Streamlining Protein Design: FAMPNN Now Available on Vecura

What is FAMPNN?

FAMPNN (Full-Atom MPNN) is an advanced graph-neural-network model designed to bridge the gap between protein sequence identity and sidechain geometry. Unlike traditional fixed-backbone design methods that often treat these elements separately, FAMPNN utilizes a dual-component probabilistic framework that couples discrete amino-acid selection with a continuous sidechain diffusion process. This approach ensures that designed sequences are not only compatible with the target backbone but also inherently pack correctly, allowing for the generation of immediately usable, full-atom protein structures.

What can users do with FAMPNN on Vecura?

With FAMPNN on Vecura, users can:

• Design novel protein sequences: Generate amino-acid sequences tailored to a specific backbone while simultaneously predicting the corresponding full-atom sidechain conformations.
• Optimize sidechain packing: Take existing backbone-only models and perform high-quality, diffusion-based sidechain packing without needing to redesign the sequence.
• Score mutation effects: Perform deep-mutational scans or evaluate specific point mutations using log-likelihood-ratio scores to assess their impact on protein fitness.
• Access detailed quality metrics: Benefit from Predicted Sidechain Error (PSCE) scores provided for every atom, allowing for immediate identification of structural regions that require further attention.

What the output means

The output provides comprehensive files including designed full-atom PDB structures, amino-acid sequences, and mutation likelihood scores. Specifically, the PDB files contain per-atom Predicted Sidechain Error (PSCE) values embedded in the B-factor column, providing a clear, machine-readable quality estimate for every residue.

This output should be used to support scientific decision making. It does not replace experimental validation.

Why this matters

The ability to simultaneously reason about sequence identity and sidechain conformation is a significant advancement in computational protein design. Traditional methods often produce sequences where sidechains may clash or fail to adopt expected rotamers on the target backbone, necessitating extensive post-hoc repacking. By explicitly modeling these interactions via diffusion, FAMPNN produces structures that are physically more consistent and ready for downstream validation or experimental synthesis.

This model is particularly powerful for researchers working in protein engineering and structural biology, as it streamlines the transition from a hypothetical backbone to a high-fidelity, full-atom design, effectively reducing the time and complexity required to reach a testable construct.

• Developed by: Protein Design Lab
• Source: GitHub (https://github.com/fnachon/fampnn)
• Reference: bioRxiv preprint