Protein Design · Nanjing University / ByteDance
DPLM-2
Generates protein sequences and structures simultaneously using a unified diffusion process.
What DPLM-2 is
DPLM-2 is a diffusion protein language model developed by Nanjing University and ByteDance. It addresses the challenge of joint sequence and structure generation in protein design, where conventional pipelines often decouple structure generation from sequence inverse-folding.
By unifying both modalities within a single diffusion framework, DPLM-2 ensures that generated protein backbones and their corresponding sequences remain mutually consistent throughout the generation process. This co-design approach provides researchers with coordinated three-dimensional atomic coordinates and sequence proposals.
The model produces all-atom coordinates alongside co-designed sequence tables, supporting workflows where backbone conformation and residue identity must be designed in tandem.
How it works
DPLM-2 unifies protein sequence and structure generation into a single diffusion process. Rather than generating a backbone structure first and subsequently predicting compatible sequences, it co-designs both representations concurrently so that sequence and structure do not diverge.
Key facts
- Version
- 2.0
- Domain
- Protein design
- Tasks
- Sequence generation, structure generation
- Modalities
- Protein sequence, protein structure
- Memory requirement
- 24 GB GPU memory
What you can use it for
- Joint sequence-structure generation
- Generate coordinated all-atom protein structures and sequences from scratch given a specified sequence length.
- Motif-constrained protein design
- Design complementary sequences and scaffold structures around predefined functional motif constraints.
- Integrated backbone and sequence co-design
- Produce atomic coordinates in PDB format alongside matching sequence tables without relying on separate inverse-folding steps.
Run it here
No installation and no GPU of your own: fill in the inputs and it runs on our servers. Results stay in your browser unless you sign in.
Not runnable yet
This model isn't available to run right now.Run this model
DPLM-2
Strengths and limits
Good at
- Generates sequence and structure concurrently in a single diffusion process to prevent mismatches between fold and residue identity.
- Outputs complete all-atom coordinates in standard PDB format alongside co-designed sequence options.
- Supports motif-conditioned generation to design scaffolds around specific functional sites.
Watch out for
- Requires 24 GB of GPU memory to run.
- Input conditioning options are restricted to specified target lengths and motif sequences.
Inputs and outputs
You provide
Target lengthoptional
Motif constraintoptional
You get back
All-atom coordinates in PDB format, viewable and downloadable.
Co-designed sequences.
Licence
Commercial use permitted · Apache-2.0
The Apache-2.0 licence allows use in commercial research and products.
Questions
Can I use DPLM-2 commercially?
Commercial use permitted. The Apache-2.0 licence allows use in commercial research and products.
How do I run DPLM-2 without installing anything?
Use the form under “Run it here” on this page: fill in the inputs and press Run model. It runs on our servers on 1× A100 40GB, typically in 1.0min, and a run costs about 420 credits. Visitors without an account get 1,000 credits a month.
What inputs does DPLM-2 require?
DPLM-2 accepts a target sequence length as a number and an optional motif sequence constraint.
What formats are used for the output data?
The model outputs all-atom structural coordinates in standard PDB format and provides co-designed sequences in tabular format.
How does DPLM-2 differ from traditional inverse-folding pipelines?
Instead of generating a backbone structure first and then predicting a sequence, DPLM-2 designs the sequence and three-dimensional structure simultaneously within one diffusion model.
What hardware is necessary to run the model?
The model requires a GPU with at least 24 GB of memory.
Written from DPLM-2's own documentation. Statements citing a figure that does not appear there are removed before publication. Licence and running details come from our registry.