ImmuneBuilder

ImmuneBuilder predicts three-dimensional (3D) structures of immune receptor proteins (antibodies, nanobodies, and T-cell receptors) to enable structural analysis and support biotherapeutic design.

Key Features:

• Specialized deep learning models: ABodyBuilder2 for antibodies, NanoBodyBuilder2 for nanobodies, and TCRBuilder2 for T-cell receptors.
• Accuracy and speed: ABodyBuilder2 predicts CDR-H3 loops with an RMSD of 2.81 Å, improving 0.09 Å over AlphaFold-Multimer and executing over 100× faster than AlphaFold2.
• Benchmark performance: On a benchmark set of 34 recently solved antibody structures, NanoBodyBuilder2 achieves an average CDR-H3 RMSD of 2.89 Å, improving 0.55 Å over AlphaFold2, with similar reported improvements for TCR predictions.
• Error estimation: Produces an ensemble of predicted structures and provides per-residue error estimates for the final models.

Scientific Applications:

• Structural insights: Enabling analysis of antigen-binding mechanisms by providing atomic-level models of antibodies, nanobodies, and TCRs.
• Biotherapeutic development: Supporting design and optimization of therapeutic antibodies, nanobodies, and T-cell receptors through accurate structural predictions.
• Rapid prototyping: Facilitating fast iteration of structural hypotheses due to prediction speed significantly exceeding that of AlphaFold2.

Methodology:

Specialized deep learning models generate ensembles of predicted structures with per-residue error estimates, and performance was benchmarked against AlphaFold2 and AlphaFold-Multimer.