Answers are grounded in this paper and cite their sources.
This study formulates Trojan DNA detection as a controlled engineering problem and develops a reproducible evaluation framework for assessing detection performance under adversarial conditions. The framework systematically examines the interaction between payload encoding, insertion strategy, feature representation, and classifier behavior. Two encoding schemes (ASCII and Base4) and five insertion strategies, including worst-case configurations, are evaluated to quantify their impact on sequence-level perturbations. Feature representation combines positional encoding and k-mer statistics, followed by dimensionality reduction using PCA within a strictly nested cross-validation framework. Classification is performed using SVM, with hyperparameters optimized via PSO. Results across multiple adversarial settings show that optimization-driven SVM models achieve consistently high detection performance while maintaining lower computational cost and improved interpretability compared to deep learning approaches. The findings demonstrate that evaluation design and encoding-aware modeling are critical for reliable deployment of DNA-based threat detection systems.
No takes yet. Share an insight, caveat, or question.
Shankar et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: