Unconstrained generation of synthetic antibody–antigen structures to guide machine learning methodology for antibody specificity prediction

Abstract

Machine learning (ML) is a key technology for accurate prediction of antibody-antigen binding. Two orthogonal problems hinder the application of ML to antibody-specificity prediction and the benchmarking thereof: The lack of a unified ML formalization of immunological antibody specificity prediction problems and the unavailability of large-scale synthetic benchmarking datasets of real-world relevance. Here, we developed the Absolut! software suite that enables parameter-based unconstrained generation of synthetic lattice-based 3D-antibody-antigen binding structures with ground-truth access to conformational paratope, epitope, and affinity. We formalized common immunological antibody specificity prediction problems as ML tasks and confirmed that for both sequence and structure-based tasks, accuracy-based rankings of ML methods trained on experimental data hold for ML methods trained on Absolut!-generated data. The Absolut! framework thus enables real-world relevant development and benchmarking of ML strategies for biotherapeutics design.

Performance comparison of TCR-pMHC prediction tools reveals a strong data dependency

“…We suggest that peptide imbalance contributes more to a better performance of the models than size, a finding that was also made in antibody-antigen prediction Robert et al. ( 17 ). It will be interesting to see whether the models perform well purely because of peptide frequency, or whether other factors such as biological or physicochemical properties may influence performance.…”

Simulation of adaptive immune receptors and repertoires with complex immune information to guide the development and benchmarking of AIRR machine learning

“…1C), in line with previous studies ( 2–4 , 63 ). In addition, AIR information may include derived features such as CDR3 length ( 64 ), physicochemical properties ( 65 ), or binding energy ( 66 ). In the case of paired chain data, the immune signal would include this information from both chains ( 3 , 15 , 16 , 67 ).…”

“…There is a potential risk that an AIRR-ML model trained on simulated data may only learn the artifacts of the simulation framework, thereby impacting the applicability of ML model-related insights to real-world scenarios. This may be addressed in the future by analyzing how ML results on synthetic data transfer to experimental data as recently demonstrated by Robert et al ( 66 ). Further work is needed on the biological understanding of immune signals for improved simulation ( 38 , 63 ).…”

A colloidal gold immunoassay strip assay for cadmium detection in oilfield chemicals

“…28 The epitope, which is present on the surface of the antigen, con-sists of a continuous sequence or a discontinuous threedimensional protein structure. 29 In nature, Cd usually exists in the +2 form. Cd 2+ , which is not immunogenic due to its small molecular weight, cannot be combined with a carrier protein to form a complete antigenic epitope.…”