Stratification of Genomic Sequence Complexity and Phenotypes for Precision Medicine

No training. No model was trained, and the method was not told which fusions are in the mix: it searched every read for 41 known fusion junctions and reports the ones it finds. It was also not told which libraries were spiked, or at what 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 replicate libraries at each of 6 spike-in levels, 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

34% of the model's weight is on sequences with the spiked mutations, among them KRAS G12A, BRAF V600E and TP53 R306*

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 at 0%, 4 at 0.8%, 17 at 4% and 9 at 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 of a median 85 million reads per library (6%). No stage: lab-made reference material, not patients.SRP296025 · study: Deveson et al. 2021, Nature Biotechnology · data description: Gong et al. 2022, Scientific Data · spiked variants: Jones et al. 2021, Genome Biology

More results17 further results · drag or use the arrowsswipe

Held-out AUC: P2G and the paperBeats the paper

P2G1.000

Paper0.91

Pregnancy · maternal plasma RNA · SRP271730

Adverse pregnancy outcomes are told apart from uncomplicated pregnancies at AUC 1.000; the paper reports 0.91

AUC 1.000SRP271730 · Del Vecchio et al. 2021, Epigenetics

Held-out AUC: P2G and the paperBeats the paper

P2G0.94

Paper0.87

Cancer · plasma cell-free DNA, bisulfite · SRP610281

Gastric cancer is found from plasma DNA at AUC 0.94; the paper reports 0.87

AUC 0.94SRP610281 · 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 0.96; the paper's miR-320c marker scores 0.66

AUC 0.96SRP345647 · Yang et al. 2022, Cellular Oncology

Held-out AUC: P2G and the paperMatches the paper

P2G0.97

Paper0.956

Cancer · plasma cell-free DNA, methylation (MRE-seq) · SRP432064

Lung cancer is found from 2 of 46 million reads per sample at AUC 0.97; the paper reports 0.96

AUC 0.97SRP432064 · Kwon et al. 2023, Scientific Reports

Main point 2 · Compared with papers

9 of 14

Key results · every highlighted question

P2G is within 0.02 of the paper's AUC, or higher, on 9 of 14 highlighted questions with a published AUC

9 of 14Paper AUCs: each study's publication

Main point 2 · Compared with papers

11 of 28

The two dots at 100% are Alzheimer's and Parkinson's disease from blood DNA (SRP722164, Pollard 2026). The paper first estimates how much of the cell-free DNA comes from each type of neuron, using the methylation of every read and a reference map of neuron methylation made outside the study. It then scores those amounts. It reports no result from the sequences alone. P2G used only the sequences, and the same reads as the paper.

Same question as the paper · reads used

P2G is within 0.02 of the paper, or higher, in 11 of 28 comparisons, using 4% to 23% of the paper's reads

11 of 28Reads in the paper: median reads sequenced per sample (public sequencing records)

Main point 2 · Compared with papers

2 of 4

Same question · the paper's own measure

On the papers' own sensitivity and specificity, P2G is higher in 2 of 4, about the same in 1 and lower in 1

2 of 4Held-out samples of each study; paper values from each paper

Cases found at 95% specificity

Colorectal cancer vs healthy100%

Melanoma vs healthy100%

Adverse pregnancy outcome vs no complication100%

Adverse pregnancy outcome vs no complication88%

Colorectal or lung cancer vs healthy88%

Severe preeclampsia vs control pregnancies86%

Colon cancer vs healthy84%

Colorectal cancer vs healthy83%

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 → SRP6907851.000

Cross-study · applied unchanged

Cross-study: five models score 0.69 to 1.000 on a study from another lab

AUC 0.69–1.000Moufarrej 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

49 of 68

Within each study · all questions

Within their own study, 49 of 68 predictions reach a held-out AUC of 0.80 or more

49 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 healthy0.93

Death after immunotherapy (deceased vs alive), bladder cancer0.88

Preterm labour vs control pregnancies0.88

Colorectal cancer vs healthy0.85

Liver cancer vs hepatitis B and healthy1.000

Lung cancer, response vs no response to sintilimab immunotherapy1.000

Head and neck cancer vs healthy0.81

Broader insights · questions the papers did not test

5 questions 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 0.96

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 0.96, a score the paper did not report

AUC 0.96UCSD triage cohort, no public accession recorded · Morey et al. 2023, Science Advances

Main point 4 · P2G: broader insights from published studiesUnseen patients

AUC 0.94

The paper reports 94.1% of lung cancers found at 96.2% specificity on its own validation set, and no AUC (Shao 2022); the square marks it. At 96.2% specificity or more, P2G finds 5 of 11 in its test samples.

Cancer · plasma cell-free DNA, 5hmC · SRP375150

Lung cancer is found from 5hmC in plasma DNA at AUC 0.94, with 9 of 11 cancers found at 95% specificity

AUC 0.94SRP375150 · Shao et al. 2022, Clinical and Translational Discovery

Held-out AUC, not tested in the paperNot in the paper

Liver cancer vs hepatitis B and healthy1.000

Melanoma vs healthy1.000

Lung cancer, response vs no response to sintilimab immunotherapy1.000

Colorectal or stomach cancer vs healthy0.93

Death after immunotherapy (deceased vs alive)0.88

Preterm labour vs control pregnancies0.88

Colorectal cancer vs healthy0.85

Head and neck cancer vs healthy0.81

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

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

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.

White Papers

Read our white papers on the benefits of a statistics-first approach to genomic analysis.

Precision Medicine Applications

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

Leadership

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.

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.

Stratification of Genomic Sequence Complexity and Phenotypes for Precision Medicine

Single Cell Demo

Exhausted vs active killer T cells in patients on immunotherapy

PD-L1 levels in lung tumour immune cells of a new patient

Predicting protein-defined cell types in bone marrow from RNA

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

Cell types predicted across tissues and donors

Pancreas cell types: at or above every published method

Brain cell types predicted across species

How cells respond when 2 genes are changed at once

Stages of egg development in the young zebrafish ovary

Cancer mutations found in individual cells

Liquid Biopsy Demo

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

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

Adverse pregnancy outcomes are told apart from uncomplicated pregnancies at AUC 1.000; the paper reports 0.91

Gastric cancer is found from plasma DNA at AUC 0.94; the paper reports 0.87

Colorectal cancer is found from vesicle RNA at AUC 0.96; the paper's miR-320c marker scores 0.66

Lung cancer is found from 2 of 46 million reads per sample at AUC 0.97; the paper reports 0.96

P2G is within 0.02 of the paper's AUC, or higher, on 9 of 14 highlighted questions with a published AUC

P2G is within 0.02 of the paper, or higher, in 11 of 28 comparisons, using 4% to 23% of the paper's reads

On the papers' own sensitivity and specificity, P2G is higher in 2 of 4, about the same in 1 and lower in 1

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

Cross-study: five models score 0.69 to 1.000 on a study from another lab

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

Within their own study, 49 of 68 predictions reach a held-out AUC of 0.80 or more

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

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

Preeclampsia is told apart from normal blood pressure at AUC 0.96, a score the paper did not report

Lung cancer is found from 5hmC in plasma DNA at AUC 0.94, with 9 of 11 cancers found at 95% specificity

Untested predictions exceed 0.80

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

Benefits of Rosa Bio Technology

Panoptic. Faster. Cheaper.

Speed

Accuracy

Cost

A Statistics-First Approach.

Why Statistics First

White Papers

Precision Medicine Applications

Leadership

Julia Salzman

Michael Rabson