biors 0.54.0

Command-line tools for bio-rs biological AI model input workflows.
biors-0.54.0 is not a library.

bio-rs

CI Release Benchmark Contracts License: MIT/Apache-2.0

bio-rs is a local, reproducible input layer for bio-AI tools. It takes raw biological sequences, structure files, molecule files, tokenizer profiles, package manifests, model artifacts, and fixture outputs, then turns them into checked contracts that can run in CLIs, CI, Python notebooks, browsers, and agent tools.

raw sequence data + package metadata
  -> validation diagnostics + provenance
  -> tokenizer IDs, model-input tensors, pipeline locks, package checks
  -> JSON contracts for Rust, CLI, Python, WASM, MCP, and service hosts

Protein, DNA, and RNA validation, tokenization, model-input generation, workflow generation, Python/WASM/MCP bindings, package artifact validation, and benchmark regression guards are supported through explicit sequence-kind profiles. Package generation and Python/Hugging Face conversion remain protein-first; see the sequence-kind support matrix before making broad full-support claims.

Status: pre-1.0 CLI and JSON contract stabilization.

Current boundaries

bio-rs is not a hosted model registry, a model training framework, or a remote inference service. It does not upload biological data or package artifacts by default. Current binary archives are limited to the platforms documented in docs/install.md; other platforms should use source builds until release artifacts are added. Package trust is local validation, checksums, schemas, and reproducible lockfiles, not a remote signing or review service.

Why bio-rs?

Bio research code often starts as a notebook and then has to survive handoff: to a CLI, a CI job, a browser demo, an agent, a package fixture, or a service owned by someone else. The brittle part is usually not the model. It is the input contract around the model: what was parsed, how residues were normalized, which symbols were warnings, which tokenizer IDs were emitted, whether the model input is actually safe to run, and which artifacts were used.

bio-rs makes that layer explicit:

  • validate protein, DNA, and RNA with stable diagnostics instead of hidden string cleanup
  • preserve token positions with known profile IDs instead of shortening inputs silently
  • emit model-input JSON only when unresolved residues have been handled
  • pin provenance, vocabulary hashes, output hashes, package checksums, and pipeline lockfiles for repeatable runs
  • expose the same deterministic core through Rust, the CLI, Python, WASM, MCP, and local-first service schemas

The goal is not to replace Python research workflows. The goal is to give those workflows a portable contract layer that is easy to inspect, test, cite, and share.

Made For Sharing

The project is built for open-source bio-AI work where examples need to be reproducible and claims need evidence:

  • Researchers can share a small FASTA, command, and JSON contract instead of a screenshot from a notebook.
  • Model/package authors can ship fixture outputs, checksums, citations, model cards, and runtime bridge plans next to their artifacts.
  • Tool builders can embed the same preprocessing behavior in CLIs, web demos, local agents, and internal services without reimplementing the parser.
  • Maintainers can point every public claim at schemas, tests, package artifact checks, and benchmark regression guards.

Quickstart

cargo install biors --version 0.54.0
biors doctor
biors seq validate --kind auto examples/multi.fasta
printf '@read1\nACGN\n+\n!!!!\n' | biors formats validate --format fastq -
printf 'CC(=O)O acetate\n' | biors molecule validate --format smiles -
printf '>dna\nACGT\n' | biors workflow --profile dna-iupac --max-length 128 -
biors batch validate --kind auto examples/
biors tokenizer inspect --profile protein-20-special
biors package validate examples/protein-package/manifest.json
biors service contract
# in another terminal: biors serve --host 127.0.0.1 --port 8787
biors dataset inspect --source uniprot --version 2026_02 --split train examples/

Full commands, demos, and install options: docs/quickstart.md

Proof

bio-rs keeps performance claims tied to reproducible in-repo benchmarks.

Current release posture: no current-version numeric throughput claim is made for 0.54.0. The committed benchmark artifact is historical performance evidence from biors-core v0.20.0; rerun and commit a fresh artifact before using these numbers as evidence for a later release.

The current code includes Criterion regression guards for fixed-length model-input construction and selected backend smoke paths, but those implementation changes are not represented by the historical FASTA table below. The benchmark artifact records which promoted surfaces have committed numeric coverage and which are explicit non-claims.

