Poster #35 - Brydon Wall(2)

Genetic Ancestry Inference in PDX Models: Benchmarking WGS and RNA-seq Approaches

Brydon P. G. Wall, MS, Department of Biostatistics, SOPH, VCU Katarzyna M. Tyc, PhD, Department of Biostatistics, SOPH, VCU; Massey BISR Amy L. Olex, PhD, Wright Center for Clinical and Translational Research, VCU My Nguyen, BS, Department of Biostatistics, SOPH, VCU Jinze Liu, PhD, Department of Biostatistics, SOPH, VCU; Massey BISR J. Chuck Harrell, PhD, Department of Pathology, SOM, VCU; Massey BISR Mikhail G. Dozmorov, PhD, Department of Biostatistics, SOPH, VCU

Patient-derived xenograft (PDX) mouse models are widely used in precision oncology for diagnostic, prognostic, and treatment predictions. Self-identified race and ethnicity (SIRE) is often considered for interventions but is not objective and may misrepresent tumor biology. Whole genome sequencing (WGS) is commonly used for genetic ancestry profiling, though the impact of mouse DNA contamination in PDX models is unclear. RNA-seq is more common in PDX studies, yet its utility for ancestry prediction remains uncertain. We compared ancestry inference using WGS and RNA-seq from PDX models to identify tools and methods suitable for continental ancestry inference. We analyzed PDX samples from multiple cancer types using WGS and RNA-seq. Human and mouse reads were separated with Xengsort, then aligned and variant-called using both GATK and the JAX ancestry pipelines. Six inference tools / pipelines (ADMIXTURE, AEon, EthSEQ, gnomAD, RAIDS, SNPweights) were applied and benchmarked against SIRE metadata. SNP yield, overlap with ancestry-informative markers (AIMs), and concordance between sequencing modalities were evaluated. RNA-seq generated on average >100x fewer SNPs than WGS after GATK processing, and the JAX pipeline detected ~20x fewer RNA-seq and ~27x fewer WGS SNPs than GATK. SNP overlap with tool-specific models varied, with some tools' models accommodating overlapping sample RNA-seq SNPs better than others. WGS-based ancestry predictions were concordant across tools, though mismatches with SIRE highlighted admixture and sociocultural influences. RNA-seq predictions were accurate with Admixture, EthSEQ, and JAX, but less so with others. In summary, we establish a systematic framework for genetic ancestry inference in PDX models. WGS provides consistent ancestry calls across the tools tested, while RNA-seq performance depends on SNP yield, model overlap, and tool choice. Incorporating objective ancestry metrics into preclinical models reduces reliance on SIRE and promotes more accurate precision oncology research.