A new approach for structural clustering of antibodies and antigens

Introduction:The clustering of functional antibodies that recognize the same epitope is a strategy that reduces the sample space, optimizing data analysis pipelines and revealing relevant biological patterns.Sequence-and structure-based methods are widely used, but since distinct sequences can adopt similar conformations, structural approaches tend to be more promising.However, Waury et al. (2024) highlights the low sensitivity of current software in identifying functional pairs, which typically rely on atomic overlap (RMSD).We propose the comparison of protein surfaces using 3D Zernike descriptors, which geometrically represent proteins through specific mathematical functions.To the best of our knowledge, these descriptors have not yet been applied to the clustering of antibodies that bind to the same epitope.Objectives: This study's objective was to evaluate the application of 3D Zernike descriptors as an atomic overlap-independent method and propose a new strategy for the structural clustering of antibodies. Methodology:We used a set of 256 functional antibody pairs proposed by Waury et al., modeled with AbodyBuilder2 and segmented into CDRH3, All CDRs, paratopes, and separately obtained epitopes.To represent the surfaces, we employed FP-Zernike, which calculates 3D descriptors and measures similarities using Euclidean distance.As a control, we used SPACE2, a specialized antibody clustering software. Results:The results showed that SPACE2 clustered only 19 pairs with a precision of 0.89.In contrast, clustering based on 3D Zernike descriptors identified 74 pairs with a precision of 0.80 when focusing on the paratope, and 43 pairs with a precision of 0.78 when considering only CDRH3.All CDRs clustered 352 pairs, but with a precision of 0.57.Additionally, the clustering of epitopes resulted in 110 pairs with a precision of 0.91. Conclusion:These results demonstrate that methods based on atomic overlap-free metrics, such as 3D Zernike descriptors, are promising and outperform traditional approaches that rely exclusively on RMSD, standing out as a robust tool for structural analyses in immunology.