GeneFamHunter: annotation-independent recovery of divergent and tandemly arrayed gene-family members from genome assemblies

Motivation: Gene families expand by tandem duplication, diverge and decay, so their members are often missing, fragmented or fused in automated genome annotations, and divergent copies escape pairwise similarity searches. These challenges are particularly pronounced in non-model organisms, for which well-annotated genomes and reference proteomes are often unavailable, limiting the applicability of protein-based gene family recovery methods. Results: We developed GeneFamHunter to recover gene-family members directly from nucleotide assemblies. It searches continuous six-frame translations of overlapping genomic windows using four complementary domain- and profile-based methods, clusters hits while distinguishing tandem paralogs, predicts and validates gene models, and constructs gene trees. Across six Quest for Orthologs genomes, window-based HMMER searches recovered all 130 reference papain-like (C1A) loci, with 21 off-target hits, compared with 45 for stop-to-stop ORF searches; whole-contig HMMER searches recovered 108 (83.1%). With 11 distantly related queries, PSI-BLAST and DELTA-BLAST recovered all reference loci, whereas RPSTBLASTN recovered 98.5-99.2% and TBLASTN, as used by BITACORA, recovered 94.6-95.4%. Across six mite genomes, GeneFamHunter recovered all 209 previously reported C1 loci and identified 37 additional candidate loci. In a separate comparison, it recovered all 128 reference UniProt C1A proteins and identified five additional loci without annotated gene models; OrthoFinder included 35 non-C1A proteins, whereas OMA recovered 98 reference proteins. Availability and implementation: GeneFamHunter is implemented in Bash for Linux and SLURM clusters and is freely available at https://github.itap.purdue.edu/HeLab/GeneFamHunter (MIT license). Benchmark code and data: https://github.itap.purdue.edu/HeLab/GeneFamHunter_benchmarks and Zenodo (DOI: 10.5281/zenodo.23744704).