Streamline Protein Design with APM, Now Available on Vecura
This update allows structural biologists and protein engineers to perform sophisticated all-atom protein complex design and folding directly within Vecura, eliminating the need to manage complex GPU-intensive infrastructure.
What is APM?
APM (All-Atom Generative Model for Designing Protein Complexes) is an advanced flow-based generative model developed by ByteDance Research and published at ICML 2025. Unlike conventional models that separate sequence and structure or operate only on coarse backbones, APM unifies all-atom representation, side-chain modeling, and inter-chain interactions into a single framework. It helps researchers perform de novo protein-complex design, forward folding, inverse folding, and conditional peptide-binder generation through a unified, high-performance pipeline.
What can users do with APM on Vecura?
With APM on Vecura, users can:
- Design novel protein scaffolds: Create single-chain or multi-chain protein structures from scratch based on specified lengths or chain configurations.
- Predict structures from sequences: Perform forward folding to generate high-resolution all-atom structural hypotheses from amino-acid sequences.
- Generate sequences for scaffolds: Utilize inverse folding to design sequences optimized for a given target 3D structure.
- Engineer peptide binders: Conditionally generate peptide ligands tailored to a receptor’s specific binding pocket.
What the output means
The output provides generated PDB structure files, optimized amino-acid sequences, and a detailed sample_table. This table contains essential metrics such as backbone TM-scores, pLDDT scores, and sequence-recovery rates, allowing users to rank and filter candidates effectively.
This output should be used to support scientific decision making. It does not replace experimental validation.
Why this matters
Computational protein design is fundamentally limited by the gap between sequence and structure generation, as well as the neglect of all-atom interactions crucial for binding. By modeling all atomic coordinates and inter-chain contacts simultaneously, APM overcomes traditional backbone-only limitations, providing a more biologically accurate approach to complex design.
This capability represents a significant shift toward end-to-end generative workflows in structural biology, streamlining the transition from design to initial validation. By removing the technical burden of managing complex inference pipelines, it empowers researchers to focus on hypothesis generation and downstream experimentation.
- Developed by: ByteDance Research
- Source: Official GitHub Repo
- Reference: ICML 2025 Paper
Try APM on Vecura.
Open the model workspace and start evaluating it with your own inputs.