Aurora’s discovered CV-death proteins map to established and emerging druggable biology. UK
Biobank (N ≈ 500,000)

KEY INSIGHT
Aurora does not just rank patients by risk; the proteins it selects are mechanism. The top features of the heart failure to CV-death model (C-index 0.76) are concentrated in pathways that are already drug targets: the renin-angiotensin-aldosterone system (renin, EDN1), metabolic stress (ANGPTL4, GDF-15), and fibrosis and remodelling (SPON1, HGF). The same run that stratifies patients also nominates target and indication hypotheses.

Risk Scores Are Cheap; Mechanism Is Valuable

Most prognostic models are black boxes: useful for stratification, silent on biology. Aurora’s penalised-Cox signatures are interpretable by construction, so each selected protein carries a direction of effect and a magnitude. When those proteins cluster in coherent, druggable pathways rather than scattering as noise, they become leads for target identification and for choosing which patients a mechanism is most likely to help.

 

The Discovered Proteins Map to Druggable Pathways

The strongest prognostic proteins are not arbitrary. They sit on pathways with approved drugs or active clinical programmes, which means a prognostic hit can double as a mechanistic hypothesis and a ready-made patient-selection marker.

 

Pathway, Marker, and Therapeutic Relevance

Pathway	Aurora markers	Therapeutic relevance
RAAS / renal perfusion	Renin, EDN1	Target of ACE inhibitors, ARBs, MRAs, and endothelin antagonists.
Metabolic / systemic stress	ANGPTL4, GDF-15	Active antibody and small-molecule programmes across cardiometabolic disease.
Fibrosis / remodelling	SPON1, HGF, SCGB3A1	Anti-fibrotic and tissue-repair mechanisms of growing interest in HF.
Congestion / wall stress	NT-proBNP, ANGPT2	Canonical readouts of cardiac load; pharmacodynamic and enrichment markers.

Direction of effect is consistent with biology: RAAS activation, metabolic stress, and fibrosis markers all rise with worse prognosis, while protective markers (for example HPGDS) move the other way. That coherence is the signal that these are real mechanism, not overfit.

 

Why It Matters for Discovery and Development

A single Aurora run yields three assets at once: a validated prognostic model, a ranked set of mechanistic hypotheses, and the patient-selection markers to test them. Because the platform generalises to any indication by editing the cohort definition, the same approach can surface target and stratification hypotheses across a portfolio, turning population-scale data into a faster route from biology to a defensible development plan.