Abstract:Nanobodies have emerged as a versatile class of biologics with promising therapeutic applications, driving the need for robust tools to predict their epitopes, a critical step for in silico affinity maturation and epitope-targeted design. While molecular docking has long been employed for epitope identification, it requires substantial expertise. With the advent of AI driven tools, epitope identification has become more accessible to a broader community increasing the risk of models misinterpretation. In this study, we critically evaluate the nanobody epitope prediction performance of two leading models: AlphaFold3 and AlphaFold2-Multimer (v.2.3.2), highlighting their strengths and limitations. Our analysis revealed that the overall success rate remains below 50% for both tools, with AlphaFold3 achieving a modest overall improvement. Interestingly, a significant improvement in the performance of AlphaFold3 was observed within a specific nanobody class. To address this discrepancy, we explored factors influencing epitope identification, demonstrating that accuracy heavily depends on CDR3 characteristics, such as its 3D spatial conformation and length, which drive binding interactions with the antigen. Additionally, we assessed the robustness of the confidence metrics of AlphaFold3, highlighting their potential for broader applications. Finally, we evaluated different strategies aimed at improving prediction success rate. This study can be extended to assess the accuracy of emerging deep learning models adopting a similar approach to AlphaFold3.

What problem does this paper attempt to address?

The problem that this paper attempts to solve is the accuracy of epitope prediction for Nanobodies. Specifically, the authors evaluated the performance of two leading AI - driven models - AlphaFold3 (AF3) and AlphaFold2 - Multimer (AF2 - M) in predicting Nanobody epitopes and explored the key factors affecting prediction accuracy. ### Problem Background Nanobodies are a class of biopharmaceuticals with broad application prospects. Due to their unique structural and functional characteristics, they show great potential in the fields of treatment and diagnosis. To accelerate the design and optimization of Nanobodies, reliable computational tools need to be developed to predict their epitopes, which is a crucial step for computational affinity maturation and epitope - directed design. Although traditional molecular docking methods can be used for epitope recognition, they require a great deal of professional knowledge and high - quality structural templates, which limits their wide application. With the development of AI - driven tools, epitope prediction has become more popular, but it also increases the risk of model misuse. Therefore, evaluating the performance of these new tools has become particularly important. ### Research Objectives 1. **Evaluate the performance of existing models**: By comparing the performance of AlphaFold3 and AlphaFold2 - Multimer on 70 unique Nanobody/antigen complexes, evaluate their accuracy and limitations in epitope prediction. 2. **Analyze factors affecting prediction accuracy**: Explore the influence of characteristics such as the length and conformation of CDR3 (complementary - determining region 3) on the accuracy of epitope prediction. 3. **Propose improvement suggestions**: Based on the analysis results, propose strategies and suggestions for improving the success rate of epitope prediction. ### Main Findings - **Overall success rate is less than 50%**: The study found that although AlphaFold3 performs well in some cases, the overall success rate is 47.1%, while the success rate of AlphaFold2 - Multimer is 32.8%. This indicates that epitope prediction remains a challenge. - **Importance of CDR3 characteristics**: The study shows that the accuracy of epitope prediction is closely related to the length and conformation of CDR3. Shorter CDR3 loops usually achieve higher prediction quality, while longer CDR3 loops are more difficult to predict due to their greater conformational flexibility. - **Influence of CDR3 conformation**: CDR3 loops with a kinked conformation are more easily and accurately predicted than those with a stretched conformation, which is related to the shape complementarity of CDR3 loops. ### Conclusions This study provides important insights into understanding the current situation of Nanobody epitope prediction and points out the limitations of current AI - driven tools. Future research can further improve the accuracy of epitope prediction by improving the model architecture, introducing more training data, and combining other technical means. ### Key Formulas Some formulas involved in the discussion are as follows: - **DockQ score**: A comprehensive index for evaluating the prediction quality of complexes, which takes into account the interface root - mean - square deviation (iRMSD), ligand root - mean - square deviation (LRMSD), and the proportion of correctly predicted reference contacts (fnat). The formula is as follows: \[ \text{DockQ}=\frac{1}{3}(\text{iRMSD}+\text{LRMSD}+\text{fnat}) \] Through these evaluation indicators, researchers can more comprehensively understand the performance of the model and find directions for improvement.

AI driven approaches in Nanobody Epitope Prediction: Are We There Yet?

AI driven approaches in Nanobody Epitope Prediction: Are We There Yet?

Characterization of a nanobody-epitope tag interaction and its application for receptor engineering

Integrating Dynamic Network Analysis with AI for Enhanced Epitope Prediction in PD-L1:Affibody Interactions

AI-Augmented Physics-Based Docking for Antibody-Antigen Complex Prediction

Improving B-cell Epitope Prediction and Its Application to Global Antibody-Antigen Docking.

Nanobody–antigen interaction prediction with ensemble deep learning and prompt-based protein language models

Using a Machine-Learning Approach to Predict Discontinuous Antibody-Specific B-Cell Epitopes

Identifying B-cell epitopes using AlphaFold2 predicted structures and pretrained language model

Structural Modeling of Nanobodies: A Benchmark of State-of-the-Art Artificial Intelligence Programs

A Comparison of Antibody-Antigen Complex Sequence-to-Structure Prediction Methods and their Systematic Biases

What has AlphaFold3 learned about antibody and nanobody docking, and what remains unsolved?

epitope3D: a machine learning method for conformational B-cell epitope prediction

A comparison of antibody–antigen complex sequence‐to‐structure prediction methods and their systematic biases

Integrating artificial intelligence-based epitope prediction in a SARS-CoV-2 antibody discovery pipeline: caution is warranted

Recent Progress in Antibody Epitope Prediction

Three machine learning approaches in the epitope prediction

AsEP: Benchmarking Deep Learning Methods for Antibody-specific Epitope Prediction

Improved deep learning prediction of antigen-antibody interactions

NbX: Machine Learning-Guided Re-Ranking of Nanobody–Antigen Binding Poses

Prediction of Paratope–Epitope Pairs Using Convolutional Neural Networks