Back

Integrated inference of cancer gene expression from cell-free plasma chromatin

Gulati, G. S.; Vasseur, D.; Nawfal, R.; Sotudian, S.; Semaan, K.; Eid, M.; Seo, J.-H.; Phillips, N.; Canniff, J.; Savignano, H.; Chhetri, S. B.; Jin, Z.; Ou, Y.; Bani, M.-A.; Lee, G.-S. M.; Trowbridge, R.; Epstein, I.; Rickards, G.; Cordeiro, P. R. D. S.; Chehade, R. E. H.; Zhang, Z.; James, B.; Massard, C.; Italiano, A.; Hollebecque, A.; Soria, J.-C.; Andre, F.; Badoual, C.; Bellmunt, J.; Singh, H.; Aguirre, A. J.; Wolpin, B. M.; Choueiri, T. K.; Baca, S. C.; Freedman, M. L.

2026-02-19 genomics
10.64898/2026.02.18.706026 bioRxiv
Show abstract

Gene expression is a defining determinant of tumor identity, behavior, and therapeutic response, yet remains challenging to measure noninvasively. Here, we introduce APEX (Associating Plasma Epigenomic features with eXpression), a framework for inferring expression from circulating cell-free chromatin. Trained on [~]270,000 gene-sample pairs from matched tumor RNA-seq and plasma cfChIP-seq across multiple cancers and validated on >15 unseen cancer subtypes, APEX accurately infers cancer gene expression across a range of tumor fractions and outperforms existing plasma-based approaches by integrating positional histone mark and DNA fragmentation patterns across promoters and gene bodies. Using plasma alone, APEX enables classification of prognostically relevant basal and classical pancreatic cancer subtypes and identifies plasma-inferred NECTIN4 expression as a biomarker of response to enfortumab vedotin in metastatic bladder cancer. Together, these findings establish APEX as a biopsy-free approach for profiling tumor transcriptional states and extend liquid biopsy beyond genomic alterations to clinically relevant gene expression programs.

Matching journals

The top 4 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.