AB-Panda: An AI-Generated Antibody Structure-Based Tool for Developability Prediction

Abstract

The developability of antibodies is a critical concern in antibody discovery, encompassing issues such as self-interaction, aggregation, and thermal stability. The use of computational and structure-based tools has greatly improved the evaluation and prioritization of initial antibody sequences. With the increasing demand for subcutaneous administration of small-volume, high-concentration antibody formulations, there is a need for more accurate prediction tools based on protein structures. Our study introduces AB-Panda, a tool based on AlphaFold2-predicted antibody structures and three innovative structure-related metrics. AB-Panda utilizes unit-area hydrophobic value (UHV), unit-area positive charge (UPC), and unit-area negative charge (UNC) to automatically identify hydrophobic and charged patches within the complementarity determining regions (CDRs) of antibodies. Through the analysis of the 919 clinical stage therapeutic (CST) antibodies, we have established recommended ranges of UHV, UPC, and UNC as reference standards for antibody developability. AB-Panda offers clear visualizations of surface hydrophobic and charge distribution, facilitating the identification of problematic amino acids and providing suggestions for further sequence engineering. Additionally, AB-Panda has been integrated into a web application, available at https://www.antibodydev.com, by combining UHV, UPC, UNC, and other established computational metrics for the early screening and optimization of antibody sequences.

Keywords: computation tool; developability prediction; unit‐area hydrophobic value; unit‐area negative charge; unit‐area positive charge.