Research update: Two candidates independently replicated in MSKCC cohort (GSE21034). Deeper analysis: 8 ciliary motor genes — 0 prostate cancer papers. Second candidate overexpressed without its 10 interaction partners. Full analysis →

Research overview

E-DICE-R applied to multi-omics discovery.

E-DICE-R is a live, authority-governed AI platform running cancer biomarker discovery workflows on public multi-omics data. Two candidates were identified and replicated in an independent cohort. A deeper analysis has since revealed an entire unstudied ciliary pathway — and a convergent evidence signal pointing toward invasive biology.

Live platform status E-DICE-R running end-to-end cancer biomarker discovery workflows on secure cloud infrastructure.
Current research phase MSKCC replication confirmed. Deeper analysis: ciliary pathway gap + candidate isolation + convergence signal identified.
Evidence standard Six independent molecular evidence streams evaluated under full governance controls — no human steering.
Partner readiness Structured, reproducible validation paths ready for research and enterprise engagement.
19,010
Genes screened — autonomous governed workflow
3.7×
Tumor overexpression — lead signal
48/48
Safety tests passing — live platform
SPIE ’26
Lead-authored peer-reviewed publication
Peer-reviewed publication

SPIE 2026 — Defense + Commercial Sensing.

E-DICE-R's foundational research is positioned through SPIE 2026 and helps frame the platform's governance architecture and discovery methodology.

SPIE
2026
Lead-authored  ·  Peer-reviewed  ·  SPIE Defense + Commercial Sensing 2026
Authority-Governed AI for Multi-Omics Discovery: E-DICE-R and the E-DICE Hybrid Orchestration Framework
George Soto, MBA  ·  Lead Author  ·  Infinity Software Architects, Inc.
Governed AI Authority computation Multi-omics Biomarker discovery Hybrid orchestration E-DICE-R framework
GOV
Authority-governed AI architecture
The paper establishes the theoretical basis for authority computation — a governance primitive that enforces deterministic execution, policy compliance, and trust constraints at every step of an AI workflow.
  • Authority computation and deterministic execution control
  • Governed optimization (EGAO) for multi-step workflows
  • Policy enforcement without human override
BIO
Multi-omics discovery methodology
Documents the six-evidence-type screening workflow applied to TCGA prostate cancer data, including the governance gates that surfaced the lead candidate without any human guidance.
  • Six molecular evidence streams under governance control
  • TCGA prostate cancer dataset — 19,010 genes
  • Zero human guidance in candidate selection
EDC
E-DICE hybrid orchestration framework
Introduces E-DICE — the Execution-Directed, Integrity-Controlled, Emergence framework — for coordinating classical, quantum-inspired, and quantum execution under unified authority governance.
  • Classical + quantum-inspired + quantum coordination
  • Integrity control and emergence governance
  • Federated execution with trust constraints
VAL
Validation and reproducibility
Establishes the reproducibility standard for governed AI discovery — demonstrating that governance-first systems can produce consistent, defensible outputs that hold up under scientific scrutiny.
  • 100% reproducibility across tested dataset cohorts
  • 48/48 safety tests passing on live platform
  • Public TCGA data — no proprietary data used
First governed AI signal

A prostate cancer signal no prior research had published.

E-DICE-R's governed workflow screened 19,010 genes across six evidence types and surfaced a statistically strong signal with no prior cancer publications found during review.

First successful governed discovery

Strong stats. Perfect reproducibility. Zero prior publications.

This is what governance-first AI can do — surface biology that conventional approaches miss, from public data alone, with no human candidate guidance and full execution control throughout.

3.7× tumor overexpression p = 2.57×10⁻²¹ 0 cancer publications found 100% reproducibility TCGA public data
Full discovery report →
19,010
Genes screened — governed workflow
Six evidence types evaluated under governance
01  ·  Transcriptomics
Gene expression
Differential expression — tumor vs. normal tissue. Minimum fold-change gate enforced by governance layer.
02  ·  Genomics
Mutation profiling
Somatic mutation frequency and pattern analysis across prostate cancer samples for oncogenic consistency.
03  ·  Clinical
Survival analysis
Kaplan-Meier and Cox modeling across TCGA cohorts. High expression associated with patient outcome differences.
04  ·  Validation
Reproducibility
Cross-dataset consistency required before literature review. Governance enforces 100% reproducibility gate.
05  ·  Literature
Database review
Automated cross-reference against PubMed, CancerGene, COSMIC. Lead candidate — zero cancer publications found.
06  ·  Druggability
Structural assessment
AlphaFold structure prediction + binding pocket detection + druggability scoring. Currently in validation.
Deeper analysis — April 2026

After replication, a broader pattern. An entire pathway unstudied.

