Stratification of Genomic Sequence Complexity and Phenotypes for Precision Medicine

Statistics-first.No reference genome required.Unbiased discovery.

The website is being updated with additional content. Please contact: BD@rosabio.com for more information.

p2single results

Single Cell Demo

Sequence-level models, tested on cells they never saw. Every score is on cells, patients or experiments kept out of training.

Results last updated 2026-11-09 15:58 PST

Open the full demo
10public studies
AUC 0.996best score on cells the model never saw
17published methods beaten in one test
2026-11-09artifact last updated
More studies8 further results · swipe
AUC per protein-defined group
CD14 monocyte0.999
CD16 monocyte0.998
naive B0.990
memory B0.989
CD8 effector T (CD57+)0.984
MAIT0.983
CD8 naive T0.974
CD4 naive T0.968
Bone marrow · surface protein

Predicting protein-defined cell types in bone marrow from RNA

AUC 0.979GSE128639 · Stuart et al. 2019
Predicted vs measured PD-L1 proteinSeparate experiment
r = 0.74

PD-L1's own RNA alone: 0.38 · high vs low PD-L1 cells: AUC 0.93

CRISPR screen · PD-L1 protein

PD-L1 protein level predicted from RNA, tested on a separate experiment

r = 0.74SRP268462 · Papalexi et al. 2021
AUC per shared cell type in both directionsAcross tissues
artery model → uterus0.88
uterus model → artery0.86
Human Cell Atlas · two tissues

Cell types predicted across tissues and donors

AUC 0.88HCA arterial vasculature (Zhao 2025 · CELLxGENE 8f17ac63-aaba-44b5-9b78-60f121da4c2f) · HCA uterus (Ulrich 2024 · GSE260658)
Macro F1 on Xin: p2single and 17 published methodsTop of 17 methods
p2single1.000
scGenoByte0.999
scRGCL0.994
CIForm0.991
scGraph0.985
scGPT0.980
ACTINN0.966
SingleR0.952
Human pancreas · cell types

Pancreas cell types: at or above every published method

45 of 46Xin 2016 · GSE81608
Transfer AUC by species pairAcross species
human → mouse (4 tests)0.932
macaque → human (56)0.853
between 2 macaque species (20)0.950
human → other humans (56)0.973
Brain · across species

Brain cell types predicted across species

AUC 0.932Allen Brain Cell Atlas HMBA-10xMultiome-BG (release 20250630) · human, macaque, mouse
Cosine similarity, Norman screenTop of 4 methods
p2single0.890
latent additive0.790
CPA0.760
linear0.600
GEARS0.440
CRISPR screens · gene response

How cells respond when 2 genes are changed at once

0.890GSE133344 · Norman 2019 · PerturBench
AUC per egg-development stage
stem cells, progenitors0.997
early meiosis0.997
meiosis0.987
late meiosis0.994
early egg stage 10.997
early egg stage 20.998
early egg stage 30.999
Zebrafish ovary · germ-cell stages

Stages of egg development in the young zebrafish ovary

AUC 0.996GSE191137 · Liu et al. 2022
Share of cells with KRAS G12V
Patient 6: tumour epithelial cells17 of 18
Patient 6: immune cells0 of 22
Other 9 patients: all cells0 of 515
Variant calling · single cell

Cancer mutations found in individual cells

NRAS G12DSRP268462 · Papalexi et al. 2021 · GSE316782 · Xi et al. 2026
P2G results

Liquid Biopsy Demo

One model built straight from raw sequencing reads, for RNA and DNA alike, with no gene list. Every score is on samples kept out of training.

Results last updated 2026-11-16 11:40 PST

Open the full demo
6 of 6published results matched or beaten
AUC 1.000best score on samples the model never saw
41study questions scored on held-out samples
2026-11-16artifact last updated
Main point 1 · Reference samplesKnown answer
16 of 16

How to read the chart. Each row is one fusion. Each column is one mix level: the percent of fusion RNA added to normal RNA. At each mix level, 3 libraries were made and sequenced. The number in a box tells how many of these 3 libraries contain the fusion. The last column shows that none of the 277 control samples contains any of the fusions. The two outlined fusions. KIF5B-RET is found in all 3 libraries that contain 5% fusion RNA. It is also found in 2 of the 3 libraries that contain only 1% fusion RNA. SLC45A3-BRAF is found in all 3 libraries that contain 5% fusion RNA. It is also found in 1 of the 3 libraries that contain only 1% fusion RNA. No other fusion is found in all 3 libraries at the 5% mix level.

