Skip to content

CHANGELOG.md

Latest commit

 

History

History
238 lines (200 loc) · 11 KB

CHANGELOG.md

File metadata and controls

238 lines (200 loc) · 11 KB
Raw

Changelog

All notable changes to SMSD Pro are documented in this file.

[7.1.1] - 2026-04-14

Bug-fix patch on top of v7.1.0. No new features, no public API breakage.

Fixed

  • Cross-language ECFP / FCFP fingerprint parity between Java, C++, and Python. Two long-standing Java drifts at radius ≥ 1 (signed vs unsigned neighbour-hash sort, and pharmacophore implicit-H count for pyrrole-type nitrogen) now produce bits byte-identical to the C++ / Python reference.
  • Java canonical SMILES writer: bond symbol for aromatic-adjacent single bonds, and implicit H count inside stereo brackets.
  • Python smsd.canonical_smiles(smi) / smsd.to_smiles(smi) raised TypeError on string input. Both now accept str or MolGraph.
  • MatchResult.overlapCoefficient returned the wrong similarity metric. Both MatchResult.overlap and MatchResult.overlapCoefficient now return Szymkiewicz-Simpson overlap as documented; the new MatchResult.tanimoto attribute exposes the Jaccard value. __repr__ shows both.
  • Canonical SMILES writer now emits [nH] for pyrrole-type aromatic nitrogen, so SMSD output kekulizes cleanly in downstream readers (pyrrole, indole, carbazole, fused benzo-pyrrole).
  • FP-level smsd.overlapCoefficient and smsd.count_overlapCoefficient camelCase aliases now return Simpson overlap as documented; smsd.tanimoto_coefficient / smsd.count_tanimoto_coefficient expose Jaccard. Empty-vs-empty convention aligned at 1.0 (trivially identical) for all similarity helpers.

Removed

  • Dead C++ fingerprint shim headers under cpp/include/fp/. The canonical C++ fingerprint API is smsd::batch::detail::* in cpp/include/smsd/batch.hpp, documented in docs/CPP.md.

Verified

  • Python pytest: 603 passed, 6 skipped, 0 failed.
  • Java JUnit: 581 tests, 0 failures, 0 errors.
  • Canonical SMILES reference set matches character-for-character across Java, C++, and Python.

[7.1.0] - 2026-04-14

Added

  • Comprehensive test suite: 597 tests across 9 test files
    • test_api_coverage.py (160 tests): MCS utils, batch ops, TargetCorpus, file I/O, scaffolds, coordinate transforms, depiction, SMARTS, enums
    • test_fingerprints.py (128 tests): all FP types, similarity metrics, edge cases (single-atom, disjoint, empty), challenging molecules (taxol, C60, cubane, morphine/codeine, enantiomers), mathematical properties
    • Dalke, stress, Ehrlich-Rarey, and Tautobase benchmark tests retained

Fixed

  • tanimoto_coefficient / overlap_coefficient crashed with sparse count input from circular_fingerprint_counts() — now auto-detects format
  • counts_to_array() only accepted dict — now accepts list[tuple] too
  • fingerprint() raised ValueError for kind='ecfp'/'fcfp'/'torsion'
  • Doc crashes: result.tanimoto → result.overlap (AttributeError), r['rgroups'] → R-group dict keys (KeyError)
  • Doc silent bugs: bond_order_mode= → match_bond_order=, removed invalid solvent=/pH=/chem= kwargs from find_mcs examples
  • decompose_rgroups → decompose_r_groups in PYTHON.md and EXAMPLES.md
  • All 196 smsd.xxx() references across docs verified against actual exports
  • Replaced all stale API names (ecfp_counts, dice_similarity, smarts_search, etc.)

[7.0.0] - 2026-04-13

Summary

Major release: unified API, clean break from legacy aliases, full Java parity.

