Cleavr · Early stage · Building in public

Systematic biomarker discovery for triple-negative breast cancer.

Cleavr tests candidate TNBC biomarkers against public sequencing data, subtype by subtype, with every claim independently re-derived and checked against the literature before it's published.

TCGA-VerifiedSubtype-Level AnalysisIndependently Re-DerivedLiterature Cross-CheckedOpen MethodologyCross-Cohort Validated TCGA-VerifiedSubtype-Level AnalysisIndependently Re-DerivedLiterature Cross-CheckedOpen MethodologyCross-Cohort Validated
Backed by research at
UNC‑Chapel Hill
NC State University
01Problem
TNBC isn't one disease, but it's still targeted like one.

Triple-negative breast cancer (TNBC) is not one disease. Lehmann's 2011 classification split TNBC into distinct molecular subtypes: basal-like, mesenchymal, luminal androgen receptor, and others, each with different biology and different drug sensitivities. Despite that, candidate biomarker and targeting studies are still mostly done one gene, one paper, one lab at a time, without systematically checking whether a target behaves consistently across subtypes before it gets built into a therapeutic strategy.

02Approach
Test it against real data before it becomes a strategy.

Cleavr tests candidate TNBC biomarkers against public, patient-level sequencing data, starting with TCGA, broken out by molecular subtype, and cross-checks every claim against the primary literature before it's published. Every number is independently re-derived from raw source files in a second pass, not computed once and trusted. Null results get reported exactly as directly as positive ones.

116
TNBC patients screened
15
Candidate genes tested
7 / 8
Replicated cross-cohort
03Status
Two studies live. Independently verified, start to finish.

Each entry below is independently re-derived from raw source data and cross-checked against the literature before publishing. Null results are reported as directly as positive ones.

Case study 01
GNRHR expression across TNBC molecular subtypes

Does the LHRH/GnRH receptor (GNRHR), the target of an active TNBC nanoparticle-targeting strategy, express differently across TNBC's molecular subtypes? If it does, that would matter for how consistently a receptor-targeted therapy could work across the disease's biological variants.

116
TNBC patients
p = 0.71
Across subtypes
p = 0.98
Survival, log-rank

Result: no statistically significant difference across subtypes, and no relationship to survival in this cohort (TCGA), a real, well-caveated null result.

Single-cohort · Complete
Case study 02
Systematic candidate screen, with cross-cohort validation

A gene-agnostic follow-up to Case Study 01: instead of testing one hand-picked gene, 15 candidates were sourced systematically from Open Targets' TNBC-associated target scores, then tested for subtype- and survival-association with Benjamini–Hochberg FDR correction applied across all 15 together, then scored on druggability (ChEMBL) and TNBC-specific literature coverage (PubMed).

15
Candidates screened
8 / 15
Significant after FDR
7 / 8
Replicated in METABRIC

The 8 FDR-significant genes were re-tested against METABRIC, a second, independent cohort (320 TNBC patients, nearly 3x the TCGA cohort) profiled on a different expression platform. 7 of 8 replicated on both. RRM2 did not (TCGA-significant, but p = 0.23 in METABRIC) and is reported as a miss, not dropped.

FTO is the standout candidate: significant in both cohorts, 91 potent ChEMBL ligands (a known druggable target), and only moderately studied in TNBC specifically (14 papers). Most other replicated genes (POLD1, POLD2, PRIM2, RRM1, TYMS, POLE) are DNA-replication-machinery genes, likely reflecting the known higher proliferation of the basal-like subtype rather than a subtype-specific mechanism on its own.

Status: computational only. Nothing here has been validated at the protein level yet, that's the next step, now underway with wet-lab flow cytometry.

Cross-cohort validated
04Updates
Milestones, as they happen.
05Roadmap
From one gene to a real discovery pipeline.
06Contact
Have a biomarker in mind?

Full methodology, data, and code for every case study are public.