AbImmPred: An immunogenicity prediction method for therapeutic antibodies using AntiBERTy-based sequence features.

Area of Science:

• Biotechnology
• Immunology
• Computational Biology

Background:

• Immunogenicity poses a significant challenge in therapeutic antibody development, often leading to adverse immune responses.
• Current wet-lab methods for assessing immunogenicity can be time-consuming and costly during early-stage development.

Purpose of the Study:

• To develop an efficient computational method for predicting antibody immunogenicity from amino acid sequences.
• To facilitate early-stage screening of antibody therapeutics and mitigate development risks.

Main Methods:

• Utilized AntiBERTy pre-trained model to extract sequence features from antibody variable regions.
• Applied Principal Component Analysis (PCA) for dimensionality reduction to two principal components.
• Employed AutoGluon to train and optimize machine learning models, selecting a weighted ensemble model via 5-fold cross-validation.

Main Results:

• The developed computational method achieved an accuracy of 0.7273 on an independent test dataset.
• The proposed method demonstrated a 9.09% improvement in accuracy compared to existing prediction methods.
• The model was trained and validated on a dataset of 199 commercial therapeutic antibodies.

Conclusions:

• The proposed computational approach effectively predicts antibody immunogenicity, offering a valuable tool for early-stage drug development.
• This method can accelerate the screening process and reduce the risks associated with developing therapeutic antibodies.
• A web server for this prediction tool is accessible via GenScript's official website.