[Paper Review] HelixFold-Multimer: Elevating Protein Complex Structure Prediction to New Heights

Xiaomin Fang, Jie Gao+5 more|arXiv (Cornell University)|Apr 16, 2024

Enzyme Structure and Function6 citations

TL;DR

HelixFold-Multimer significantly improves protein complex structure prediction, especially antigen-antibody, nanobody-antigen, and peptide-protein interfaces, outperforming AlphaFold-Multimer in several scenarios and is publicly available on PaddleHelix.

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.

Motivation & Objective

• Enhance accuracy of multi-chain protein complex structure predictions beyond monomer-focused models.
• Improve cross-chain interaction modeling by integrating domain expertise into architecture, features, and training.
• Enable reliable predictions for therapeutic protein interactions, including antigen-antibody and peptide-protein complexes.
• Provide public, usability-focused versions (general and antigen-antibody) on the PaddleHelix platform.

Proposed method

• Develop two versions of HelixFold-Multimer (general and antigen-antibody) building on HelixFold and HelixFold-Single foundations.
• Benchmark against AlphaFold and RoseTTAFold using DockQ as the primary metric.
• Use pLDDT, PTM, iPTM, and a confidence score to assess model reliability and correlate with DockQ.
• Evaluate on curated datasets: heteromeric protein complexes, protein-peptide complexes, antigen-antibody, and nanobody-antigen interfaces.
• Analyze performance across species and antigen sequence identity to training data to understand generalization.

Experimental results

Research questions

• RQ1 Can HelixFold-Multimer surpass AlphaFold-Multimer in predicting heteromeric protein complexes and peptide-protein interfaces?
• RQ2 How does HelixFold-Multimer perform on antigen-antibody and nanobody-antigen interfaces compared to baselines?
• RQ3 What is the relationship between model confidence metrics (CFS, iPTM, pLDDT) and DockQ accuracy for antibody-related predictions?
• RQ4 Does antigen species origin and sequence identity to training data affect predictive accuracy?

Key findings

• Median DockQ for heteromeric complexes with HelixFold-Multimer is 0.304, comparable to AlphaFold (0.316).
• HelixFold-Multimer achieves 57.8% accuracy (DockQ > 0.23) on heteromeric complexes versus AlphaFold’s 53.6%.
• For protein-peptide docking, HelixFold-Multimer has a median DockQ of 0.295 and a DockQ > 0.23 success rate of 68.9%, outperforming AlphaFold (0.262 median, 54.1% success).
• Antibody-antigen interfaces: HelixFold-Multimer mean DockQ 0.390 with 52.7% DockQ > 0.23 versus AlphaFold 0.195 mean and RoseTTAFold 0.? (lower).
• Nanobody-antigen interface: median DockQ 0.703 and mean 0.538 with 69.2% DockQ > 0.23, outperforming AlphaFold and RoseTTAFold.
• VH-VL antibody interfaces: HelixFold-Multimer median DockQ 0.823 vs AlphaFold 0.774 and RoseTTAFold 0.653; 59.5% DockQ > 0.8 vs 37.4% (AlphaFold) and 19.1% (RoseTTAFold).
• Confidence metrics (CFS and iPTM) correlate strongly with DockQ, indicating reliable confidence guidance for antibody design.