Closed wthrif closed 1 year ago
Hi @wthrif. The mutations we are referring to in figure 19 are not mutations compared to human germline, but humanizing mutations made in the 25 experimentally verified humanized sequences compared to their parental sequence. So here, a high Sapiens score is expected.
We intentionally haven't used Sapiens as an "immunogenicity" score, because OASis provides a more granular and interpretable metric. But you could definitely use Sapiens this way, I would be interested in seeing the results. A Sapiens score of 1 would be a perfectly human sequence, although I assume you would never achieve a full 1. Correspondingly, a non-human antibody sequence wouldn't achieve a 0 but something much higher, since antibodies are still conserved across species.
Thanks! Sorry I didn't read your paper closely enough. I'm just comparing it to other methods that decide if a whole antibody is immunogenic or not, so that's why I just looked at the score, but I love the visualization tool you all have provided to seeing where an AB is non-human. The Sapiens score actually works pretty well as an immunogenicity score!
The paper says: We observed that most mutations achieved the first or second highest Sapiens score (Supplementary Figure 19). Which makes me think higher is more immunogenic, but I'm getting more accurate results with lower is better.