Skip to content

Changelog

All notable changes to NeKo are documented here. The format follows Keep a Changelog and version numbers generally follow Semantic Versioning. NeKo 1.9.0 is a transitional release with a small number of explicitly documented breaking changes; version 2.0.0 is reserved for the separately developed backend rewrite.

[1.10.3] – 2026-10-28

Added

  • Network.seed_mapping records what happened to each requested seed: its database symbol (aliases such as FAK map to PTK2) or None when it was dropped. The initial history state stores the same mapping.
  • NetworkVisualizer.render() accepts format= ("pdf" by default; also "svg", "png" or any other Graphviz format).

Fixed

  • Network(genes) without resources loads OmniPath, as documented, instead of failing with TypeError: 'NoneType' object is not subscriptable.
  • Custom interaction tables: text labels in an effect column are parsed by meaning. Before this fix, any non-empty text, including "inhibition", was read as a stimulation, and missing effects were read as inhibitions. Unrecognised effect values now raise ValueError that names them, and missing values are undefined. form complex sets form_complex.
  • Boolean flag columns (is_stimulation, is_inhibition, is_directed, form_complex) accept TRUE/FALSE, t/f and y/n in any case, and raise ValueError for other text instead of reading it as True.
  • Seeds and nodes missing from the database are reported with a UserWarning instead of a printed message, so they can be caught, logged or turned into errors.
  • Documentation: corrected the render() example and the default-resource note in the network-building notebook, and documented the accepted values for custom interaction tables.

[1.10.2] – 2026-10-21

Added

  • Atomic, validated interaction additions and effect changes between existing endpoints, with manual origin, evidence, PMIDs and edit explanations.
  • Structured topology explanations in the existing network history, JSON recovery, and provenance companions for SIF and Boolean exports.
  • Selected database identity on networks and regression coverage for mutation validation, provenance preservation and history consistency.
  • Lightweight history status, retained-state count and paginated summary access.

Changed

  • Avoid repeated full-table serialization during path deduplication and repeated inclusion-reason scans during automatic network growth.
  • Reuse mutation fingerprints and normalize unique edge groups without parallel evidence merging overhead.
  • Capture export history once without copying database resources; write compact JSON companions by default with optional pretty printing.

Fixed

  • Array-like arguments (NumPy arrays, pandas Series/Index, frozensets) no longer make recorded operations fail; unsupported values are recorded by type and repr instead of rejecting the call.
  • A failure while recording history no longer leaves a stale transition context for a later save_state().
  • SIF import treats exported None/nan reference comments as missing and uses the SIF file fallback instead of storing them as references.
  • Deep-copied history tables keep nested provenance when index labels repeat.
  • Networks pickled before history tracking can grow again; their next change becomes the root state.
  • describe_history() and history_digraph() hide internal explanation metadata and show user-supplied metadata again.
  • History and export companion JSON files follow the umask instead of 0600.
  • Multi-variant BNet exports write the history once to <stem>.bnet.history.json; their companions reference it (neko-export schema version 2) instead of each embedding a full copy. Single-model exports keep the self-contained schema version 1 companion.

[1.10.1] – 2026-10-01

Fixed

  • SIF import keeps each edge's # Reference PMID: comment as its reference across export/import round trips. Unannotated interactions keep the SIF file fallback, and a comment is never carried over to another edge.

[1.10.0] – 2026-08-13

NeKo 1.10.0 improves the performance, reproducibility, and configurability of network completion while preserving compatibility with legacy connection parameters through explicit migration warnings.

Added

  • Explicit path_policy values (one_shortest, all_shortest, and all_bounded) for complete_connection.
  • Explicit reuse_policy values (none, discovered_paths, and induced_subgraph).
  • Shortest-path predecessor-DAG selection for the union of all shortest paths.
  • Visible migration warnings with exact replacements for legacy connection parameters.
  • Expanded API documentation and regression coverage for connection policies.

Changed

  • complete_connection now checks the two directed orientations sequentially, allowing newly exposed paths to prevent redundant resource searches.
  • New completion policies require a finite positive maxlen.
  • Resource edge lookup, path insertion, cascade insertion, and induced closure use indexed and batched mutation.
  • Neighbor ordering is stable for reproducible unweighted shortest-path choice.
  • Ontology phenotype and pathway-recreation tutorials are reproducible with the current APIs.

Fixed

  • Restored the semantic distinction between independent searches and reuse of discovered paths.
  • Working Effect="undefined" edges are no longer treated as signed paths.
  • Removed deduplication of stale DataFrame objects after path mutation.
  • History metadata no longer attempts identifier translation for policy strings.
  • Node renaming changes display labels without rewriting canonical identifiers.

Deprecated

  • algorithm, minimal, and connect_with_bias in complete_connection; use path_policy and reuse_policy.

