ImmuneBuilder: Deep-Learning models for predicting the structures of immune proteins.

Area of Science:

• Structural biology
• Immunoinformatics
• Computational biophysics

Background:

• Immune receptor proteins are crucial for immune system function and hold significant potential as biotherapeutics.
• Understanding the precise structure of these proteins is essential for characterizing their antigen-binding capabilities.

Purpose of the Study:

• To introduce ImmuneBuilder, a novel deep learning framework for predicting the structures of antibodies, nanobodies, and T-cell receptors.
• To demonstrate the accuracy and speed advantages of ImmuneBuilder compared to existing state-of-the-art methods like AlphaFold2.

Main Methods:

• Development of specialized deep learning models: ABodyBuilder2, NanoBodyBuilder2, and TCRBuilder2.
• Training and validation on benchmark datasets of solved antibody, nanobody, and T-cell receptor structures.
• Utilizing ensemble predictions to provide residue-level error estimates.

Main Results:

• ImmuneBuilder achieves state-of-the-art accuracy in predicting immune receptor structures, outperforming AlphaFold2 in speed by over 100x.
• ABodyBuilder2 showed a 0.09Å improvement in CDR-H3 loop prediction accuracy over AlphaFold-Multimer for antibodies.
• NanoBodyBuilder2 and TCRBuilder2 also demonstrated significant accuracy improvements for nanobodies and T-cell receptors, respectively.

Conclusions:

• ImmuneBuilder offers a highly accurate and efficient solution for immune receptor structure prediction.
• The tool's speed and accuracy accelerate research in antibody engineering and immunotherapeutics.
• ImmuneBuilder is freely available via download and a webserver, along with precomputed structural models.