ChromBPNet Remote Tool (MCP Server)#
TOU validation and deployment status (2026-08-16)#
TensorFlow safely loaded a generated Keras-v3 fixture; loopback prediction and variant-effect calls returned bounded finite contract results. No reviewed trained ChromBPNet artifact was available, so scientific inference, public publication, cross-user isolation, and production performance remain unvalidated. Authenticated private Platform import and owner testing passed on 2026-08-16; public publication and independent-caller authorization/isolation remain untested.
Operations:
run_chrombpnet_predict,run_chrombpnet_variant_effectStart:
python -m tooluniverse.remote.chrombpnet.chrombpnet_toolEndpoint:
http://127.0.0.1:8032/mcpProvider configuration: set
CHROMBPNET_MODEL_PATHto one administrator-reviewed.kerasartifact. Caller-selected paths and legacy.h5loading are rejected.TOU check:
tu doctor --forward http://127.0.0.1:8032/mcp --jsonPrivate relay:
tu serve --share --forward http://127.0.0.1:8032/mcp --name validation-chrombpnet --workers 1
Non-loopback binding requires TOOLUNIVERSE_API_TOKEN; otherwise keep the server on loopback. The relay requires TOOLUNIVERSE_SERVICE_KEY. The current result is an artifact-contract pass, not a trained-model pass.
New-user check: python scripts/remote_validation/setup_skill_preflight.py --implementation chrombpnet. Add --check-provider-env before launch, --live after launch, and --check-connect-prereqs before sharing. Live preflight checks exact MCP discovery only; it does not run or validate a model. The pinned relay SDK is not on PyPI and currently requires authorized GitHub repository access plus a configured SSH key, so a working local MCP server does not by itself prove that a new operator can share it.
The authenticated 2026-08-16 Platform matrix found all 30 private owner relays online and all 41 operations discoverable. This implementation was imported as unpublished owner draft(s), configured with a 120-second timeout, and invoked through /expert-sessions/{id}/test. Across the set, 38 unique operations passed return-schema and semantic validation; the three USPTO operations returned exact provider HTTP 403 and remain credential-blocked. Public publication, independent-caller authorization/isolation, broad saturation, and persistent supervision were not tested.
See the complete setup and verification guide.
Serves ChromBPNet (Pampari et al., Nature Methods 2025) — base-resolution, bias-corrected deep learning of chromatin accessibility from DNA sequence — as the ToolUniverse remote tools run_chrombpnet_predict and run_chrombpnet_variant_effect.
ChromBPNet predicts ATAC-seq/DNase-seq accessibility from a 2,114 bp sequence with the Tn5/DNase enzyme bias regressed out. It is the modern successor to DeepSEA/Basset for non-coding regulatory variant interpretation (GWAS/eQTL fine-mapping) and TF-motif discovery, and underlies the ENCODE accessibility model zoo. The model has two output heads: a 1,000 bp accessibility profile (shape) and a scalar log total count (magnitude).
Note on DeepSEA: the classic DeepSEA model (and HumanBase/FUMA front-ends) is browser-only with no maintained programmatic API. ChromBPNet is the maintained, installable, bias-corrected equivalent and is what this tool wraps.
Served remotely because it carries a heavy TensorFlow/Keras stack and requires a trained, cell-type-specific model. The provider selects one reviewed Keras v3 .keras artifact with CHROMBPNET_MODEL_PATH; callers cannot select model files.
Operations#
run_chrombpnet_predict— predicted accessibility (log total counts + base-resolution profile) for one sequence.run_chrombpnet_variant_effect— ref-vs-alt count log2 fold-change (magnitude effect) + profile Jensen-Shannon divergence (shape effect), the canonical ChromBPNet variant scores.
Models#
Trained, cell-type-specific models live in the HF ENCODE ChromBPNet zoo — e.g. kundajelab/encode-chrombpnet-DNASE-ENCSR000EMK-ENCSR816AQM. Those releases use legacy Keras .h5 files, which this remotely callable service intentionally does not deserialize. A provider may review and convert a trusted model to the Keras v3 format in an isolated administrative workflow, then configure the resulting artifact. Unreviewed legacy models must not be loaded merely to convert them.
Deploy#
pip install -r requirements.txt # TensorFlow/Keras 3 + NumPy
export CHROMBPNET_MODEL_PATH=/provider/models/reviewed-model.keras
python chrombpnet_tool.py # starts the MCP server on 127.0.0.1:8032
The server requires .keras, invokes load_model(..., safe_mode=True), and fails closed for legacy or incompatible artifacts. GPU is recommended. Expose remotely only behind TOOLUNIVERSE_API_TOKEN (SMCP bind guard). Scientific execution still requires validation against a reviewed converted model; no such model was available in the current validation environment.
Register in ToolUniverse#
Tool definition: src/tooluniverse/data/remote_tools/chrombpnet_tools.json
(type: RemoteTool).