ABAG-docking benchmark: a non-redundant structure benchmark dataset for antibody-antigen computational docking.

Area of Science:

• Structural Biology
• Computational Chemistry
• Immunology

Background:

• Accurate prediction of antibody-antigen complex structures is essential for developing novel therapeutics and diagnostics.
• Existing antibody-antigen docking (ABAG-docking) datasets have limitations in diversity and scope.
• Advancements in computational methods require robust and comprehensive benchmark datasets for rigorous evaluation.

Purpose of the Study:

• To create and characterize a comprehensive, non-redundant benchmark dataset for antibody-antigen complex structures.
• To provide a diverse set of cases categorized by docking difficulty, interface properties, and structural characteristics.
• To facilitate the development and evaluation of advanced computational methods for predicting antibody-antigen interactions.

Main Methods:

• Reviewed existing ABAG-docking datasets.
• Constructed a new benchmark dataset by adding 112 cases (14 sdAb, 98 mAb) to existing benchmarks, increasing the proportion of difficult cases.
• Categorized complexes based on docking difficulty, interface properties, and structural characteristics.
• Developed a pipeline for periodic dataset updates.
• Utilized multiple prediction methods (ZDOCK, ClusPro, HDOCK, AlphaFold-Multimer) for dataset analysis.

Main Results:

• Introduced a comprehensive benchmark dataset with 112 new antibody-antigen complex cases, including single-domain antibodies (sdAbs) and monoclonal antibodies (mAbs).
• The dataset features diverse antibody types (human/humanized, sdAbs, rodent) and increased representation of difficult docking cases.
• Initial analysis using multiple prediction tools demonstrated the dataset's utility for evaluating docking performance.

Conclusions:

• The new benchmark dataset provides a valuable resource for advancing computational antibody-antigen docking methods.
• This dataset will enable the development of more accurate tools for predicting and designing antibody-antigen complexes.
• The availability of this dataset promotes rigorous evaluation and improvement of algorithms in drug discovery and vaccine design.