MSKCC replication confirmed both candidates. A systematic literature search then revealed something more significant: This candidate's entire protein family — 8 canonical ciliary motor genes — has no published prostate cancer research. The second candidate overexpresses without its 10 known interaction partners. The lead candidate and its pathway partner converge through independent evidence paths toward cytoskeletal biology.

Stage 1 — Complete
Governed discovery
Candidate screening — TCGA
Stage 2 — Complete
MSKCC replication
Both genes replicated independently
Stage 3 — New finding
Ciliary gap + isolation
8 genes, 0 papers
Stage 3 — New finding
PRR36 / ARHGEF38
Convergent paths, cytoskeletal signal
New finding — April 2026

An entire ciliary pathway. Zero papers. Second candidate overexpressed alone, in isolation.

The second candidate's entire protein family — 8 canonical ciliary motor genes — has no indexed prostate cancer publications. Every gene returns zero. This is not one unstudied gene; it is an unstudied biological pathway. And This candidate overexpresses without any of its 10 known interaction partners present. Only multi-evidence governed screening surfaces this type of pattern.

8 ciliary genes — 0 prostate cancer papers Second candidate isolated — 10 partners absent from 942 genes Two evidence paths converge independently MSKCC replicated — both genes confirmed
Full analysis on discovery page →
8
Ciliary genes
0 papers
DNAH5 — Expression Pattern
Overexpressed in isolation from its protein complex
This candidate has 10 high-confidence STRING interaction partners (scores 0.935–0.997), all ciliary components. None appear in the 942 passing genes. Two testable interpretations:
  • Moonlighting: possible non-ciliary function in prostate tumor cells
  • Pathway disruption: ciliary signaling dysregulated in a cancer-specific way
  • Biological role uncharacterized — testable in wet lab
Candidate Convergence
Two independent evidence paths converge on the same gene pair
The lead candidate's sole STRING partner is ARHGEF38 — a Rho GEF cytoskeletal regulator. ARHGEF38 independently appears in the 942 passing genes through a separate evidence path. Two unrelated paths converging is unlikely by chance:
  • Points toward cytoskeletal regulation and cell motility biology
  • Mechanistically linked to invasion and metastasis pathways
  • Most actionable functional lead from the current analysis
Preliminary interpretation — not a clinical claim

What these signals may suggest — and what they do not yet prove

Lead pathway
Points toward cytoskeletal control and cell movement — biology mechanistically tied to invasion, metastasis, and progression. If these genes track with aggressive disease, they could inform earlier detection of high-risk patients, treatment decisions, and new drug targets in cytoskeletal pathways. Consistent with a more invasive phenotype — not yet clinically demonstrated.
Isolated overexpression
May reflect cellular stress, altered intracellular transport, or dedifferentiation — patterns seen in aggressive tumors, but less mechanistically direct. If this candidate is being repurposed in prostate tumors, it could represent a novel therapeutic target distinct from common pathways. Biological role uncharacterized — wet lab needed.
!
Current scientific framing: "Consistent with a more aggressive biology — not yet proven." Association with Gleason score, metastatic staging, or survival outcomes requires separate validation. These are mechanistic observations from multi-omics data.
Discovery methodology

How the governed workflow found the lead candidate.

The E-DICE-R governance layer executed an autonomous multi-omics screen under quality gates, without pre-selected hypotheses or manual steering.

