Discovering a biomarker means finding a molecule whose level reliably differs between disease and health by more than chance and technical noise could explain — which is fundamentally a signal-versus-noise problem.
delta(m) = mean(disease_m) - mean(healthy_m)
candidate if |delta| > effectSize*noise
- Molecule panel size — how many candidate genes/proteins/metabolites are screened simultaneously.
- Cohort size — number of patient samples in disease vs matched healthy groups — statistical power.
- Effect size — true underlying biological difference between disease and healthy states for a given molecule.
- Noise / batch effect — technical variability across samples/batches that can hide or mimic a true signal.
Biomarker discovery pipelines like this are how PSA, HER2 and, more recently, circulating-tumor-DNA panels moved from a differential-expression signal in a discovery cohort to routine clinical tests.