Request Access
Model v22.86 Evidence Cycles.External Cohort set34.Jun 2026

Neurodevelopmental Biomarker Research

Building a reproducible biomarker identifier for rare disorders

Universa Biomarker Intelligence now exposes a governed v24 research-triage scorer backed by adjudicated source features, set41 surveillance, manifest replay, and endpoint readiness controls.

Pass99 final governanceResearch triage ready: set41 v24_sf 99.3% F1 / 100.0% recall / 98.6% specificity @ 0.71

Pass99 final governance approves v24_source_feature_integration for research triage at threshold 0.71; set41 surveillance remains the current audit signal. Clinical use prohibited.

Final sign-offResearch triage ready
Candidate threshold0.71
Manifest replaypassed
Use boundaryClinical use prohibited
Model Readiness APIFinal Sign-offManifest Replay
4131ClinVar Variants
3152GEO Datasets
6868SRA Archives
1929Disease-Target Links
12Reactome Pathways
115Evidence Cycles
Pipeline

Three-stage governed evidence pipeline with full lineage

The platform moves from source ingestion through model training into mandatory external validation, with each pass logged and every model artifact retained for audit.

The current research program addresses a recurring rare-disease problem: candidate biomarkers can look strong in narrow discovery windows, then fail under independent replication. Universa makes that gap visible by pairing training snapshots with disjoint external holdouts before any model-status change.

External cohort gating is mandatory. No model advances without repeated disjoint validation.
  1. 01

    Multi-source evidence aggregation

    ClinVar annotations, GEO expression datasets, SRA archives, Open Targets disease links, and structure-aware context are normalized into reproducible feature rows.

  2. 02

    Versioned classification training

    The latest selected artifact is source_feature_integration_stacked_mlp; governed scoring now uses threshold 0.71, while 0.38 remains visible as a comparability anchor.

  3. 03

    Independent external cohort gating

    The latest cohort, set41, contains 144 rows. Best historical 0.38 performer: v24_sf; final operation is governed by Pass99 readiness checks.

Performance

v24_sf external cohort results

Two validation regimes are reported together for transparency. Training holdout is reference context; external cohort set41 is the active governance signal, and candidate calls use threshold 0.71.

External (set41)Holdout (set11)
F1 Score
99.3%/ 93.3% holdout
Recall
100.0%/ 87.5% holdout
Specificity
98.6%/ 100.0% holdout
Platform Sections

Audit surfaces stay one click from the executive readout

  1. 01

    Executive Findings

    Single combined briefing for the latest validated model posture and external evidence. View findings.

  2. 02

    Data & Sources

    ClinVar, GEO, SRA, Open Targets, external literature pulse, and Docker pipeline outputs. View data.

  3. 03

    Model Registry

    Training snapshots, holdout metrics, and 25 stored model artifacts. Review models.

  4. 04

    Structure Atlas

    Gene-to-structure coverage across UniProt, AlphaFold, and RCSB PDB with an interactive molecular viewer. View atlas.

  5. 05

    Research Intelligence

    3D knowledge graph, API contracts, molecular runtime probes, toolchain workbench, and artifact launchpad. Explore.

Current Research Focus

Residual monitoring after a zero-FN set41 gate

Set41 left no false negatives at the governed threshold. Monitoring now concentrates on recurring hard-boundary positives and false positives before any set42 expansion.

PTPN14MAP3K7

For research triage and prioritization only. Not for standalone clinical diagnosis.

Open the latest research brief

Review combined findings, model lineage, and cohort-level evidence from the current cycle.

Findings BriefJSON API