1.2.0

Hardware support for Blackwell GPUs plus a batch of correctness and crash fixes for the V1 GPU docking engine (unidock/).

⚠️ The grid-buffer and macrocycle fixes change docked results for the affected inputs — they correct previously wrong / undefined behavior. See notes below.

Added

• Blackwell / CUDA 12.8 support — builds for sm_100 (B100/B200/GB200) and sm_120 (RTX 50/B40); CUDA bumped to 12.8 (#184).

Fixed — correctness (results change for affected inputs)

• Grid buffer sized for the default box — MAX_NUM_OF_GRID_POINT 80³ → 81³ (531,441), so the default 30 Å / 0.375 Å box fits exactly. Previously the bounds check was a release-stripped assert and the grid silently overflowed, corrupting ~20% of results for boxes ≥ 30 Å and crashing (SIGSEGV) on larger boxes (#195, #177; closes #174, #14).
• Large boxes now error cleanly — boxes exceeding the GPU grid buffer print a clear message and exit instead of silently overflowing / SIGSEGV (#177).
• Macrocycle (Meeko CG/G) affinities — the ring-closure "glue" energy no longer leaks into the reported affinity (it's a sampling artifact used only to close the ring during optimization). Fixes absurd, non-reproducible scores for macrocycle ligands (#196; closes #102, #192).
• Out-of-bounds poses are skipped instead of being reported with FLT_MAX (3.4e38) affinities (#182; closes #144).

Fixed — crashes

• CUDA error 700 (illegal address) in --gpu_batch from a cudaMemset pointer bug (#178; addresses #115/#118/#136).
• Segfault when a ligand fails to parse in a batch — the bad ligand is skipped with a warning instead of crashing the batch (#181; closes #112).
• SIGSEGV on unsupported atom types in SDF (e.g. boron) — SDF atoms are validated like the PDBQT path and unsupported ones are skipped (#180; closes #138).
• Intermittent SIGSEGV from a NaN / non-unit axis in angle_to_quaternion — guarded with a renormalize / identity-quaternion fallback (#179; closes #160).
• CUDA error 700 with flexible residues (--flex) — ≤ 1 flex torsion now docks on GPU; > 1 exits with a clear "supports at most 1" message instead of crashing (#176; closes #159).

Build / CI

• Fix -DFETCH_BOOST=ON against system Boost ≥ 1.90 — link the fetched Boost::math into the CUDA target so nvcc doesn't pick up the system headers (#189, #197).
• Repair the (manually-dispatched) benchmark workflow: call the current run_test.py entry point and download the dataset (#194).
• Pin Python 3.12 for the tools/CI images (openbabel compatibility) (#183, #188); dependency bumps (#185, #186, #187).

Changed / Removed

• License unified to Apache-2.0 (#153, #155).
• Removed the built-in receptor processor (it didn't support most user proteins) (#164).

Known limitations

• Macrocycle sampling: the broken ring isn't held as tightly closed as AutoDock during the GPU search (glue equilibrium ~1.5 Å vs ~0.03 Å), so absolute affinities for macrocycles can be weaker than AutoDock even though scores are now sane and reproducible (tracked in #37).
• GPU flexible-receptor docking supports at most 1 flex torsion (#175 prototype for more).

Docker: dptechnology/unidock:1.2.0 (and :latest) are built and pushed by the release workflow.

Full changelog: 1.1.3...1.2.0

1.1.3

This release contains minor bug fixes.

What's Changed

• feat:support bias by @dp-yuanyn in #131
• fix: add kernel type info by @ysyecust in #132
• Enhance run info by @lucifer1004 in #134
• Bump docker/build-push-action from 5 to 6 by @dependabot in #135
• new version for pdb2pdbqt by @zhangplpl in #117
• Optimize CUDA codes in precalculate.cu by @kongexp in #137
• fix: typo in Dockerfile by @k-ujihara in #145
• Fix uninitialized f0 and re-enable sm75/sm86/sm89 by @lucifer1004 in #120
• fix: update boost::filesystem include header by @caic99 in #146
• chore: update project version to 1.1.3 by @caic99 in #147

New Contributors

• @dependabot made their first contribution in #135
• @zhangplpl made their first contribution in #117
• @kongexp made their first contribution in #137
• @k-ujihara made their first contribution in #145

