Parapred: antibody paratope prediction using convolutional and recurrent neural networks

Area of Science:

• Immunology
• Computational Biology
• Bioinformatics

Background:

• Antibodies are crucial for vertebrate immunity, research, and diagnostics.
• Identifying antibody paratopes (antigen-binding sites) from amino acid sequences is challenging.

Purpose of the Study:

• To develop a novel computational method for accurate paratope prediction using only antibody sequence data.

Main Methods:

• A sequence-based probabilistic machine learning algorithm, Parapred, was developed.
• Parapred utilizes a deep-learning architecture incorporating local and global sequence features.

Main Results:

• Parapred significantly improves upon existing state-of-the-art paratope prediction methods.
• The algorithm accurately predicts paratopes using only hypervariable region sequences, without antigen information.
• Parapred predictions enhance the speed and accuracy of rigid docking algorithms.

Conclusions:

• Parapred offers a powerful, sequence-based approach for antibody paratope prediction.
• The tool has implications for antibody engineering, drug discovery, and diagnostic development.
• Parapred is publicly available as a webserver and for download.