HelixFold-Multimer: Elevating Protein Complex Structure Prediction to New Heights

TL;DR

HelixFold-Multimer tackles the persistent challenge of predicting protein complex structures, notably antigen-antibody and cross-species interfaces, by extending the HelixFold lineage with two specialized versions: a general model and an antigen-antibody model. It integrates domain-informed architectural and training strategies to improve cross-chain interaction modeling, achieving strong results on heteromeric and peptide-protein docking and substantial gains in antibody-antigen predictions, including VH-VL interface accuracy. Comparative benchmarks show that while AlphaFold-3 often attains high accuracy on general complexes, HelixFold-Multimer significantly outperforms baselines on antigen-antibody and peptide-protein tasks, underscoring its utility for therapeutic design. Public availability on PaddleHelix enables researchers to perform rapid inferential analyses of protein interactions, with experiments indicating that epitope specification and confidence metrics can further enhance predictive reliability, paving the way for accelerated drug development and immune-targeted design.

Abstract

While monomer protein structure prediction tools boast impressive accuracy, the prediction of protein complex structures remains a daunting challenge in the field. This challenge is particularly pronounced in scenarios involving complexes with protein chains from different species, such as antigen-antibody interactions, where accuracy often falls short. Limited by the accuracy of complex prediction, tasks based on precise protein-protein interaction analysis also face obstacles. In this report, we highlight the ongoing advancements of our protein complex structure prediction model, HelixFold-Multimer, underscoring its enhanced performance. HelixFold-Multimer provides precise predictions for diverse protein complex structures, especially in therapeutic protein interactions. Notably, HelixFold-Multimer achieves remarkable success in antigen-antibody and peptide-protein structure prediction, greatly surpassing AlphaFold 3. HelixFold-Multimer is now available for public use on the PaddleHelix platform, offering both a general version and an antigen-antibody version. Researchers can conveniently access and utilize this service for their development needs.

Figures (8)

• Figure 1: Comparison between AlphaFold 2.3, AlphaFold 3 and HelixFold-Multimer for heteromeric protein complexes. The box plots on the left show DockQ score distributions, while the bar graph on the right reflects the percentage of accuracy (DockQ $>$ 0.23).
• Figure 2: Comparison of AlphaFold 2.3, AlphaFold 3 and HelixFold-Multimer in protein-peptide docking accuracy. The box plots on the left show DockQ score distributions, while the bar graph on the right reflects the percentage of accuracy (DockQ $>$ 0.23).
• Figure 3: Overall comparison for antibody-related complex structure predictions. The left figure depicts the distribution of DockQ scores, serving as a metric for the quality of predicted structures. The right figure showcases the percentage of accurate predictions specifically concerning antibody-antigen interfaces. The data underscores a notable improvement in prediction accuracy with the HelixFold-Multimer model compared to RoseTTAFold, AlphaFold 2.3, and AlphaFold 3.
• Figure 4: Evaluation of antibody VH-VL interfaces: Box plots on the left illustrate the performance comparison among RoseTTAFold, AlphaFold 2.3, AlphaFold 3 and HelixFold-Multimer. The bar graph on the right quantifies the percentage of predictions with very high accuracy (DockQ > 0.8).
• Figure 5: Impact of epitope specification for antigen-antibody structure prediction.