Breaking Changes

  • Removed smsd.mcs() — use smsd.find_mcs()
  • Removed smsd.substructure_search() — use smsd.find_substructure()
  • Removed smsd.all_mcs() — use smsd.find_mcs(mol1, mol2, max_results=N)
  • Removed camelCase aliases: overlapCoefficient, tanimoto, count_overlap_coefficient, count_tanimoto

Added

  • Unified Python API: find_mcs(mol1, mol2, max_results=1) and find_substructure(query, target, max_results=1)
  • Java convenience methods: SearchEngine.findMCS(g1, g2) and SearchEngine.findSubstructure(query, target) with MolGraph and IAtomContainer overloads
  • Raw C++ bindings renamed to _native_* prefix (clearly internal)

Changed

  • All internal calls updated to unified API names
  • mcs_from_smiles(), mcs_rdkit(), substructure_rdkit(), depict_mcs(), depict_substructure() use new API
  • __all__ cleaned of all deprecated entries
  • overlapCoefficient([], []) returns 1.0 (trivially identical empty sets)

Platforms

  • macOS (arm64 Apple Silicon, x86_64), Linux (x86_64, aarch64), Windows (AMD64)
  • GPU: Metal (Apple Silicon), CUDA (Volta+)
  • Java 25+, C++17, Python 3.10-3.13

[6.12.0] - 2026-04-07

Summary

Correctness, performance, and API cleanup release.

Included

  • Fixed memory leak in SearchEngine cache
  • Fixed overflow in graph-bound computation for large molecular graphs
  • Added missing CIP Rule 3 (Z > E) per IUPAC 2013 in Java and C++
  • Thread-safety: volatile on lazy-init fields in MolGraph
  • Updated tautomer weights: nitroso-oxime 0.95, nitro-aci 0.95, pyridone 0.95
  • Added selenium to tautomer compatibility, iodine to scoring
  • Corrected SAH test SMILES (thioether, not ester connectivity)
  • Relaxed formal charge matching in the default MCS profile
  • Renamed Mcs* types to MCS*, tanimoto to overlapCoefficient
  • Improved MCS construction throughput via faster compatibility graph traversal
  • Reduced allocation pressure throughout the MCS pipeline
  • Faster convergence on symmetric ring systems
  • Faster substructure search domain initialisation
  • Improved throughput on Apple Silicon with native vector operations
  • Stage-aware pipeline routing to skip unnecessary MCS stages
  • MCSStageTimers profiling API for pipeline diagnostics
  • TargetCorpus and batch_find_substructure() Python APIs
  • Reduced allocations per query; thread-local SMILES parsing
  • SDF batch cap at 100K molecules

[6.11.1] - 2026-04-04

Bug Fixes

  • ECFP initial invariants: corrected circular fingerprint atom invariants to include bond-order and mass contributions in both binary and count ECFP variants (C++ and Java)
  • Path fingerprint canonical hash: corrected path fingerprint to use a single canonical hash direction, fixing bit density inflation
  • FCFP pyrrole-N misclassification: aromatic nitrogen acceptor classification now uses direct hydrogen count, fixing incorrect non-acceptor assignment for pyridine-N (pyridine N has a free lone pair; pyrrole N does not)
  • Thread safety: prewarmGraph() now initialises the pattern fingerprint before entering parallel regions, preventing data races on lazy-init fields
  • Dead code removal: removed unused internal accumulator from binary ECFP path

[6.11.0] - 2026-04-04

Summary

Performance, precision, and depiction release: faster MCS engine, publication-quality SVG renderer (ACS 1996 standard), comprehensive layout engine, 35+ new Python bindings.