Full Changelog: 1.1.2...1.1.3

1.1.2

What's Changed

• docs: remove unidock_ad4 in README by @dp-yuanyn in #98
• fix: ligandprep skip failed cases by @dp-yuanyn in #99
• Update version number in CMakeLists.txt by @ysyecust in #101
• Docs: add conda-forge install instructions by @caic99 in #103
• Docs: fix typo in conda-forge instructions by @caic99 in #104
• Docs: add introduction to uni-dock benchmarks by @caic99 in #105
• Fix path error in README.md by @ysyecust in #108
• build: add unidock_tools dependency in toml by @dp-yuanyn in #110
• fix:unidock pipeline one ligand failed case by @dp-yuanyn in #116
• [feature] unidock tools add required dependencies by @Hong-Rui in #119
• fix: env readd mgltools and remove tests using ad4 temporary by @dp-yuanyn in #121
• test: readd ad4 tests by @dp-yuanyn in #122
• fix: unidock pipeline ligand index list too long err by @dp-yuanyn in #127
• build: change docker build runner from github to self by @dp-yuanyn in #128
• feat: update ligand group method by @ysyecust in #125

Full Changelog: 1.1.1...1.1.2

1.1.1

What's Changed

• fix: mcdock write same conf in confgen stage by @dp-yuanyn in #79
• Format code and remove unused code by @ysyecust in #81
• Test: use public image for testing unidock_tools by @caic99 in #83
• Refactor CMakeLists.txt by @wzf03 in #80
• Fix: array index out of range by @caic99 in #86
• fix:mcdock rigid docking; test:remove mcdock rmsd assert by @dp-yuanyn in #87
• Build: update conda build script for CUDA 12 and uni-dock subfolder by @caic99 in #90
• feat: unidock_pipeline add support for score_only and local_only; doc: update unidock perf png by @dp-yuanyn in #91
• doc: update unidock perf images by @dp-yuanyn in #92
• Build: disable sm_90 arch for forward compatibility on CUDA 11.2 by @caic99 in #95
• fix: unidock_pipeline read only one ligand for big sdf by @dp-yuanyn in #96
• Docs: add cuda toolkit requirement by @caic99 in #97

New Contributors

• @wzf03 made their first contribution in #80

Full Changelog: 1.1.0...1.1.1

1.1.0

What's Changed

Features

• v1.1 by @ysyecust @dp-yuanyn @zhengh96 @caic99 in #77
  • Templatize compute kernels
  • Update docs, CI/CD pipeline, and Dockerfiles
• unidock_tools support PDB-format protein file, support ligand file without hydrogen and 3D conf by @YNYuan in #19
• Hypnopump/rough edges by @hypnopump in #22
• sync torsion tree update from nashtools by @Hong-Rui in #43
• Warp parallel by @lucifer1004 in #56
• CUDA stream optimisation by @envidian in #63
• Feature/hbond by @ysyecust in #64
• [GPU setting] device id by @envidian in #71
• [paired_batch_size] Write score by @envidian in #72
• Update MAX_NUM_OF_RIGID_ values in SmallConfig, MediumConfig, and Lar… by @ysyecust in #78

Bug Fixes

• Fix warning by @ysyecust in #17
• remove obsolete debug info by @caic99 in #16
• Fix typo in logging by @lucifer1004 in #36
• Style/formatcode by @ysyecust in #46
• fix ligand_index option bug by @UR-Free in #48
• Minor fix for logical AND by @prabindh in #49

Docs

• Update docs for boost version requirement by @caic99 in #18
• Update FAQ in README.md for uni-dock by @caic99 in #38
• docs: update installing instructions by @caic99 in #47
• Update README.md by @baixiaokuang in #52
• Docs: Add Uni-Dock logo by @caic99 in #68

CI/CD

• cicd trigger in main by @YNYuan in #26
• unidock_tools ci by @YNYuan in #24
• fix ut by @dp-yuanyn in #66

New Contributors

• @ysyecust made their first contribution in #17
• @caic99 made their first contribution in #16
• @YNYuan made their first contribution in #19
• @UR-Free made their first contribution in #20
• @hypnopump made their first contribution in #22
• @lucifer1004 made their first contribution in #36
• @prabindh made their first contribution in #49
• @baixiaokuang made their first contribution in #52
• @envidian made their first contribution in #63
• @dp-yuanyn made their first contribution in #66

Full Changelog: https://github.com/dptech-corp/Uni-Dock/commits/1.1.0

1.0.0

The executable of Uni-Dock using vina & vinardo scoring function unidock and ad4 scoring function unidock_ad4 is provided below.
Use conda to install Uni-Dock: conda install -c https://conda.mlops.dp.tech/caic -c conda-forge unidock. An offline conda package is provided in the attachment: run conda install -c conda-forge unidock-1.0.0-5_5.tar.bz2 after downloading the file.