Quantifying the nativeness of antibody sequences using long short-term memory networks

Area of Science:

• Biotechnology
• Immunology
• Bioinformatics

Background:

• Antibody engineering is crucial for developing therapeutic candidates with optimal developability.
• Maintaining native-like antibody sequences enhances overall library quality.
• Deep learning advancements offer new opportunities for analyzing biological sequence data.

Purpose of the Study:

• To develop a deep learning model for quantifying the nativeness of antibody sequences.
• To leverage large antibody sequence datasets for improved antibody engineering.
• To provide a tool for evaluating antibody libraries and guiding humanization processes.

Main Methods:

• A bi-directional long short-term memory (LSTM) network was developed.
• The LSTM model was trained on extensive human antibody sequence data.
• Model performance was evaluated by its ability to distinguish antibody species and assess library quality.

Main Results:

• The developed LSTM model effectively quantifies antibody sequence nativeness.
• The model demonstrated superior performance in distinguishing human antibodies from non-human species compared to other methods.
• The approach proved applicable for evaluating synthesized antibody libraries and humanizing mouse antibodies.

Conclusions:

• Deep learning, specifically LSTM networks, can accurately assess antibody sequence nativeness.
• Quantifying nativeness is a valuable consideration in antibody design and engineering.
• This method offers a practical tool for enhancing antibody developability and therapeutic potential.