Historical FASTA benchmark reference

Historical FASTA benchmark baseline (recorded on biors-core v0.20.0; not current-version performance evidence):

Dataset Matched workload bio-rs core mean Biopython mean bio-rs speedup
Human proteome Parse + validation 0.036s 0.584s 16.09x
Human proteome Parse + tokenization 0.061s 0.587s 9.68x
100MB+ FASTA Parse + validation 0.294s 3.994s 13.59x
100MB+ FASTA Parse + tokenization 0.492s 4.040s 8.22x
Many short records Parse + validation 0.007s 0.204s 28.35x
Many short records Parse + tokenization 0.010s 0.205s 20.54x
Single long sequence Parse + validation 0.005s 0.176s 34.48x
Single long sequence Parse + tokenization 0.007s 0.177s 26.67x

Benchmark details:

  • Datasets:
    • UniProt human reference proteome (UP000005640, 9606)
    • 100MB+ large FASTA generated by repeating the same real proteome to isolate large-input throughput
    • 20,000 short 48-residue records generated from the same proteome residue stream
    • one 960,000-residue sequence generated from the same proteome residue stream
  • Matched workloads:
    • pure parse
    • parse plus validation
    • parse plus tokenization
  • Current best recorded raw throughput:
    • human proteome parse + validation: 315.4M residues/s, 360.6 MB/s
    • 100MB+ FASTA parse + validation: 350.8M residues/s, 401.1 MB/s
    • human proteome parse + tokenization: 189.0M residues/s, 216.1 MB/s
    • 100MB+ FASTA parse + tokenization: 209.7M residues/s, 239.8 MB/s
  • Benchmark doc: benchmarks/fasta_vs_biopython.md
  • Benchmark script: scripts/benchmark_fasta_vs_biopython.py

This benchmark measures biors-core directly and excludes CLI startup and JSON serialization overhead. It is still workload-specific, not a broad claim that bio-rs is faster than Biopython across every FASTA workload or researcher input shape. Until the artifact is refreshed for 0.54.0, the numeric table above remains a historical reference.

What Works Today

biors-core provides the Rust engine and data contracts. biors provides the CLI surface.

Sequence handling

  • FASTA parsing and normalization with buffered reader APIs
  • FASTQ parsing and validation with shared format metadata, sequence/quality length checks, Phred+33 quality character validation, and DNA IUPAC sequence diagnostics
  • Protein/DNA/RNA validation with per-record kind detection (--kind auto)
  • Line and record-index diagnostics with residue warning/error reporting

Structure handling

  • PDB fixed-column ATOM/HETATM parsing with StructureRecord, Chain, Residue3D, Atom, and Coordinate contracts
  • Chain extraction, REMARK 465 missing-residue preservation, coordinate validation, occupancy checks, and missing element warnings
  • Coordinate-derived protein sequence extraction and SEQRES mapping through biors structure validate --format pdb and biors structure sequence --format pdb
  • mmCIF is reviewed as the next structure parser candidate but is not exposed as executable parser support yet

Molecule handling

  • SMILES parsing with branch, ring-closure, bracket atom, bond-order, and disconnected component validation
  • SDF V2000/basic V3000 parsing with coordinates and data-item preservation
  • MOL2 parsing with atom types, partial charges, and substructure metadata
  • MoleculeRecord, AtomGraph, BondGraph, and MoleculeMetadata contracts
  • Conservative valence validation, deterministic canonical graph keys, formula and mass descriptors, simple drug-discovery descriptors, and biors-ecfp-lite-v0 hashed fingerprints
  • CLI support through biors molecule validate --format smiles|sdf|mol2 and biors molecule inspect --format smiles|sdf|mol2

Tokenization

  • protein-20 tokenization with stable IDs
  • protein-20-special tokenization with UNK/PAD/CLS/SEP/MASK special tokens
  • dna-iupac and rna-iupac tokenization with stable canonical base IDs
  • dna-iupac-special and rna-iupac-special tokenization with UNK/PAD/CLS/SEP/MASK special tokens
  • JSON tokenizer config loading and inspection
  • Hugging Face tokenizer config conversion
  • Positional token alignment preserved with explicit unknown-token IDs

Model input

  • model-input CLI: profile-aware input_ids, attention_mask, and truncation metadata for protein, DNA, and RNA token profiles
  • workflow CLI: profile-aware validation → tokenization → model input with readiness issues and reproducibility provenance
  • pipeline CLI: no-config validate → tokenize → export, or config-driven (TOML/JSON) workflows with lockfile generation
  • debug CLI: step-by-step per-record inspection with compact residue markers
  • Checked and unchecked model-input builders with safety checks for unresolved residues
  • Python, WASM, MCP, package artifact validation, and regression benchmarks cover nucleotide model-ready workflows. Package skeleton/conversion helpers remain protein-first; see Protein, DNA, and RNA support.