01
Full dataset ingestion — 19,010 genes
TCGA prostate cancer RNA-seq data loaded into the governed workflow. No pre-filtering, no hypothesis about which genes to prioritize. Every gene in the public dataset enters the pipeline equally.
→ TCGA public data — no proprietary data used
02
Governance gate 1 — differential expression
Tumor vs. normal tissue expression analysis. Governance enforces a minimum fold-change threshold. Candidates that do not meet the statistical threshold are eliminated without human review. 95%+ filtered at this stage.
→ 95%+ candidates eliminated — no human override
03
Multi-evidence scoring — mutation, survival, reproducibility
Remaining candidates evaluated across somatic mutation profiling, Kaplan-Meier survival analysis, and cross-dataset reproducibility. All three gates must pass. The governance layer enforces each gate independently.
→ 942 candidates advanced to literature review
04
Automated literature review — novelty filter
Cross-reference against PubMed, CancerGene, COSMIC, and primary cancer databases. Candidates with extensive prior publication are deprioritized — novelty is valued. The lead candidate returned zero cancer publications at time of review.
→ Lead candidate — 0 cancer publications found
05
Structural druggability assessment
AlphaFold-based protein structure prediction, binding pocket detection, and druggability scoring. This stage evaluates whether the candidate has a realistic path to therapeutic development.
→ Currently in validation
06
Lead candidate identified
PRR36 (Proline Rich Region 36) emerged as the lead candidate: 3.7× tumor overexpression, p = 2.57×10⁻²¹, 100% reproducibility, zero cancer publications. The result of a fully governed, zero-human-guidance AI discovery workflow.
→ First successful governed AI discovery
Method and process — our proprietary IP

We claim the method that identifies, validates, and governs its use.

E-DICE-R is a proprietary governed multi-omics pipeline. The IP is the structured, reproducible, authority-scored process — not the gene. PRR36 is a result produced by that method. The method is what we sell. The method is what we protect.

E-DICE-R — Governed Pipeline

The method and process

E-DICE-R converts multi-omics discovery into a governed, repeatable decision system for identifying and prioritizing therapeutic and diagnostic targets. The invention defines a structured pipeline that ingests heterogeneous data, applies authority-based scoring and validation, and produces ranked, auditable outputs.

  • Method: defined pipeline for ingesting, processing, scoring, and ranking targets from multi-omics data
  • Decision engine: authority-based filtering, evidence weighting, and cross-cohort validation
  • Outputs: ranked candidates for diagnostics, biomarkers, and therapeutic targeting
  • System: architecture enforcing reproducibility, auditability, and deterministic workflows
  • Pharma linkage: integration into target selection, validation, and drug development workflows
Lead Candidate — Exploratory Signal

What our method found

An early governed discovery run surfaced the lead candidate as a statistically strong exploratory find from a 19,010-gene prostate cancer screen — independently replicated in the MSKCC cohort (GSE21034). The full methodology, signal details, and interpretation scope are documented in the discovery summary.

View discovery summary →
E-DICE-R invention position

E-DICE-R turns multi-omics discovery into a governed decision system.

The invention is defined across three interlocking claims — method, system, and pharma workflow — each protecting a distinct layer of the pipeline. Together they form a complete, defensible IP position around the governed discovery process.

M
Method claim
A governed method for ingesting, processing, scoring, and ranking targets from multi-omics data with authority-based scoring and non-compensating validation gates.
  • Structured target-identification pipeline
  • Evidence weighting and cross-cohort validation
  • Ranked outputs for diagnostics, biomarkers, and therapeutic targeting
S
System claim
A deterministic, auditable architecture that enforces governance, maintains a complete evidence chain, and guarantees replayable outputs.
  • Reproducibility and auditability by design
  • Deterministic workflows and evidence traceability
  • Architecture-level protection around the ranking engine
P
Pharma workflow claim
A pharmaceutical workflow layer that connects governed outputs to target selection, validation tracking, and downstream drug-development decision support.
  • Structured evidence packages for portfolio review
  • Target selection and validation tracking
  • Application-level use of identified targets
SDK
E-DICE-Edge · E-DICE-Edge Partner Vault
Deployable molecular governance for partner pipelines
A programmable governance layer for pharmaceutical and multi-omics workflows
Authority-based scoring Deterministic audit workflows External dataset integration Partner Vault

E-DICE-Edge SDK exposes the molecular governance layer of E-DICE-R and E-DICE-R as a deployable integration for external pipelines. Through E-DICE-Edge Partner Vault, partners can apply authority-based scoring, non-compensating validation gates, and deterministic audit workflows directly to their own datasets.

Research partner access

Replicated. Deeper pattern found. Wet lab is next.

Both candidates independently replicated in the MSKCC cohort. A deeper analysis surfaces an unstudied ciliary pathway and a convergent cytoskeletal signal. If you are a research partner, clinical collaborator, or investor, now is the moment to engage.

Full discovery report Request investor brochure Contact George directly
Live cloud deployment
SPIE 2026 peer-reviewed
MSKCC independently replicated
Ciliary gap — 8 genes, 0 papers