Repository navigation
Resistify v2.0.0
Exciting times! Resistify is now really really fast.
New features
- The previous NLRexpress
jackhmmerbackend has been completely replaced by a new classifier that uses ESM2 8M embeddings. It is substantially faster and in most cases more accurate. It can be accelerated via GPU and MPS, but CPU inference is also very fast. Motif predictions will change as a result of this new model. As part of this, the TIR motifs have temporarily been removed. This is not expected to change classifications. - The plotting method has been changed - plots are now generated as SVGs (one per sequence), and are generated by default in the results directory. Plots can be hidden with
--no-draw. - More HMMs have been added to the default NLR database, including Pfam-A LRR models. This should improve domain annotation accuracy slightly, but otherwise should not affect classification accuracy.
- HMMER dependency has been replaced by
pyhmmer- it should be slightly faster and more stable. - There has been a lot of refactoring - refer to the README.md for up-to-date result formats.
Bug fixes
- Probably added a few with this update - raise an issue if you find any...