37f65c068b7723cd7809ee2d31d7861c-AuthorFeedback.pdf

We have updated our table to make this clearer. We have expanded the discussion of alignment-based3 features in the Background section to make it clearer what they are. For the single-prediction tasks, we use a learned attention-weighted sum over the amino-acid21 representations. In Table 2 of the paper we chose to report simple global metrics.44 However, we report many detailed metrics in the appendix, including some of the metrics the reviewer suggested:45 precision-recall, AUPRC, and accuracyfor long-range contacts.