Can a screening or biomarker signal remain useful when its population, instrument, assay, clinical task, or biological source changes?

A locked seven-protein plasma assay separated vasculitis remission in an independent cohort, while a separate tissue-inference framework showed how missing proteins and tissue attribution affect interpretation of biofluid proteomics. Plasma and tape-strip studies further show that depletion, extracellular-vesicle enrichment, and collection site change the observable protein signal without guaranteeing diagnosis-level specificity. A pediatric plasma-proteomics study included immune thrombocytopenia as a clinically difficult comparator for MYH9-related disease, reinforcing that case-control separation and differential diagnosis are distinct tasks. Two urine-proteomics studies produced strong internal discrimination with different platforms and comparator designs, while neither tested a locked panel in an independent cohort containing the full set of realistic disease alternatives. Depression studies using a small type 2 diabetes cohort and a larger imaging-linked population produced different protein lists despite converging on metabolic context. In lung-screening research, CT dose and nodule measurements varied across scanners, units, and reconstruction settings, while a small breath-analysis study reported group separation that still requires prespecified classification and independent performance testing. In Giardia, a time-resolved whole-secretome study and a separate extracellular-vesicle study showed that sampling time, carrier compartment, and medium condition shape which released proteins are observed; the studies did not establish a portable clinical biomarker.

Reliable discrimination and a plausible biological explanation are complementary. Confusing them can make a promising panel look more portable than the evidence supports.

Prospective replication, prespecified thresholds, locked acquisition, specimen, preparation, timing, carrier-fraction, and culture workflows, clinically realistic disease mimics, calibration across instruments, laboratories, and populations, comparison with simpler models, and evidence connecting measured signals to their biological source.