ABodyBuilder3: improved and scalable antibody structure predictions

Area of Science:

• Computational Biology
• Structural Biology
• Immunoinformatics

Background:

• Antibody structure prediction is crucial for therapeutic development.
• Existing models face challenges in accurately modeling CDR loops.
• Scalable and accurate prediction tools are needed.

Purpose of the Study:

• To introduce ABodyBuilder3, an advanced antibody structure prediction model.
• To improve the accuracy of Complementarity-Determining Region (CDR) loop modeling.
• To enhance the estimation of prediction uncertainties.

Main Methods:

• Leveraging language model embeddings for enhanced prediction.
• Implementing refined relaxation strategies for structural refinement.
• Incorporating a predicted Local Distance Difference Test (pLDDT) for uncertainty quantification.

Main Results:

• Achieved state-of-the-art accuracy in CDR loop modeling.
• Demonstrated improved antibody structure prediction through advanced techniques.
• Integrated pLDDT for reliable uncertainty estimation in predicted structures.

Conclusions:

• ABodyBuilder3 represents a significant advancement in antibody structure prediction.
• The model offers improved accuracy and uncertainty assessment for antibody design.
• The software and model weights are publicly available for research use.