DJR-MCP Finder v0.1 — Model V0

DJR-MCP Finder v0.1 — Model V0

This is the first formal GitHub release of DJR-MCP Finder: a reproducible protein-language-model pipeline for classifying user-supplied protein FASTA sequences through the frozen all-ESM-C-6B V0 hierarchy.

Included

• the frozen user-inference-v0 CLI and three checksum-verified linear heads
• bilingual English/Chinese landing documentation with a workflow diagram, quick start, output schema, and interpretation boundaries
• reproducibility metadata through CITATION.cff, frozen model metadata, parity records, and checksums
• a scoped MIT license for project-authored material, plus repository-level and release-specific third-party notices
• GitHub Actions coverage for Python 3.10/3.12 core tests, the formal V0 package, CPU-only CLI smoke checks, and critical Ruff rules

Quick start

cd user-inference-v0
python -m venv .venv
source .venv/bin/activate
python -m pip install -e '.[dev]'
python -m pytest -q
djrmcp-predict validate-fasta examples/synthetic_example.faa
djrmcp-predict model-info

Real inference additionally requires python -m pip install -e '.[inference]', the pinned ESM-C 6B checkpoint, and preferably a CUDA GPU with at least 24 GB of memory. See user-inference-v0/README.md for the full command and offline-cache workflow.

Validation at release

• core suite: 180 passed, 1 skipped
• formal V0 package: 33 passed
• critical Ruff checks: passed
• CPU-only FASTA validation and model-integrity smoke checks: passed
• all GitHub Actions checks on PR #4: passed

Scientific boundary

V0 has no new prospective external Test. Its scores are calibrated under the development-data distribution and are not prevalence-adjusted posterior probabilities in natural proteomes. Large-scale discovery still requires independent false-positive assessment and structural/manual validation.

The separate user-inference-v0.1 directory remains a candidate implementation and is not the formal model released by this tag.