mobiorigin 0.1.3__tar.gz

mobiorigin-0.1.3/CHANGELOG.md

@@ -0,0 +1,146 @@
+# Changelog
+
+All notable MobiOrigin changes are documented here. The project uses semantic versioning for the standalone `mobiorigin` package interface.
+
+## Unreleased
+
+## 0.1.3 — 2026-08-27
+
+PyPI transport and supply-chain update for the unchanged frozen
+`mobiorigin-dev1-mob-selective-v1` classifier.
+
+### Added
+
+- Atomic, resumable retrieval of the exact frozen dev1 model bundle from a
+  versioned GitHub release asset.
+- SHA-256 and byte-count verification of all three checkpoints, marker
+  normalization, model manifest, and the complete transport archive.
+- PyPI Trusted Publishing workflow with tag/version matching, separate build and
+  publishing jobs, OIDC authentication, attestations, and a fail-closed 100 MB
+  per-file gate.
+- Offline model setup through `--model-archive` and configurable model storage
+  through `MOBIORIGIN_MODEL_DIR`.
+
+### Changed
+
+- Python distributions no longer duplicate the 121 MB frozen model payload.
+  Guided installation retrieves the exact bytes once and verifies them before
+  making them available to prediction.
+- The database helper now prepares and checks model artifacts before marker
+  databases, keeping the standard installation route automatic.
+- `mobiorigin doctor` verifies the resolved model directory as part of the full
+  installation check.
+
+### Scientific boundaries
+
+- Model transport changed; checkpoint bytes, hashes, architecture, feature
+  definitions, ensemble, normalization, threshold, and predictions did not.
+- Prediction remains offline after installation. Missing or changed model bytes
+  stop execution rather than triggering a fallback model or network request.
+
+## 0.1.2 — 2026-08-26
+
+Installation, database automation, and integrated analysis update for the
+unchanged frozen `mobiorigin-dev1-mob-selective-v1` classifier.
+
+### Added
+
+- Guided Conda or Mamba installation with a post-installation doctor check and
+  bundled deterministic demonstration.
+- Automatic, resumable setup and cryptographic verification of the comprehensive
+  annotation resources, with explicit acceptance of applicable third-party terms.
+- mobileOG-db as the default MGE resource, while retaining legacy ISfinder as an
+  optional user-supplied resource.
+- A bundled eight-contig assembly example that demonstrates chromosome, plasmid,
+  phage, and unclassified outputs without presenting an accuracy claim.
+- Deterministic SVG, HTML, and tabular visualizations for prediction and annotation
+  outputs.
+
+### Changed
+
+- `mobiorigin run` now performs prediction, comprehensive biological annotation,
+  and integrated visualization by default in one atomic output directory. A
+  `--skip-annotation` option preserves the lightweight prediction-only route.
+- Expanded `--threads` from the original validated 1–8 range to 1–128 for
+  DIAMOND, AMRFinderPlus, and other external searches. Deterministic neural-network
+  inference remains single-threaded, and the model and scientific policy are unchanged.
+
+### Fixed
+
+- Separated the CPU-only MobiOrigin runtime from MOB-suite's incompatible legacy
+  NumPy/pandas database-building stack.
+- Added a non-overwriting database setup helper with Linux/WSL, Intel macOS, and
+  Apple Silicon/Rosetta handling plus actionable failure messages.
+- Prevented Conda from selecting CUDA by pinning the documented runtime to a
+  cross-platform CPU PyTorch build.
+- Added CI installation smoke tests, an isolated MOB-suite dependency solve, and
+  installation-contract unit tests.
+
+### Scientific boundaries
+
+- The classifier checkpoints, feature definitions, ensemble, selective threshold,
+  and frozen external-validation results are unchanged from version 0.1.1.
+- Biological annotations remain independent supporting evidence. They do not
+  override prediction labels or probabilities and are not clinical risk scores.
+- The bundled assembly is a software demonstration. It is not an accuracy,
+  prevalence, or biological-discovery dataset.
+
+## 0.1.1 — 2026-08-23
+
