) and add it to the desktop:

Interface illustration

Interface illustration

Open it, select the JSON input, and set the output directory:

Interface illustration

Use a valid JSON file that matches the application requirements. Update PDB, ligand, and output references for your task. Ensure write permission and use separate outputs for different batches or templates.

View status and progress in Job Management (作业管理):

Interface illustration

Open job details for standard output, errors, and logs:

Interface illustration

Demonstration:

Interface illustration

Results and troubleshooting ​

The output directory may contain generated structures, logs, scoring/filtering files, and per-input or per-template subdirectories. Inspect logs, JSON parameters, paths, and permissions if results are absent.

GPU jobs can wait when resources are busy. Input paths must identify readable files on the cluster. Output directories must be writable; shared access requires administrator configuration. Start with a platform example, then adapt parameters one at a time to your structure, ligands, and design goals.

Copyright 2021 Biodesign Center. All rights reserved.