XenoScan1 enables alignment-free discovery of horizontal gene transfer through reference-relative genomic discordance

Boctor, Joseph, Bauer, Denis, Alattas, Hussain, Jampasa, Sakda, Fouad, Walid, Weiss, Emiliana, Wilson, Laurence, Sala, Samuele, Masuda, Reika, Murphy, Daniel
BioData Mining · Springer Nature · 2026-11-20 · DOI: 10.1186/s13040-026-00624-3

Abstract

Horizontal transfer of genetic material (HGT) is important for genomic plasticity, but detecting HGT is hindered by phylogenetic saturation and compositional amelioration. Most existing methods that detect HGT do so by using phylogenetic or GC content analysis. This approach becomes less effective as the signal degrades over time and leads to misclassification. In this study, 3,069 loci were analysed, and only 16.4% of the mobile genetic elements had higher GC divergence than core genomes (Mann–Whitney U, P = 1.70 × 10⁻²⁷). To address these limitations, we introduce XenoScan1, a machine learning tool that quantifies reference-relative genomic discordance using a dynamic reference-genome manifold. The reference manifold comprises 17,708 species’ genomes. Core reference genomes are represented as 322-dimensional vectors containing GC and oligonucleotide percentages. Query sequences are converted to 323-dimensional query tensors and classified with a calibrated Random Forest ensemble. XenoScan1 achieved ROC–AUC = 0.9577, identifying horizontal genetic acquisitions with 98.3% precision and 92.0% recall (F1 = 0.951). At a detection threshold of P _HGT ≥ 0.72, codon-bias false positives are suppressed (MCC = 0.706). On the experimental datasets, conventional tools showed a precision-recall trade-off, with F1-scores as low as 0.143. XenoScan1 allows alignment-free, fast and reliable detection of HGT.

Read at publisher →

Similar articles