Included

  • Core engine: cache-performance improvements on hot MCS computation paths (15-25%)
  • Pre-indexed candidate sets in the MCS solver — eliminates repeated linear scans per frontier atom
  • Publication-quality SVG depiction engine (ACS 1996 standard):
    • Jmol/CPK element colors, asymmetric double bonds, wedge/dash stereo bonds
    • Bond-to-label clipping, H-count subscripts, charge superscripts
    • Full customization via DepictOptions (bond_length, colors, fonts, sizes)
    • Side-by-side MCS pair rendering with atom-atom mapping numbers
  • Multi-phase 2D layout pipeline: template match, ring-first, chain zig-zag, force refinement, overlap resolution, crossing reduction, canonical orientation, bond-length normalisation
  • Distance-geometry 3D coordinate generation with iterative coordinate refinement
  • 40+ ring scaffold templates (pharmaceutical scaffolds, PAH, spiro, bridged)
  • Full 2D/3D coordinate transform suite (translate, rotate, scale, mirror, center, align, project, lift)
  • 35+ new Python bindings with GIL release for thread safety
  • Java: explicit per-atom type matching for robust handling of exotic valence states
  • 9 precision chemistry tests (azulene, pyrene, pyridinium, cyclopentadienyl, boron, sulfoxide, phosphate, E/Z stereo)
  • 27 new layout engine tests (2D/3D generation, transforms, overlaps)
  • Comprehensive Python documentation with examples and cautions

[6.10.2] - 2026-04-03

Summary

Correctness release: fixed MCS connectivity filter for non-induced mode, added regression tests for challenging molecule pairs.

Included

  • Corrected connected-component filter to enforce common-bond reachability in both query and target molecules (non-induced MCS mode)
  • Added GOLDEN_843 regression tests in Python and Java (timeout and size)
  • Version bump to 6.10.2

[6.10.1] - 2026-04-03

Summary

Stability and correctness release: hardened MCS repair pipeline, deterministic tests, CI/CD fixes.

Included

  • Rewrote MCS mapping repair to iterative bounded loop — eliminates unbounded recursion on large molecules (vancomycin, CoA, paclitaxel)
  • Correct duplicate-target handling in mapping repair
  • Removed all timing-dependent test assertions — algorithmic correctness is now fully deterministic and machine-speed independent
  • Removed forkedProcessTimeoutInSeconds from Surefire (was killing fork JVM)
  • Switched Python publish to manual dispatch only (no auto-publish on release)
  • Fixed GitHub Actions artifact version references and Node.js 24 opt-in
  • Fixed atomWeights array length for benzene queries
  • Adjusted MCS thresholds and completeRingsOnly tests for edge cases

[6.9.0] - 2026-04-02

Summary

Core chemistry correctness, native I/O hardening, and benchmark alignment release.

Included

  • Direction-stable native/public MCS handling for hard asymmetric pairs
  • Symmetric ringMatchesRingOnly semantics across C++, Python, and Java
  • Mode-matched benchmark leaderboards with explicit defaults, strict, and ring-only comparison modes
  • Release-documentation cleanup and benchmark/report alignment
  • Native MDL MOL V2000 metadata preservation for molecule name, program line, comment, and SDF properties
  • Native MDL MOL V3000 reader and writer for core graph round-trip
  • Native patent-style R-group molfile support via R# pseudo-atoms and M RGP
  • Stronger SMILES and SMARTS attachment-point handling, including labeled placeholders such as [R1]
  • Additional stereo and CIP round-trip coverage across SMILES and molfile paths
  • Python bindings for native mol block read/write APIs
  • Java MolGraph metadata parity fields for release alignment

[6.8.1] - 2026-04-02

Summary

Release alignment and parity cleanup.

Included

  • Repo-wide version bump
  • Python packaging alignment
  • Java and C++ release metadata alignment
  • Benchmark and documentation refresh

[6.8.0] - 2026-04-01

Summary

Open-source release of the SMSD Pro cheminformatics toolkit.

Included

  • Substructure search engine
  • Maximum common subgraph (MCS) computation
  • Circular fingerprints (ECFP/FCFP) with tautomer awareness
  • SMARTS pattern matching
  • Molecular similarity and screening
  • CIP R/S and E/Z stereodescriptor assignment
  • Batch processing with optional GPU acceleration
  • Java 21+ / C++17 / Python 3.8+ support
  • CLI, SDF batch, and JSON export