Mining adaptive immune receptor repertoires for biological and clinical information using machine learning
Victor Greiff, Gur Yaari, Lindsay G. Cowell
Abstract
The adaptive immune system stores invaluable information about current and past immune responses and may serve as an ultrasensitive biosensor. Given the immune system's critical role in a wide variety of disease types, this has broad implications for biomedicine. Machine and deep learning is being leveraged to decipher how information is encoded in adaptive immune receptor repertoires to enable prediction from adaptive immune responses and fast-track vaccine, therapeutics, and diagnostics development. Recent advances include predicting the presence of immunity after vaccination or infection, predicting the presence of disease, and designing antibody-based therapeutics. Outstanding challenges encompass increasing our knowledge of the feature space structure that encodes relevant immune information, addressing the lack of ground truth–labeled data, and improving our handling of genetic and environmental confounding factors.