Antibody Clustering Using a Machine Learning Pipeline that Fuses Genetic, Structural, and Physicochemical Properties

Area of Science:

• Immunology
• Structural Biology
• Bioinformatics

Background:

• Antibody V domain clustering is crucial for immunology, disease linkage, and new medicine discovery.
• Existing clustering methods lack consensus and utilize diverse data types like sequences and structures.
• Antibody drug conjugates (ADCs) represent a significant area for new therapeutic development.

Purpose of the Study:

• To develop a novel antibody V domain clustering method.
• To leverage antigen interaction site comparisons for enhanced specificity.
• To utilize deep learning and data mining for advanced classification.

Main Methods:

• Comparison of antibody-antigen interaction sites.
• Extraction of multidimensional information from 3D structures of antibody-antigen complexes.
• Transformation of antibody V domain frameworks into a binary format using structural features.
• Implementation of a three-level hybrid classification scheme.

Main Results:

• A new method for antibody V domain clustering based on interaction site comparison.
• Specific clustering analysis using deep learning and data mining techniques.
• Identification of new antibody signatures in V domain binding activity.
• Generation of multilevel information on antibodies and antibody-antigen complexes.

Conclusions:

• The proposed method offers a more specific approach to antibody V domain clustering.
• This technique combines sequence, structural, and interaction pattern data for comprehensive classification.
• The developed clusters provide valuable insights for immunological research and therapeutic development.