Fix dependency#5
Open
MarkAStevens04 wants to merge 48 commits into
Open
Conversation
- infer.py and SLURM/build scripts (submit.sh, vina_resume_build.sh, setup_qv2gpu.sh) - CRBN warhead-anchored docking pipeline (anchor_dock.py, batch_anchor_dock.py, etc.) - receptor prep (clean.py, fix_model.py) and decoy generation - wandb logger + config updates - .gitignore: exclude Miniconda installer, scratch symlinks, and docking data artifacts Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Pivot the docking-reward work from CRBN/5HXB (which hit a discrimination ceiling: anchored docking is blind to GSPT1 cooperativity) to the CDK12-DDB1/CR8 system (6TD3), where the glue interface is ligand-mediated and capturable by docking. - 6TD3 prep + validation: clean_6td3.py (tiers + native CR8), redock_cr8.py (blind redock recovers native at 1.23 A, ranked #1; DDB1 bonus -2.2 kcal/mol). - Discrimination: make_decoys_cdk.py (purine-warhead decoys), dock_cluster.py (4-GPU batched docking + Tier1 score_only differential). DDB1 differential separates real glues from decoys 85.6% vs 7.3%. - Cross-system: dock_cluster_crbn.py computes the analogous GSPT1 differential; compare_systems.py lines up both. 6TD3 +78 pts vs 5HXB -3 pts on the same metric. - Cluster jobs run on Balam debug_full_node (4x A100); ~21x (CRBN) / 6.6x (6TD3) vs sequential, mostly from multi-ligand batching (see Logs/004). - Logs/: experiment journal (001-004) with methods, results, and result locations. - test-data/: curated known-glue datasets (DDB1_CDK12, CRBN_GSPT1). Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Consolidate the docking-oracle pre-processing for future contributors: - Add pre-processing/README.md: the 5-step pipeline (prep -> validate -> decoys -> dock -> compare), env vars, local-vs-cluster usage, and a guide for adding a new glue system. - Remove redundant sequential runners (dock_6td3_batch.py, batch_anchor_dock.py, run_*.sh) now subsumed by the batching dock_cluster*.py drivers, and collapse three compare scripts into compare_systems.py. - Delete abandoned pre-processing-v2/ and stale Vina exploration artifacts. - Make the gnina launcher path overridable via $GNINA (was hardcoded). - Point the 6TD3 job summary at compare_systems.py; update Logs references. Verified: all kept scripts compile, imports resolve, and compare_systems.py reproduces the cross-system result from the committed CSVs. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
The Vina-GPU oracle runs from the $SCRATCH install (reached via the quickvina_dir symlink, per configs/proxies/target/docking_base.gin); the repo-side build dir and logs were one-off build artifacts. Verified the $SCRATCH install is self-contained and its QuickVina2-GPU binary still links before removing: - delete vina_workspace/ (141 MB build dir), vina_*build*.log, the 163 MB Miniconda3 installer, and the ad-hoc vina_resume_build.sh (the canonical build recipe is external/setup_qv2gpu.sh). - gitignore these so they don't reappear. Runtime unchanged: quickvina_dir -> $SCRATCH/vina_gpu/Vina-GPU-2.1 intact. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Docking 6td3 cross system
Rewrites all four existing logs from the old Objective/Method/Results/Conclusion/Files structure into the new story + audit layer format: Question, Context & Summary, Answer, Relevance to our Publication, Next Experiments, then # Re-creation with Relevant Files, Versions, Resources, Method, and Results. All factual content preserved. Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
ROC and precision-recall curves for 6TD3 and 5HXB using existing docking result CSVs. 6TD3: AUC=0.890, AP=0.872. CRBN: AUC=0.627, AP=0.726. Adds experiment log 005 and updates README index. Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Establish a clear old/new boundary so future work extends RGFN without editing it, and scaffold the structure for new oracles, rewards, batch selection, benchmarks, and dataset tooling. Key finding: rgfn/ and configs/ were already ~pristine upstream; our only edits were 3 operational overrides (now documented in docs/PATCHES.md). So this is mostly additive scaffolding + relocating our own files. - Add glue/ package (oracles, rewards, samplers, proxies, datasets) with a registry.py import hub for gin discovery and a working example proxy adapter. - Add scripts/train.py + scripts/infer.py wrappers that import glue (so gin sees our components) then delegate to the upstream entry points via runpy. - Add configs/glue/ overlay dir; scaffold benchmarks/, models/, data/synthetic/. - Move pre-processing/ -> research/preprocessing/ (internals intact); fix the repo-root-relative path refs in the moved docking scripts. - Move submit.sh -> scripts/hpc/submit.sh. - Add CLAUDE.md (boundary rules + how to extend) and docs/ (ARCHITECTURE, REFACTOR_LOG, PATCHES) so future agents reference this structure. - Fix .gitignore: drop the `scripts/` ignore (was hiding our entry points), repoint moved paths, ignore docking outputs / weights / synthetic datasets. - Declare glue + gin_config packages in pyproject; update README. Validated statically (Balam down, laptop lacks heavy deps): py_compile, gin include integrity, glue import chain, bash -n, no dangling refs. Real import/train/dock verification pending on Balam (see docs/REFACTOR_LOG.md). Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
…ture [REFACTOR] Separate upstream RGFN from our work via glue/ package
…2 vs Tier 2 - Tier 1.
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
pip install -e .threw an error because of a dependency conflict. Resolved the conflict with this code.