MultiSAAl: Sequence-Informed Antibody-Antigen Interaction Prediction Using Multiscale Deep Learning.

Area of Science:

• Biochemistry
• Computational Biology
• Immunology

Background:

• Antibody-antigen interaction prediction is crucial for developing new therapeutics but is often limited by experimental costs and the inability of static structure-based methods to capture dynamic binding.
• Accurate sequence-based prediction methods are needed to overcome the limitations of structure-based approaches in modeling the dynamic conformational changes critical for antibody-antigen binding.

Purpose of the Study:

• To develop and validate MultiSAAI, a novel sequence-informed computational framework for predicting antibody-antigen interactions.
• To explicitly model the distinct contributions of antibody heavy and light chains in antigen binding.
• To improve the accuracy and efficiency of therapeutic antibody discovery through advanced computational prediction.

Main Methods:

• MultiSAAI integrates language model embeddings, physicochemical properties, geometric constraints, and residue substitutability to characterize interactions.
• A multiscale network architecture is employed to evaluate both global residue-pair compatibility and local amino acid fitness at the binding interface.
• The framework incorporates site-specific information and biologically grounded binding principles to reflect actual interaction mechanisms.

Main Results:

• MultiSAAI achieved an AUROC score of 0.772 on a generic antibody-antigen interaction dataset.
• The model demonstrated superior performance on a SARS-CoV-2 dataset, achieving an AUROC score of 0.947.
• MultiSAAI outperformed established methods like A2binder and AbAgIntPre in benchmark tests.

Conclusions:

• MultiSAAI offers a powerful sequence-based approach for predicting antibody-antigen interactions, surpassing current methods.
• The framework's ability to model dynamic interactions and incorporate biological principles enhances its predictive accuracy.
• MultiSAAI shows significant potential for accelerating high-throughput therapeutic antibody discovery.