v2.0.0
Calling this version 2.0.0 to recognize that MHCflurry has evolved quite a bit in the last year. This release itself is not a major departure from the 1.6.0 version, however.
Version 2.0 introduces new models (BA/AP/PS). These follow the same overall design as in 1.6.0 but incorporate updated training data and some small tweaks:
- BA predictor: MS hits are assigned < 100 nM affinity instead of < 50 nM for training
- AP predictor: flanking window size decreased to 5 amino acids per flank from 15. Not seeing an accuracy decrease and it makes the models faster.
The code has been ported to tensorflow 2.0. Users will need to upgrade to a recent tensorflow to use MHCflurry.
Other changes:
- Support for percentile ranks for PS predictions
- Revamped Docker support
- Small fixes and docs updates