eLife Assessment
This study provides an important method to model the statistical biases of hypermutations during the affinity maturation of antibodies. The authors show convincingly that their model outperforms previous methods with fewer parameters; this is made possible by the use of machine learning to expand the context dependence of the mutation bias. They also show that models learned from nonsynonymous mutations and from out-of-frame sequences are different, prompting new questions about germinal center function. Strengths of the study include an open-access tool for using the model, a careful curation of existing datasets, and a rigorous benchmark; it is also shown that current machine-learning methods are currently limited by the availability of data, which explains the only modest gain in model performance afforded by modern machine learning.