Evaluating Kinship Estimation Methods for Reduced-Representation SNP Data in Non-model Species

Accurate kinship estimation between close relatives is crucial in conservation and restoration but remains challenging in wild populations due to structure and inbreeding. The efficacy of kinship inference using reduced-representation sequencing data (e.g., DArTseq, RADseq) is also uncertain. We evaluated the sensitivity and precision of six kinship methods (Goudet's beta dosage, KING Homo, KING Robust, PC-Relate, PLINK, RelateAdmix) at detecting parent-offspring and sibling relationships. Analyses were conducted on 3395 individuals and 363 families from six non-model Australian plant species: Acacia terminalis, Acacia suaveolens, Banksia serrata, Banksia aemula, Hakea sericea and Hakea teretifolia. Method performance varied across species and filtering parameters. Goudet's beta dosage and RelateAdmix performed well in low-structure, noninbred species but were less reliable in structured or inbred contexts. PLINK offered a balance of sensitivity and precision but was sensitive to filtering and often underestimated relatedness. KING Robust was highly precise but missed many true relatives. PC-Relate showed high false positives and is not recommended for similar applications. We recommend PLINK for general use, Goudet's beta dosage and RelateAdmix for low-structure species, and KING Robust for high-precision needs. Comparing multiple methods is advisable, as each has different assumptions and complementary strengths. Further theoretical development is needed for species with high inbreeding.