+Publication and biological-annotation update for the unchanged frozen
+`mobiorigin-dev1-mob-selective-v1` classifier.
+
+### Added
+
+- Prediction-independent `mobiorigin annotate` workflow integrating CARD, SARG,
+  official AMRFinderPlus, VFDB, MGE, BacMet2, and MOB-suite evidence without
+  changing classifier labels or probabilities.
+- Publication-quality annotation tables, provenance, checksums, and HTML reports,
+  including transparent A–E biological evidence-priority tiers that are explicitly
+  not clinical risk scores.
+- Post-hoc exploratory external comparisons with PlasClass, PlasFlow v1, PLASMe,
+  and Platon under a separately frozen statistical contract.
+- Label-free operational evidence from two deterministic real-assembly subsets,
+  covering all 12 dataset–tool runs and 10 pairwise operational comparisons.
+- Validation tables, editable vector figures, methods, limitations, and updated
+  repository documentation.
+
+### Changed
+
+- Package and citation metadata now identify the expanded publication bundle as
+  version 0.1.1.
+- Distribution metadata includes the annotation and operational-validation
+  documentation.
+
+### Scientific boundaries
+
+- The frozen classifier, three model checkpoints, marker normalization, ensemble,
+  and selective threshold are unchanged from version 0.1.0.
+- Secondary comparator findings are exploratory and do not alter the preregistered
+  MobiOrigin-versus-geNomad co-primary evidence.
+- Real-assembly results support runtime, call-rate, coverage, agreement, and
+  biological-evidence reporting only; they do not support ground-truth accuracy or
+  superiority claims.
+
+## 0.1.0 — 2026-08-21
+
+Initial research release of the frozen `mobiorigin-dev1-mob-selective-v1` candidate.
+
+### Added
+
+- Standalone `mobiorigin predict` interface for chromosome, plasmid, phage, and explicit unclassified predictions.
+- Deterministic 9,557-dimensional sequence-feature extraction and 17-dimensional MOB protein-marker extraction.
+- Three frozen neural-network checkpoints combined by an equal-weight softmax mean.
+- Frozen plasmid selective-abstention rule with threshold `0.19835489988327026`.
+- Safe tensor-only checkpoint loading and exact model, normalization, and database identity verification.
+- Atomic prediction outputs with provenance and SHA-256 checksums.
+- `mobiorigin setup-databases` for atomic retrieval or offline installation of the exact marker databases.
+- Prospective external validation against geNomad 1.12.0/database 1.9 using 3,000 source-disjoint records.
+- Aggregate validation tables, vector figure, methods, and claim boundaries.
+
+### Scientific boundaries
+
+- The frozen external cohort is closed to retrospective tuning and record-level error mining.
+- geNomad outputs are not model features or training targets.
+- MobiOrigin does not use hard biological overrides or post-hoc probability transfer.
+- Third-party biological database records are retrieved for local use and are not bundled in the Python distribution.

mobiorigin-0.1.3/CITATION.cff

@@ -0,0 +1,27 @@
+cff-version: 1.2.0
+message: "If you use MobiOrigin, please cite this software release."
+title: "MobiOrigin: sequence-and-marker classification of bacterial replicons"
+type: software
+authors:
+  - family-names: Raza
+    given-names: Shahbaz
+    email: shahbaz.invincible3182@gmail.com
+repository-code: "https://github.com/Raza-pl/MobiOrigin"
+url: "https://github.com/Raza-pl/MobiOrigin"
+version: 0.1.3
+date-released: 2026-08-27
+license: GPL-3.0-only
+keywords:
+  - bioinformatics
+  - plasmid
+  - phage
+  - chromosome
+  - replicon classification
+  - machine learning
+abstract: >-
+  MobiOrigin is a CPU-oriented sequence-and-marker classifier that assigns
+  bacterial DNA fragments to chromosome, plasmid, phage, or an explicit
+  unclassified state. Version 0.1.3 retains the exact frozen three-seed dev1
+  model through an independently verified model bundle and provides an integrated
+  prediction, annotation, and visualization workflow with automatic database
+  setup and publication evidence from prospective and operational comparisons.