DNAI
Research use only. Designed for hypothesis generation and translational research.
For Translational Researchers

From Molecular Data to Testable Hypotheses.

DNAI provides an interpretable simulation platform for molecular tumor boards, biomarker labs, and academic medical centers. Stratify cohorts, discover biomarkers, and validate hypotheses against external datasets — all with quantified uncertainty.

33 cancer types
50 Hallmark pathways
Structured abstention
Request Access View Architecture

Built for the Research Workflow

Unlike clinical decision tools, DNAI is designed for exploratory research — where the value is in hypothesis generation, not clinical action.

Interpretable

Every simulation traces to specific genes, pathways, and physics parameters. 50 Hallmark pathway scores, driver gene rankings, and integrated gradients — not a black-box score.

Honest About Limits

Structured abstention tells you when the model cannot reliably simulate. OOD detection, per-cancer reliability tiers, and Information Sufficiency Scores flag unreliable results.

Multi-Omics Native

RNA-seq, DNA mutations, CNV, methylation, and histopathology — fused probabilistically. Works with partial data via PoE; WSI required for BRCA and other histology-dependent types.

Research Applications

Four core use cases for translational oncology teams

Cohort Stratification

Identify molecular subgroups within your cohort

Encode your cohort into DNAI's 328-dimensional structured latent space. Cluster patients by pathway activity profiles rather than single-gene markers.

Example: "Which sarcoma patients have cell-cycle dysregulation vs. immune-hot signatures? How does simulated sensitivity differ between these groups?"
50 pathway dimensionsPer-patient latent vectorsExportable for downstream analysis

Biomarker Discovery

Find molecular features that associate with outcomes

Use the Driver Module (AUROC 0.933) for driver gene identification and integrated gradients for pathway-level attribution. Identify which molecular features most strongly influence simulated outcomes.

Example: "Which molecular features distinguish patients with high vs. low simulated sensitivity to CDK4/6 inhibition in our DDLPS cohort?"
Driver gene rankingPathway attributionIntegrated gradients

Hypothesis Generation

Simulate "what if" scenarios

Run virtual experiments by simulating treatment outcomes across molecular subgroups. Identify which combination strategies, dose schedules, or patient subsets show differential response in silico.

Example: "If we add an MDM2 inhibitor to CDK4/6i in MDM2-amplified DDLPS, does the simulation show greater sensitivity than CDK4/6i alone?"
Treatment comparisonSubgroup analysisIn silico experiments

Retrospective Validation

Test model performance on completed studies

Provide de-identified data from a completed trial or cohort study. DNAI simulates outcomes and compares against actual results — producing performance metrics, per-subtype analysis, and calibration curves.

Example: "How well does DNAI's simulation match actual outcomes from our completed DDLPS trial? Which subtypes does it handle best?"
C-index evaluationCalibration curvesPer-cancer metrics

Technical Foundation

Validated architecture with transparent limitations

0.704
Global C-index
Path A, internal val, survival ranking
0.633
CPTAC external C
Ridge encoding, 5 cohorts, 1,031 patients
0.009
Calibration ICI
Isotonic, per-horizon
0%
Physics violations
Biologically bounded

What's validated

Survival ranking (C-index) across 33 cancer types, 18 reliable
External validation on 5 independent cohorts (6,000+ patients)
Driver gene identification (AUROC 0.933 on IntOGen benchmark)
Per-horizon calibration (ICI < 0.01 at 1yr/2yr/3yr/5yr)

Known limitations

15 cancer types below C=0.60 — abstention recommended
BRCA survival is WSI-dependent (C=0.37 without histopathology)
External validation degrades with platform shift (microarray vs. RNA-seq)
Drug response, synergy, and resistance are simulated — not clinically validated

How We Work With Research Teams

Zero-risk retrospective validation first. Prospective collaboration if results warrant it.

1

Retrospective Validation

You provide de-identified data from a completed study. We run DNAI simulations and compare to actual outcomes. Zero cost, zero risk.

Timeline: 4–6 weeks from data receipt
2

Prospective Pilot

If validated — simulate outcomes in parallel with an active study. Compare simulated vs. actual results as they emerge.

Co-authorship on resulting publications
3

Tool Access

Research collaborators get platform access for their own cohort analysis, biomarker discovery, and hypothesis testing.

Free for academic research collaborators

What you provide

De-identified multi-omics data (RNA-seq minimum)
Clinical outcomes for validation
Scientific guidance on disease context

What you get

Disease-specific performance metrics
Co-authorship on publications
Free research tool access
Per-patient latent vectors for your own downstream analysis

See the Research Platform

Explore the tools for molecular analysis and simulation

Patient Dashboard

Patient Dashboard

Cohort overview

Platform Tour

Platform Tour

Full pipeline walkthrough

Traceability

Traceability

Glass-box audit trail

Validation

Validation

Calibration + metrics

Explore full platform tour

Intended use: DNAI is intended solely as a research tool for hypothesis generation and exploratory molecular analysis. It is not approved, cleared, or validated for clinical decision-making or diagnostic use.

Start with a retrospective validation

Send us de-identified data from a completed study. We'll run DNAI simulations and return performance metrics — no cost, no commitment.

Request Access Explore Technology