How to Use OpenDDE

OpenDDE is an open-source all-atom biomolecular foundation model for co-folding mixed molecular systems. It is designed for AlphaFold3-style structure prediction across proteins, DNA, RNA, ligands, and ions, making it useful when the biological question is a complete complex rather than an isolated protein chain.

On Neurosnap, researchers provide Input Sequences, optional Input Molecules, optional Ions, and OpenDDE runtime controls such as checkpoint, protein MSA usage, recycling, diffusion steps, sample count, precision, and seed. The service returns ranked mmCIF structures together with scalar confidence metrics and per-rank JSON sidecars for any detailed confidence arrays or matrices emitted by OpenDDE.

OpenDDE follows the modern all-atom co-folding paradigm: sequence and molecular inputs are represented together so folding, docking, and complex assembly are solved as one structure-generation problem. The preview release exposes a general-purpose checkpoint plus an antibody-antigen checkpoint, which lets researchers select the model variant that best matches broad complex prediction or immune-complex modeling.

On Neurosnap, template and RNA-MSA generation are disabled, so RNA inputs use OpenDDE's single-sequence representation without downloading the large RNA search databases. When Use MSA is enabled, Neurosnap generates protein alignments with its internal private MSA server and supplies the resulting A3M files directly to OpenDDE; users should expect longer runtimes while those alignments are generated. Diffusion Samples, Diffusion Steps, and Number Recycles control the tradeoff between runtime, refinement depth, and structural diversity.

Researchers should interpret OpenDDE results as ranked structural hypotheses. Confidence scores such as pLDDT, pTM, ipTM, ranking score, and any emitted error matrices help determine whether the complex is globally credible, whether chain placement is plausible, and which regions need caution before docking follow-up, mutagenesis planning, or experimental validation.

• Multimodal biomolecular input: OpenDDE accepts proteins, DNA, RNA, ligands, and ions in one prediction problem, which better matches real structural-biology and drug-discovery systems than protein-only folding.
• Checkpoint choice: The general-purpose and antibody-antigen checkpoint options let researchers align the model variant with either broad co-folding or immune-complex questions.
• Practical MSA-free path: The default Neurosnap workflow can run without MSA, template, or RNA-MSA preprocessing, reducing setup overhead for exploratory structure prediction.
• Ranked confidence outputs: Structures are returned with scalar scores and detailed per-rank JSON sidecars, supporting comparison across generated samples before downstream modeling or experiments.