OpenDDE

Overview

Managed version: SubSeq currently pins OpenDDE 1.1.1 source and runtime dependencies. Numerical behavior may change after a future managed release upgrade.

• Protein, DNA, RNA, ligand, ion, and mixed-complex structure prediction.
• Typed assembly building for common systems.
• Native OpenDDE JSON for prepared or advanced inputs.
• Batch prediction from native JSON or SubSeq sequence artifacts.
• The general-purpose opendde.pt and antibody-antigen opendde_abag.pt checkpoints, verified against the OpenDDE 1.1.1 model manifest.

Workflows

Canonical Job Configuration

The current typed fields are returned by GET /api/v1/program/params?program=opendde and submitted as the params JSON object to POST /api/v1/job/submit.

Choose gpus as 1, 2, or 4 (default 1). With 2 or 4, the GPUs cooperate on each prediction to help larger complexes fit. The hourly rate is 2x or 4x the default catalog rate; total cost also depends on runtime. Speed and capacity depend on the input. SubSeq manages distributed launch and compatible kernels automatically.

Local Feature Policy

• SubSeq runs OpenDDE without hosted preprocessing services.
• Protein and RNA MSA are off unless a prepared input explicitly references a local A3M for every applicable entity.
• Template inference is disabled because the upstream path fetches template structures from a third-party service.
• Native JSON may reference prepared A3M files mounted under the selected input source.
• SubSeq seq JSON with a precomputed MSA keeps its relative MSA reference during conversion.
• When RNA MSA is enabled without protein MSA in a mixed system, SubSeq supplies query-only protein alignments so OpenDDE can activate RNA features without a hosted protein search.
• Do not submit paths outside the mounted input and service-managed reference trees.

Outputs And Results

/outputs/<job_name>/seed_<seed>/predictions/
  <job_name>_sample_<rank>.cif
  <job_name>_summary_confidence_sample_<rank>.json
  <job_name>_full_data_sample_<rank>.json  # when atom confidence is enabled

• Native mmCIF structures and confidence JSON remain available in the raw output manifest.
• Rank 0 is the highest-ranked sample for a job and seed.
• Final structures are also published as standardized SubSeq structure artifacts for preview and pipeline handoff.
• plddt is local confidence on a 0-100 scale; higher is better.
• gpde is predicted global distance error in angstroms; lower is better.
• ptm estimates global topology confidence; iptm focuses on interfaces. Higher is better.
• ranking_score orders samples; detected clashes can apply a large penalty.
• The result reader recognizes optional protein-binding ranking and van der Waals clash fields when OpenDDE emits them.

The Results view pairs each predicted structure with its recognized metrics. Use the raw manifest for full confidence matrices, per-chain values, and other detailed sidecars.

Common Examples

• Protein monomer: Build Assembly with one protein entity, the general checkpoint, one sample, and local feature searches off.
• Antibody-antigen complex: Build Assembly or Custom JSON with the antibody-antigen checkpoint.
• Protein-ligand complex: Build Assembly with protein and ligand entities; optionally enable training-free guidance.
• Pipeline prediction: place OpenDDE after a sequence-producing step and use Batch Input to consume canonical SubSeq seq JSON artifacts.

Minimal native OpenDDE JSON

[
  {
    "name": "tiny",
    "modelSeeds": [101],
    "sequences": [
      {
        "proteinChain": {
          "sequence": "ACDEFGHIK",
          "count": 1
        }
      }
    ]
  }
]

Caveats

• Future managed OpenDDE upgrades may change output fields, checkpoint behavior, or reproducibility.
• Each job is limited to five model seeds and 64 total entity instances; batch sources are limited to 64 jobs. There is no fixed residue/base limit; inputs must fit the available memory with the selected settings.
• Single-sequence prediction may be less accurate than prediction with an informative prepared MSA.
• High local confidence does not guarantee a correct interface, ligand pose, or biological assembly.
• Ranking and confidence values are computational prioritization signals, not measured affinity, activity, or experimental validation.

Advanced Submit

Advanced submit is available for direct OpenDDE arguments through POST /api/v1/job/submit-advanced. Prefer canonical configuration for normal jobs and batch pipelines.

curl -X POST https://subseq.bio/api/v1/job/submit-advanced \
  -H "Authorization: Bearer <api_key>" \
  -F program=opendde \
  -F 'args=--input=/inputs/input.json' \
  -F 'args=--model=general' \
  -F 'args=--gpus=1' \
  -F 'args=--samples=1' \
  -F 'args=--steps=200' \
  -F 'args=--cycles=10' \
  -F 'args=--use-msa=false' \
  -F 'args=--use-templates=false' \
  -F 'args=--use-rna-msa=false' \
  -F 'args=--atom-confidence=false' \
  -F 'args=--dtype=fp32' \
  -F 'args=--trimul-kernel=torch' \
  -F 'args=--triatt-kernel=torch' \
  -F 'folder=@input.json;filename=input.json'

Further Reading

• OpenDDE 1.1.1 release
• OpenDDE 1.1.1 inference instructions
• OpenDDE 1.1.1 input JSON format