Load GCS Scaffold Folder¶
Creates a lightweight reference to a Google Cloud Storage (GCS) folder that contains pre-computed scaffold .pt files for fold conditioning. It does not download any data; instead, it forwards the GCS path so downstream GPU services can load the files directly. Intended for folders with pairs of *_ss.pt and *_adj.pt files generated by RFDiffusion tooling.
Usage¶
Use this node when you already have scaffold secondary structure and adjacency tensors stored in GCS and want to feed them into scaffold-guided protein design or fold-conditioning steps. Connect its output to nodes that accept SCAFFOLD_PT_FILES; those downstream nodes will read the files directly from GCS.
Inputs¶
| Field | Required | Type | Description | Example |
|---|---|---|---|---|
| gcs_folder_path | True | STRING | GCS folder path that contains scaffold .pt file pairs (e.g., *_ss.pt and *_adj.pt). Must be a valid GCS URI beginning with gs://. A trailing slash will be added automatically if missing. | gs://my-bucket/scaffolds/my_project_run/ |
Outputs¶
| Field | Type | Description | Example |
|---|---|---|---|
| scaffold_pt_files | SCAFFOLD_PT_FILES | A reference object pointing to the specified GCS folder. Downstream nodes use this reference to load the scaffold .pt files directly from GCS. | {'type': 'gcs_path', 'gcs_folder_path': 'gs://my-bucket/scaffolds/my_project_run/'} |
Important Notes¶
- The folder should contain pairs of files: scaffold_name_ss.pt (secondary structure) and scaffold_name_adj.pt (block adjacency).
- The node validates that the path starts with gs:// and normalizes it to end with a trailing slash.
- Files are not loaded into memory by this node; it only passes a reference to downstream services for efficient processing of large folders.
- This node does not list, validate, or verify file presence in the folder; validation occurs when downstream nodes attempt to read the files.
- Ensure your environment has permission to read the specified GCS path; permission issues will surface in downstream steps that perform the actual read.
Troubleshooting¶
- Invalid GCS path format error: Provide a path starting with gs:// and a valid bucket/folder (e.g., gs://bucket/path/).
- Downstream node reports missing _ss.pt or _adj.pt: Verify the folder contains correctly named file pairs for each scaffold.
- Permission denied when downstream reading files: Ensure appropriate IAM permissions for the service account or credentials used to access the GCS bucket.
- Downstream timeouts or slow reads: Large folders are supported, but network or permission issues can cause delays; verify GCS availability and connectivity.
- Unexpected folder interpretation: Ensure the provided path points to a folder (not a single file). A trailing slash is added, but the path must be a directory containing the .pt files.