Releases: HannesStark/boltzgen
v0.3.1
What's Changed
- Feat/per residue constraints by @buerbaumer in #180
- feat: implement ipSAE metric by @profdocpizza in #176
- feat: Add gemmi 0.6.x/0.7.x version compatibility by @buerbaumer in #175
- Fix gontivimab scaffold path reference by @ekiefl in #188
- Version 0.3.1 by @timodonnell in #196
New Contributors
- @buerbaumer made their first contribution in #180
- @profdocpizza made their first contribution in #176
- @ekiefl made their first contribution in #188
Full Changelog: v0.3.0...v0.3.1
Contributors
Assets 2
v0.3.0
v0.3.0 — protein-redesign protocol, small molecule constraints, and metrics improvements, bug fixes
- Add support for symmetric inverse folding and protein-redesign protocol
- Add covalent bond constraints from file and SMILES specification
- Add absolute metrics computation and small molecule normalization to scoring
- Add per-chain sequences to metrics for multi-chain designs
- Add ranking metrics to PDF output
- Add design mask templates option
- Fix bug when running inverse folding only
- Fix off-by-one error in size bucket filtering
- Fix mmCIF parsing for files with multiple models
- Fix refolding step when there are no designed residues (#122)
- Fix bond length loss issues
- Fix activation checkpointing
- Switch to templateV2 module for correct inter-chain template masking
v0.2.0
Multichain antibody fixes.
v0.1.9
Fix delta sasa and plip metrics to only count between the ~chain_design_mask and design_mask. Previously we counted between ~design_mask and design_mask meaning that we count e.g. the interactions between designed nanobody CDR loops and and the scaffold.
add option to skip analyze step if cif path does not exist. This can happen when using --reuse if a process was interrupted between writing the refold .npz file and the refold .cif file.
v0.1.8
Fix some antibody-related docs, file names, and add the "antibody-anything" protocol to boltzgen cli (it is equivalent to nanobody-anything for now)
Other changes:
- Clamp analyze worker thread pools to prevent oversubscribing by @rhyschappell in #97
- Add SLURM example by @timodonnell in #106
- Add weights_only=False to model checkpoint loading by @keejkrej in #92
- training fixes by @timodonnell in #107
New Contributors
Full Changelog: v0.1.7...v0.1.8
Contributors
Assets 2
v0.1.7
Fix antibody things. Sorry about that.
v0.1.6
add scfv and fab scaffolds
add boltzgen merge command for slurm. You can now run separate chunks and then merge the outputs from the files with the boltzgen merge command. instructions in readme are coming soon.
v0.1.5
add bioarxiv citation
add antibody scaffolds
add training ids
fix small bugs
Version 0.1.4
This release is very similar to the currently-deployed version 0.1.3 on PyPI but with a fix to schema parsing in 497d9ec
The main purpose of this release is testing automatic deployment to PyPI.
Changes since initial release
- Minor bug fixes while setting up the environment by @y1zhou in #12
- Allow offline mode for inference by @nishanthmerwin-benchling in #6
- fix: download command in Dockerfile by @chAwater in #25
- Fix README install instructions: “from source” vs pip install boltzgen by @rhyschappell in #57
Contributors since initial release
- @y1zhou made their first contribution in #12
- @nishanthmerwin-benchling made their first contribution in #6
- @chAwater made their first contribution in #25
- @rhyschappell made their first contribution in #57
Full Changelog: https://github.com/HannesStark/boltzgen/commits/v0.1.4
