Data4Cure AI Omics Foundation Models
AI Solution · RNA1 family

Foundation models for omics.

The RNA1 family of foundation models learns the language of gene expression — powering bulk RNA-seq harmonization, disease subtyping, and preclinical-clinical translation.

Overview

Omics Foundation Models

Preclinical models — cell lines, organoids, xenografts — are indispensable for drug discovery and persistently hard to connect back to patient biology. Subtypes don’t transfer cleanly, drug-response predictions fail in patients, and the gap is structural. RNA1-DA closes it: a single molecular space where clinical tumors and preclinical models can be compared, clustered, and transferred — in both directions.

RNA1 is pre-trained on 180K+ bulk RNA-seq samples; the RNA1-DA domain-adaptation layer deconvolves cancer-cell signal and aligns tumors, cell lines, organoids and xenografts into one embedding — enabling forward and reverse translation across the pre-clinical / clinical gap.

How it works

Embed. Align. Translate.

RNA1-DA is not a batch correction — it’s a domain-adaptive foundation model that aligns clinical and preclinical samples at multiple biological scales, in both directions.

01 · Embed

One shared space

RNA1 embeds 130K+ clinical and preclinical RNA-seq samples across 6,800+ datasets and 25 cancer types into a shared foundation-model space.

02 · Align

Across sample types

A deconvolution layer separates cancer-cell signal from bulk tumor; a domain-adaptation layer aligns tumors, cell lines, organoids and xenografts into one molecular space.

03 · Translate

In both directions

Disease identity, molecular subtypes, genetic dependencies and drug response transfer between clinical and preclinical systems — with provenance.

By the numbers
130K+
clinical & preclinical samples aligned
6,800+
datasets, across 25 cancer types
62–88%
preclinical disease-classification accuracy, outperforming comparators
0.60
drug-response Spearman vs 0.35 baseline (~2×)

13 TCGA + 61 RNA1-derived subtypings transferred between clinical and preclinical systems · forward and reverse translation · presented at AACR 2026 (Abstract LB434).

Case study · collaboration with Champions Oncology

Validated on 1,500+ PDX models.

In collaboration with Champions Oncology, RNA1-DA was applied to 1,549 TumorGraft PDX models — aligned to 130K+ public clinical and preclinical samples with no Champions-specific training, providing independent-cohort validation. Subtypes, driver-alteration biology and drug response all transferred between PDX and patient samples.

1,549
Champions PDX models aligned
75%
disease-classification accuracy
87%
subtype–alteration associations recapitulated (1,964/2,270)
r = 0.78
cross-system concordance of variant prevalence
Get in touch

See Omics Foundation Models in action.

Walk through this with an applications scientist — focused on the questions that matter to you.