Validation and compatibility

  • The refactor passed 148 deterministic tests, including every legacy BFS/DFS × minimal × connect_with_bias mapping and the corresponding explicit-policy topology.
  • On the pinned 18-gene SIGNOR benchmark, performance-only changes preserved 100% of nodes, signed directed edges, initial seeds, and seed-pair directed reachability for connect_nodes, all_bounded + discovered_paths, connect_subgroup, connect_component(mode="OUT"), upstream connection, and Atopo-complete construction.
  • all_bounded + discovered_paths improved from 113.9 seconds to 5.3 seconds (21.6×), while retaining its 111-node/771-edge topology exactly.
  • Historical BFS selected equal-length routes through unordered set iteration. The deterministic one_shortest policy can therefore choose a different but equally short bridge topology; it retained 100% of initial seeds and directed seed-pair reachability in the reference benchmark.
  • For users who need robustness across all equal-length alternatives, all_shortest + none contained 100% of the historical BFS topology, while all_shortest + discovered_paths retained 97.3% of historical nodes and 95.7% of historical signed edges. Both retained 100% of seeds and directed reachability and remained 6.6–8.0× faster than historical BFS.

[1.9.0] – 2026-07-31

NeKo 1.9.0 is the final major feature release planned for the current backend. Existing tutorial notebooks and the principal Network workflows remain supported. See Upgrading to NeKo 1.9 for migration details.

Added

  • Cache-backed reviewed-human UniProt identifier mapping with bounded, retried downloads and a UniProt REST fallback.
  • A shared cache location controlled through NEKO_CACHE_DIR.
  • Validated SIGNOR database caching and entity dictionaries for complexes, protein families, phenotypes, and stimuli.
  • SIGNOR ChEBI display-name resolution backed by the official ChEBI compound table.
  • Official Gene Ontology API integration with validated pagination, taxon filtering, descendant controls, and optional removal of automatic assertions.
  • Structured GOTerm and GOGene records.
  • Deterministic regression coverage for identifier mapping, SIGNOR, Gene Ontology, tissue mapping, ChEBI, phosphosites, and export behavior.

Changed

  • Replaced PyPath-based identifier translation and removed the pypath-omnipath/Paramiko dependency chain; lightweight pypath-common utilities remain.
  • SIGNOR entities are normalized by default to readable typed identifiers such as COMPLEX:, PROTEIN_FAMILY:, PHENOTYPE:, and STIMULUS:.
  • SIGNOR directness evidence is stored separately from graph direction.
  • Gene Ontology phenotype lookup now uses authoritative GO accessions and the official GO API instead of the former URL-based workflow.
  • Tissue expression queries use validated Human Protein Atlas annotations, with managed caching for the HPA cancer dataset.
  • Duplicate and opposing regulatory evidence is consolidated before export.
  • BNet variant generation is lazy and can be bounded with n=.
  • Runtime dependency constraints support pandas 2 and 3 and current Python releases.
  • Package maturity metadata is now Beta.

Fixed

  • Preserved custom phenotype labels during identifier translation and export.
  • Prevented ambiguous BNet models when distinct labels collide after identifier sanitization.
  • Rejected null, empty, or unknown BNet endpoints with descriptive errors.
  • Created parent directories automatically for BNet and SIF exports.
  • Preserved references while consolidating bimodal regulatory evidence.
  • Normalized phosphosite identifiers consistently.
  • Retried and rejected invalid or truncated SIGNOR downloads.
  • Corrected installed version discovery to query the nekomata distribution.

Breaking changes

  • Removed the documented neko._annotations.gene_ontology.fetch_nodes_from_url helper.
  • Removed Ontology.modify_url_ontology; use Ontology.get_term, Ontology.fetch_go_genes, or Ontology.get_markers with a GO accession.
  • signor() now normalizes proprietary SIGNOR entities by default. Pass normalize_entities=False when raw SIGNOR identifiers are required.
  • BNet export now raises ValueError for identifiers that become ambiguous after sanitization instead of writing an invalid model.

No compatibility wrappers are provided for the removed ontology interfaces.

[1.1.0] – 2025

Added

  • Branching network history with automatic NetworkState snapshots and HTML/SVG rendering
  • NetworkState class for point-in-time network snapshots
  • BFS / DFS graph traversal algorithms in neko.core.algorithms
  • connect_to_upstream_nodes method
  • Performance benchmarking scripts

Changed

  • Interaction lookup tables pre-processed for O(1) neighbour queries
  • connect_nodes signature updated for clarity
  • pandas pinned to 2.2.2 for stability

Fixed

  • Edge colouring bug in NetworkVisualizer.vis_comparison
  • Handling of complex node names containing colons

[1.0.0] – 2024

Added

  • Initial public release on PyPI as nekomata
  • Core Network class with add_node, add_edge, remove_node
  • connect_nodes, connect_subgroup, connect_component, complete_connection
  • connect_genes_to_phenotype via Gene Ontology
  • NetworkVisualizer with Graphviz and yFiles backends
  • Exports class: BNet and SIF formats
  • Ontology class for GO-term and tissue mapping
  • 11 tutorial notebooks
  • Sphinx documentation hosted on GitHub Pages

[0.x] – Pre-release

Internal development versions at Institut Curie / Sysbio-Curie.