Reference samples · fusion RNA · Seraseq Fusion RNA Mix v3 (SRP293279)

All 16 fusions in the reference mix are found with no model training, 9 already at a 1% spike-in, and none in 277 control samples

16 of 1618 libraries: the Seraseq Fusion RNA Mix v3 reference diluted into normal RNA, 3 repeat libraries at each of 6 mix levels (0, 1, 5, 10, 50 and 100% fusion RNA mixed into normal RNA), 101 to 227 million read pairs per library (median 159 million). 277 control samples with no spiked fusion: 90 plasma RNA (7.3 billion reads) and 187 tumour DNA libraries (42.2 billion reads). No stage: lab-made reference material, not patients.SRP293279 · Seraseq Fusion RNA Mix v3 reference material
Main point 1 · Reference samplesKnown answer
R² 0.93
Reference samples · tumour DNA · SEQC2 (SRP296025)

P2G predicts the tumour DNA percent of 38 held-out libraries (R² 0.93)

R² 0.9338 libraries kept out of training: 8 with no tumour DNA, 4 with 0.8%, 17 with 4% and 9 with 20% tumour DNA. Tumour DNA percent: how much of a library's DNA comes from the SEQC2 reference tumour DNA, which the study diluted into normal DNA. P2G used 5 million reads per library, 6% of the median 85 million sequenced. No stage: lab-made reference material, not patients.SRP296025 · study: Deveson et al. 2021, Nature Biotechnology · data description: Gong et al. 2022, Scientific Data
More results15 further results · swipe
Held-out AUC: P2G and the paperBeats the paper
P2G0.94
Paper0.91
Pregnancy · maternal plasma RNA · SRP271730

Adverse pregnancy outcomes are told apart from uncomplicated pregnancies at AUC 0.941; the paper reports 0.91

AUC 0.941SRP271730 · Del Vecchio et al. 2021, Epigenetics
Held-out AUC: P2G and the paperBeats the paper
P2G0.90
Paper0.87
Cancer · plasma cell-free DNA, bisulfite · SRP610281

Gastric cancer is found from plasma DNA at AUC 0.904; the paper reports 0.870

AUC 0.904SRP610281 · Long et al. 2026, BMC Cancer
Held-out AUC: P2G and the paperBeats the paper
P2G0.96
Paper0.658
Cancer · plasma vesicle small RNA · SRP345647

Colorectal cancer is found from vesicle RNA at AUC 0.956; the paper's miR-320c marker scores 0.658

AUC 0.956SRP345647 · Yang et al. 2022, Cellular Oncology
Main point 2 · Compared with papers
13 highlighted

Numbers match the dots in the chart above; the missing numbers are under Smaller test sets.

Key results · every highlighted prediction

13 highlighted predictions reach a held-out AUC of 0.80 or more: 4 are within 0.02 of the paper's AUC or higher (glioblastoma, two breast cancer studies and gastric cancer) and 9 have no published AUC

13 highlightedPaper AUCs: each study's publication
Main point 2 · Compared with papers

Numbers match the dots in the chart above.

Same prediction as the paper · reads used

P2G is within 0.02 of the paper's AUC, or higher, for glioblastoma, breast cancer, gastric cancer and colorectal cancer, using 6% to 23% of the paper's reads

Reads in the paper: median reads sequenced per sample (public sequencing records)
Cases found at 95% specificity
Gastric cancer vs benign stomach disease100%
Melanoma vs healthy100%
Epilepsy vs healthy100%
Colorectal or appendix cancer with spread to the belly lining vs no cancer93%
Lung cancer vs healthy92%
Glioblastoma vs people without symptoms90%
Colon cancer vs healthy89%
Cancer spread to the brain vs people without symptoms88%
Sensitivity at fixed specificity

At 95% specificity, P2G achieves >=80% specificity

≥80%Test samples of each study
AUC · Cross-study · applied unchangedAnother lab
Stanford → UCSD discovery0.69
UCSD discovery → Stanford0.70
UCSD triage → Stanford0.73
Li 2017 → Puddu 20260.69
Song 2017 → Sun 20261.000
Cross-study · applied unchanged

Five models score 0.69 to 1.00 on a study from another lab (six scores; one colorectal model is scored two ways)

AUC 0.69–1.00Moufarrej 2022; Srinivasan 2020; Li 2017; Puddu 2026; Song 2017; Sun 2026
Main point 3 · Across studiesAnother lab
57%
Cross-study · what the models that work on another study use

31 cross-study scores between different labs reach 0.65 or more; the cancer models rely mostly on sequences in genes (57%)

57%Cross-study scores between different labs
Main point 3 · Across studies
41 of 68

Paper comparison: the published AUC for the same prediction is set against P2G in the charts of main point 2; this chart shows each study's own held-out result for the 41 predictions at 0.80 or more on 20 or more test samples, including those with no published AUC.

