Answers are grounded in this paper and cite their sources.
Background and Objectives: Prognostic scores built from ion channel gene sets are published regularly, some nominating the channels as drug targets. That arbitrary gene sets predict cancer outcome is established; we asked whether these models contain the proteins their titles name, and what decides this. Materials and Methods: Eleven signatures in nine carcinomas, from 17 PubMed records, were reconstructed from full texts, their 93 gene slots classified under one explicit criterion and each candidate pool’s delimitation taken from its paper’s Methods. Each was re-evaluated in its Cancer Genome Atlas cohort under nested Cox models adding purity, composition, differentiation and proliferation; ten were compared against size-matched random gene sets and two transferred to independent cohorts. Results: Forty-seven of 93 gene slots (50.5%) are pore-forming subunits, a figure that moved five times here and spans 30.1–59.1%. What decides this is how the candidate pool was delimited: an enumerated biological inventory gives 83.3% to 100%, an annotation-database keyword 16.7% to 25%, and single-gene correlation none. That ordering is the robust finding, surviving two narrower schemes and an independent external annotation; the separation of the bands holds under the primary criterion alone. Discrimination runs the other way: three of the four models with more than 75% pore-forming subunits fail to beat their random comparators—an observation, not a testable relationship. One model published as voltage-gated contains no voltage-gated channel under any scheme. Six signatures add 0.000 to 0.097 concordance units to age and stage; no paper adjusted for purity or composition, or stated its coefficients’ scale. Conclusions: What an ion channel signature contains is ordered by how its candidate pool was assembled, not by what the paper is called. None is near clinical use, and a paper naming a gene family in its title should report what fraction of its model belongs to it.
No takes yet. Share an insight, caveat, or question.
Yohan Seo (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: