Graylab/IgFold

What it solves

IgFold provides a fast and accurate way to predict the 3D structures of antibodies from their amino acid sequences. This is critical for understanding how antibodies interact with targets, which is essential for drug discovery and biological research.

How it works

IgFold uses deep learning trained on a massive dataset of natural antibodies. It leverages a pre-trained language model called AntiBERTy to process antibody sequences. The system can predict the backbone and CB atoms of antibody variable domains (Fv) and can further refine these predictions into full-atom structures using tools like OpenMM or PyRosetta. It also supports using template structures to guide the prediction of mutants or specific variants.

Who it’s for

It is designed for structural biologists, biochemists, and pharmaceutical researchers who need to model antibody structures quickly and accurately without relying solely on expensive experimental methods.

Highlights

• Fast Prediction: Capable of predicting structures in seconds on a CPU.
• Flexible Input: Supports paired Fv, nanobodies, or single chains via FASTA files or command-line arguments.
• Full-Atom Refinement: Integrates with OpenMM and PyRosetta to build and relax full-atom structures.
• Template Support: Allows the use of existing PDB structures as templates to guide predictions, with the ability to ignore specific CDR regions.
• Standardized Output: Generates structures in PDB or mmCIF formats with predicted RMSD stored in the B-factor column.