Within each study · all predictions

Within their own study, 41 of 68 predictions reach a held-out AUC of 0.80 or more

41 of 68Held-out samples of each study
AUC on the other center's samples
UCSD triage → Stanford0.73
UCSD discovery → Stanford0.70
Stanford → UCSD discovery0.69
UCSD discovery → Stanford0.68
Stanford → UCSD discovery0.67
UCSD discovery → Stanford0.67
UCSD discovery → Stanford0.64
Stanford → UCSD triage0.62
Cross-study · preeclampsia · Stanford and UCSD

10 models trained on one center score 0.60 to 0.73 when tested on the other (Stanford and UCSD)

AUC 0.60–0.73Moufarrej 2022 (Stanford); Srinivasan 2020 and Morey 2023 (UCSD)
Held-out AUCNot in the paper
Melanoma vs healthy1.000
Colorectal or stomach cancer vs healthy0.860
Colorectal cancer vs healthy0.846
Preterm labour vs control pregnancies0.804
Broader insights · predictions the papers did not test

4 predictions the papers did not score reach an AUC of 0.80 or more on 20 or more test samples

AUC 1.000Held-out samples of each study
Main point 4 · P2G: broader insights from published studiesUnseen patients
AUC 0.848

The paper reports sensitivity and specificity of a 3-microRNA ratio, not an AUC (Morey 2023).

Pregnancy · serum small RNA · UCSD triage cohort

Preeclampsia is told apart from normal blood pressure at AUC 0.848; the paper reports sensitivity and specificity only

AUC 0.848UCSD triage cohort, no public accession recorded · Morey et al. 2023, Science Advances
Held-out AUC, not tested in the paperNot in the paper
Melanoma vs healthy1.000
Lung cancer, response vs no response to sintilimab immunotherapy1.000
Liver cancer vs hepatitis B and healthy0.92
Colorectal or stomach cancer vs healthy0.86
Colorectal cancer vs healthy0.85
Preterm labour vs control pregnancies0.80
Death after immunotherapy (deceased vs alive)0.75
Pregnancy complications vs normal pregnancy0.69
Not tested in the paper; predicted by us

Untested predictions exceed 0.80

AUC 1.000Held-out samples of each study
Main point 4 · P2G: broader insights from published studies
2 genes

Stage. Melanoma (SRP239782): stage IV for all 32 test patients. Colorectal or stomach cancer (SRP343186): stages I to IV, written out in the table of predictions the papers did not test. Bladder cancer (SRP619729): advanced, 3 stage III and 17 stage IV among the 20 patients who died. Colorectal cancer, methylation (PRJNA529690): stages I to III. Lung cancer, 5hmC (SRP375150): the study does not say which patient had which stage (across its 97 patients: stage I 22, stage II 7, stage III 11, stage IV 57). Preterm labour and preeclampsia are not cancers, so there is no stage.

New insights · genes not yet linked to the disease

The 2 genes with the most model weight per study that no paper links to the disease yet

2 genesModels behind each AUC above
Held-out AUC
Lung cancer, response vs no response to sintilimab immunotherapy1.000
Liver cancer vs hepatitis B and healthy0.920
Fewer than 20 test samples · listed apart from the main points

These results are shown as hollow dots on the charts and are not used for the main points

AUC 1.000Held-out samples of each study; paper values from each paper

Benefits of Rosa Bio Technology

  • ROSA's "statistics first" Semore platform allows detection of genomic complexity missed by other methods
  • Allows faster, cheaper, smarter genomics analysis
  • Readily scalable
  • Agnostic to nucleic acid type: DNA (WGS and targeted), RNA, short and long read, exome, non-coding regions, ctDNA, bisulfite, etc.
  • Detects structure-based changes, not based on gene panels
  • Can be used with tissue sample or liquid biopsy (plasma cfDNA)
  • Software = AI ready
  • Diverse applications: Dx, Tx, other

Panoptic. Faster. Cheaper.

Speed

Faster processing, lower overhead.

Accuracy

Assess all sequence variation in a sample with no bias.

Cost

High-efficiency analysis tools.

A Statistics-First Approach.

A new paradigm for analysis of nucleic acid sequence information. Unlike traditional methods, we require no reference genome for sequence alignment.

Our approach enables truly unbiased discovery, identifying patterns and variations that conventional tools simply cannot detect.

Why Statistics First

Our statistics-first approach analyzes raw data directly, eliminating reference genome bias and enabling detection of all genomic complexity.

Precision Medicine Applications

Patient stratification using sequence variation that predicts treatment response and disease progression.

Leadership

JS

Julia Salzman

Founder

Associate Professor in the Department of Biomedical Data Science, and by courtesy in Biology and Statistics at Stanford University. A.B. in Mathematics from Princeton University magna cum laude and Ph.D. from Stanford University in Statistics. Developed statistical algorithms that discovered circular RNA and statistics-first genomic algorithms. Research funded by NIH, NSF, CZI. Recognized with Alfred P. Sloan Fellowship, McCormick-Gabilan Fellowship, and NSF CAREER Award.

MR

Michael Rabson

Co-Founder & CEO

More than 25 years of experience in life sciences. Former partner at Wilson Sonsini Goodrich & Rosati for 15 years. Served as General Counsel of Maxygen (MAXY) and Cytokinetics (CYTK). Former patent examiner at USPTO focused on biotechnology. Member of the patent bar. Author of scientific publications in Science, Nature, PNAS. Ph.D. from Yale University and J.D. from Yale Law School.