Batch and dataset operations

  • batch validate: multiple files, recursive directories, quoted globs
  • dataset inspect: dataset descriptors, sample mapping, file SHA-256 provenance
  • cache inspect and guarded cache clean for local artifact store

Package management

  • Manifest inspection, validation, and migration (v0 → v1)
  • Schema compatibility checks and canonical diffs
  • SHA-256 checksum verification and fixture verification
  • Python project to bio-rs package skeleton conversion
  • Runtime bridge planning reports, backend execution abstraction contracts, and guarded external-process backend adapters
  • Optional Candle backend crate for CPU safetensors linear-probe inference
  • Model artifact metadata and runtime/model compatibility checks in package bridge reports
  • Local-first biors serve HTTP runtime plus service interface contracts for service hosts
  • Typed validation issue codes and manifest enums

External interfaces

  • biors-python: PyO3 bindings for Python integration and notebook workflows
  • biors-wasm: WebAssembly/JavaScript bindings with TypeScript definitions, browser file validation/tokenization helpers, and local-only browser demo
  • biors-mcp-server: local MCP server crate for agent-callable sequence tools
  • service contract: JSON route/schema contract for caller-owned service hosts
  • biors serve: local-first HTTP API for health, OpenAPI, and inline FASTA batch validation

Utilities

  • diff: canonical JSON/raw comparison with SHA-256 hashes
  • doctor: core CLI, WASM, Python, package, release, and benchmark readiness
  • completions: shell completion generation
  • JSON success/error envelopes for all commands

Documentation

Not yet

These are roadmap directions, not current capabilities:

  • hosted web workflows
  • pretrained model-specific inference backends
  • package registry or plugin ecosystem
  • general-purpose chemistry tooling
  • mmCIF structure parsing beyond reviewed candidate requirements
  • tautomer normalization, force-field chemistry, conformer generation, and RDKit/Open Babel canonical SMILES equivalence
  • no-code or low-code workflows
  • hosted workflow workspace management

Development

Run checks:

scripts/check.sh

Run the faster local commit gate:

scripts/check-fast.sh

The check suite runs:

  • cargo fmt
  • shell and Python syntax checks for repo scripts
  • benchmark Markdown regeneration check
  • release workflow publish-order invariant check
  • Rust checks
  • biors-core wasm32-unknown-unknown build check
  • tests
  • cargo clippy with warnings denied

Reproduce the FASTA benchmark:

cargo build --release -p biors-core --example benchmark_fasta
python3 -m venv .venv-bench
. .venv-bench/bin/activate
pip install biopython
python scripts/benchmark_fasta_vs_biopython.py
cat benchmarks/fasta_vs_biopython.json

The benchmark script updates both benchmarks/fasta_vs_biopython.json and benchmarks/fasta_vs_biopython.md. scripts/check-benchmark-docs.sh verifies that the Markdown report still matches the JSON artifact.

Compare two benchmark artifacts:

python scripts/compare-benchmark-artifacts.py before.json after.json

Run the Rust library example:

cargo run -p biors-core --example tokenize

Repository Map

Most users only need the biors CLI or one binding package. This map is for contributors and integrators who need to understand where public surfaces live.

packages/
  rust/
    biors/                 CLI
    biors-backend-candle/  Optional Candle runtime backend
    biors-core/            Core engine + contracts
    biors-mcp-server/      Local MCP server
    biors-python/          PyO3 bindings
    biors-wasm/            WASM/JS bindings

schemas/
  JSON contracts for CLI, package, pipeline, service, tokenizer, and workflow outputs

examples/
  protein.fasta
  multi.fasta
  model-input-contract/
    protein.fasta
    protein-20-special.config.json
    protein-20-special.expected.json
    reference-python-parity.json
  python/
    esm_from_biors_json.py
    pandas_numpy_friendly.py
    protbert_from_biors_json.py
    reference_preprocess.py
  protein-package/
    models/
    docs/
    manifest.json
    observations.json
    fixtures/
    observed/
    tokenizers/
    vocabs/
    pipelines/
  pipeline/
    protein.toml
    protein.yaml
    protein.json
    pipeline.lock

The full schema inventory and command-to-schema mapping live in docs/cli-contract.md. Release tests keep that contract inventory aligned with the files under schemas.

Protein-20 alphabet

A C D E F G H I K L M N P Q R S T V W Y

Token IDs follow that order, starting at 0.

Contributing

See CONTRIBUTING.md for local setup, checks, and PR expectations.

License

Dual licensed under MIT OR Apache-2.0. If you use bio-rs in research software or publications, cite the repository and version via CITATION.cff.