Data
Versioned dataset inventories, provenance, split definitions, integrity checks and explicit source, target and independent-validation roles.
BioNucleiArchitectureResearch structure
The research architecture of BioNuclei.
BioNuclei is organized as an inspectable chain from data and model computation through cross-domain evaluation, robustness analysis and retained evidence. This page describes that research structure rather than the separate MCP ecosystem.
BioNuclei research structure
Four connected parts of the scientific programme.
Each part has a defined role in the evidence chain. The architecture is a way to organize the research, not a substitute for experimental evidence.
Segmentation model
The scientific model is evaluated as part of a controlled pipeline, with configuration and execution records retained alongside results.
Robustness evaluation
Cross-domain transfer, failure analysis, controlled robustness experiments and prespecified statistical evaluation test what changes when the image domain changes.
Evidence and provenance
Artifacts, metrics, configurations, manifests and verification records provide the traceability needed to inspect each scientific claim.
Capability structure
From raw inputs to an inspectable result.
The research pipeline can be read as explicit stages, with each handoff preserving the information required for reproducibility and audit.
Dataset inventory
Record declared images, labels, versions and acquisition provenance.
Split discipline
Keep source training, target evaluation and independent validation roles explicit.
Model execution
Run the declared segmentation configuration without silently changing the protocol.
Domain evaluation
Measure transfer behavior on declared evaluation populations.
Failure analysis
Characterize where robustness breaks rather than reporting only aggregate performance.
Evidence record
Preserve metrics, parameters, artifacts and verification records for inspection.
Scientific boundary
Computation stays with the scientific pipeline.
BioNuclei's research architecture separates the scientific experiment from future interoperability layers. Models and scientific software perform the analysis; later MCP components may provide interfaces around established capabilities.
Inside BioNuclei
Scientific computation
Dataset handling, segmentation, evaluation, robustness analysis, statistics and evidence generation belong to the research programme.
Separate future layer
MCP interoperability
BioMCP and related MCP directions are separate future interfaces for coordinating established scientific capabilities. They are not the scientific foundation of BioNuclei.
Status
BioNuclei is the scientific programme. MCPs are a separate direction.
The website keeps the boundary explicit: datasets, research and scientific architecture belong to BioNuclei; BioMCP, BioFM, BioWF and BioSkills are presented separately as future interoperability and ecosystem directions.