Skip to content

Repository files navigation

BRIDGE

Development-aware transcriptomic evidence for cell-therapy products.

Parkinson's disease hPSC-derived midbrain dopaminergic (hPSC-mDA) products are the first biological instance.

BRIDGE connects human fetal references and a pre-transplant cell product to five evidence domains and traceable outputs.

Biological question

Given a pre-transplant cell product, which intended developmental and regional identities are supported by its transcriptome, where does the full composition diverge, and which uncertainties should be tested next?

Five evidence domains

Domain Evidence question
Target identity Are the intended lineage and cell states represented?
Regional fidelity Does the product support the intended anatomical identity rather than an off-axis fate?
Developmental compatibility How does the product align with a researcher-defined developmental window?
Off-target control What target, adjacent, off-axis, and unresolved states make up the whole product?
Proliferation & Stress Response Without reassigning cell identity or composition, which stage-conditioned proliferation, stress-response, death-associated, or residual pluripotency-like transcriptomic signals require review?

Current biological progress

The first pilot asked whether fetal ventral-midbrain references can identify intended states in a pre-transplant product while refusing unrelated neural and non-neural cells.

Biological question Data examined Current finding Meaning for product evaluation
Can broad fetal ventral-midbrain states be recognized? Donor-aware Chen vMB scRNA-seq splits Several methods recover broad states, with uneven performance across labels Exploratory state composition is possible, but no state is released for formal reporting
Can fine RG/Nb-derived states be separated? Seven priority L2 states Some methods separate these states internally, but external support and marker review are incomplete Fine regional or developmental claims remain unavailable
Can unrelated cells be rejected? Cortical organoid, neural crest, motor-neuron and mesenchymal OOD data Tested inductive methods can force these cells into known ventral-midbrain labels Formal target, regional-fidelity and off-target conclusions are blocked
Can markers provide an independent check? Internal marker/program cards Negative-marker coverage is incomplete and all seven L2 marker cards remain unfrozen Marker evidence remains a shadow interpretation channel

The next scientific step is the P0-02 External-Source Freeze Candidate: review the 25 state definitions and marker cards together with the ProductDefinitionCard and StateRoleMap, then sign the FreezeGate before any locked runner is implemented or run. Unresolved Nb boundaries remain provisional or unavailable.

Repository status

Package Status
P0-01 Input Audit & QC Executable candidate
P0-02 Cell-State Evidence Executable shadow; no state or method frozen
P0-08 Evidence Sufficiency Executable deterministic candidate over versioned upstream evidence; no score or real-case conclusion
P0-09 Evidence Compiler & Reconciler Executable deterministic candidate for immutable evidence graphs and bounded read-only queries; no score or claim verification
P0-03–P0-07, P0-10–P0-12 Scientific contracts only; executors pending

Minimal usage

python -m pip install -e ".[qc,evidence]"
bridge-tool list
bridge-tool describe P0-02
bridge-tool validate --request request.json
bridge-tool run --request request.json

Each run preserves the input and records the applicable measurement, reference, method, artifact, and checksum provenance.

Scientific boundaries

  • BRIDGE reports research-use transcriptomic evidence, uncertainty, and evidence gaps.
  • Missing, unresolved, and out-of-reference evidence is not treated as a negative result.
  • Candidate or shadow evidence does not establish clinical efficacy, safety, potency, GMP release, or an absolute product ranking.
  • Post-transplant graft evidence is analyzed independently and is not back-propagated into the pre-transplant profile.
  • No frozen P0 domain score is currently published.

Documentation