Preprocessing of amino acid chains of antibody structure for machine learning analysis

Antibody classification represents a task of growing importance in bioinformatics. In recent years, the identification of antibodies capable of recognising and neutralising SARS-CoV-2 has become a central focus in immunological research and bioinformatics. Antibody representation presents several challenges, as antibody structure and function are highly variable, which complicates the development of a universal classification framework. Antibodies are composed of heavy and light chains that contain hypervariable complementarity-determining regions, which define their specificity. These structural variations create substantial challenges for sequence alignment, feature extraction, and classification. In this research, three methods for representing amino acid sequences were compared: TF–IDF, Atchley Factors, and ProtVec. These representations were evaluated using decision trees, logistic regression, and support vector machines. A separate dataset was generated for each representation. The results suggest that the representation based on Atchley Factors achieved comparatively stronger performance in the task of antibody classification. Smart citations: https://scite.ai/reports/10.61467/2007.1558.2026.v17i1.1030 Dimensions.Open Alex.