ImmUQBench: A Benchmark on Uncertainty Quantification of Protein Immunogenicity Prediction

Discovering antigen proteins, capable of eliciting desired immune responses, is of paramount importance in developing immunogenic therapeutics for combating various diseases, particularly autoimmune disorders, infectious diseases, as well as cancers. Accurate and generalizable immunogenicity prediction with recent AI/ML advancements that can guide antigen design has emerged as a crucial subject in computational therapeutic discovery. However, due to insufficient labeled data, existing approaches tend to be overly simple. Many immunogenicity prediction models do not generalize well, making their predictions unreliable. Uncertainty Quantification (UQ) approaches are commonly used to address the aforementioned challenges when applying AI/ML methods with limited training data, aiming to reduce the risk of catastrophic errors. In developing AI/ML immunogenicity prediction models, these errors may lead to significant waste in cost and time for consequent therapeutic development for new immunogenic antigen proteins. We here present ImmUQBench, a benchmark for evaluating different well-known UQ methods for antigen immunogenicity prediction. Our work has the potential to facilitate more effective and reliable therapeutic antigen design, by providing insights into the efficacy of different UQ methods on immunogenicity predictions.