Back

Detection of Melanoma Using Deep Serum Proteome Profiling and Machine Learning

Hall, D.;Gonzalez, B.;Schepmoes, A.;Filmore, T.;Sagendorf, T.;Drucker, B.;Rodland, K.;Lacson, J.;Dalgard, C.;Brownell, I.;Lee, J.;Shriver, C.;Keung, E.;Grenier, L.;Petyuk, V.;Liu, T.

2026-06-16 Cancer Biology
10.64898/2026.06.11.731698 bioRxiv
Show abstract

Early detection determines melanoma outcomes, yet current screening relies on visual inspection that misses molecular changes preceding clinical diagnosis. To identify serum protein signatures of melanoma development, we leveraged the Department of Defense Serum Repository, analyzing 390 longitudinal serum samples from 73 melanoma cases and matched controls across four timepoints (4 and 2 years prior, diagnosis, and 2 years post-diagnosis) using data-independent acquisition mass spectrometry with Seer Proteograph nanoparticle enrichment. We quantified 3,364 proteins and applied machine-learning strategies combining cross-sectional case-control comparisons with longitudinal tracking of within-individual changes. Cross-sectional analysis at diagnosis achieved AUC 0.823 (95% CI: 0.723-0.923), identifying eight consensus features selected in >50% of cross-validation folds: PRSS1, GSK3B, FAM20C, CES1, CCL14, EPHA10, LMAN2, and ITIH1. These signatures, spanning proteases, immune activation, and extracellular matrix remodeling, demonstrate proof-of-concept for serum-based melanoma detection and provide candidate biomarkers for validation.

Matching journals

The top 8 journals account for 50% of the predicted probability mass.

50% of probability mass above

"Similar papers" are the closest papers from that journal in the model's embedding space. They show what the match is built on, but the ranking comes mostly from a classifier over the whole training set, not from these examples alone.