HeFS: Helper-Enhanced Feature Selection via Pareto-Optimized Genetic Search

01 / Research question

A feature subset can miss complementary information left among the features discarded by an existing method.

02 / Innovation

Find complementary helper sets with biased initialization, proportion-guided mutation and Pareto-optimized genetic search.

03 / Mechanism

HeFS preserves an existing feature subset and searches its residual space for complementary helpers. Biased initialization, ratio-guided mutation and Pareto selection jointly optimize accuracy and feature complementarity.

04 / Experimental setup

Manuscript v1 Table 1 lists eighteen datasets, despite inconsistent total counts elsewhere. Section 4.2 uses ten random runs and five-fold validation, a k=5 nearest-neighbor evaluator, population 30 and 100 generations. Baselines include RF/Lasso and six evolutionary selectors. GSE64951 has 94 samples/54,675 features; DLBCL has 77/5,469.

Author manuscript v1 · 21 October 2025 · §§4–5 / Table 2 · Source ↗

05 / Results

Table 2 improves DLBCL BHOA accuracy from 97.50% to 99.88%, but Semeion FTGGA falls from 97.80% to 96.62%.

06 / Ablation & limitations

Figure 3 replaces initialization/mutation mechanisms with random variants. Gains depend on the parent selector, dataset and feature expansion. Small high-dimensional cohorts and unclear nesting of search and validation limit generalization claims; feature complementarity is not biological causality. These are preprint experiments, separate from the final article.

Feature selection is a general method, not a peptide-specific result. The paper is advance-online/in press in 2026; the volume citation is not yet assigned here.

07 / My contribution

Co-corresponding author.

08 / Paper & code