FcBinder: A Machine Learning Model for Predicting Antibody Fc Segment Binding Peptides

Abstract:
Phage display technology, a high-throughput screening technique, is widely used in the discovery and development of therapeutic antibodies. In peptide phage display experiments that use intact antibodies as the target, the primary aim is often to select the peptides with high affinity for the antibody variable domains. However, peptides that bind to the Fc fragment (constant domains) may also appear among the screened peptides and are classified as target-unrelated peptides (TUPs) requiring exclusion. Despite the availability of bioinformatics tools for identifying common TUPs, no tool specifically identifies Fc-binding peptides. Such a tool can also be invaluable for studies where Fc-binding peptides are considered target-specific. To address this gap, we developed FcBinder, an ensemble machine-learning model that predicts the Fc segment binding ability of peptides. FcBinder, trained on data from the Biopanning Data Bank (BDB), integrates four sub-models and achieves high specificity (0.98), accuracy (0.95), MCC (0.89), and AUC (0.99), outperforming other machine learning classifiers such as Logistic Regression, K-Nearest Neighbors, GaussianNB, XGBoost, Decision Tree, and Random Forest. It also reliably predicts Fc-binding peptides from non-biopanning data. FcBinder offers a robust computational tool for refining phage display results and advancing therapeutic antibody development. To facilitate accessibility, we have developed an online web server for FcBinder, which is freely available at https://i.uestc.edu.cn/FcBinder/.