Skip to content

Release Notes

All notable changes to KBank are documented on this page. This project adheres to Semantic Versioning and the Keep a Changelog specification.

[Unreleased]

Added

  • CSR Chemical Glycan Graphs & Monosaccharide Registry (kbank.glycan, kbank.sugars):
  • Structure-of-Arrays (SoA) Sugars chemical registry encoding monosaccharide base rings, stereochemistry (Anomer, Chirality), ring forms, and non-stoichiometric substituents.
  • Compressed Sparse Row (CSR) directed multigraphs (Glycan, Glycans) with canonical root-invariant topological hashing and Weisfeiler-Lehman graph kernels.
  • Structural Parsers & Multi-Database Curation (kbank.io, kbank.db):
  • CSDBParser, IUPACParser, and KEGGParser for bidirectional chemical graph parsing and serialization.
  • Multi-database curation pipeline (build_curated_glycan_records) uniting 1,530 bacterial surface polysaccharide structures across CSDB, KEGG, ECODAB, EPS-DB, EK3D, and GlyTouCan paired de novo with SeroGen biosynthetic gene clusters (GlycanRecords).
  • Biosynthetic Alignment & De Novo Structure Prediction (kbank.biosynthesis):
  • BiosyntheticAssigner and PanGraphConsensusPrior Expectation-Maximization homology priors for bipartite enzyme-to-linkage assignment.
  • 10-head factorized 1D-CNN BiosyntheticClassifierSuite (ClusterClassifier) predicting donor, acceptor, anomer, linkage position, and substituent specificity from protein sequences.
  • GlycanPredictor combining RelationalMPNN message passing with a branch-and-bound Constrained Spanning Arborescence solver for de novo repeating-unit structure reconstruction.
  • Interactive Benchmarks & Documentation:
  • Reproducible 16:9 Marimo slide decks (curate_glycans.py, train_cluster_classifier.py, train_glycan_predictor.py) and unified SeroVerse Zensical documentation.

Comments