Abstract:
The effective use of antibodies in personalized medicine hinges on their ability to bind
strongly and specifically to their target antigens. This binding capability is critical for
designing precise and successful therapeutic treatments tailored to individual patients.
Identifying factors that predict binding affinity is essential to developing therapeutic
antibodies with enhanced efficacy. However, experimental methods to measure antibodyantigen interactions are often costly and time-consuming, posing a challenge due to the
diverse structures of antibody regions that influence binding. This study employs advanced
machine learning and deep learning techniques, including Neural Networks, XGBoost,
Random Forest, and Support Vector Machines (SVM), to classify and predict antibodyantigen binding affinities based on structural antibody data. Using data from antibodyantigen complexes sourced from the Structural Antibody Database (SAbDab), key features
are extracted, including structural, energetic, and interaction-based metrics, to differentiate
antibodies with high and low binding affinities. The models are evaluated using
classification metrics (accuracy, precision, recall, F1 score) and regression metrics (MSE,
R² score), with Random Forest emerging as the top performer in binding affinity prediction.
By integrating neural networks with traditional machine learning approaches, this research
provides a more efficient alternative to experimental methods, enhancing our
understanding of molecular-level interactions between antibodies and antigens. This
approach contributes to the optimization of therapeutic antibody design, further advancing
the field of personalized medicine
