Abstract

Whole-genome sequencing (WGS) offers a comprehensive view of genetic variation in human disease, but translating sequencing data into reliable biological and clinical insight depends on both the deployment of appropriate computational tools and the availability of rigorous benchmarking resources to evaluate them. This dissertation addresses both requirements through two complementary lines of work. In a cohort of 788 patients with idiopathic peripheral neuropathy (IPN) from the Peripheral Neuropathy Research Registry, a WGS-based pipeline integrating ExpansionHunter Denovo and ExpansionHunter with unsupervised genotype clustering identified biallelic RFC1 AAGGG repeat expansions in 2.3% of all IPN cases and 6.9% of patients with pure sensory neuropathy, with 97% concordance against repeat-primed PCR validation. These findings establish RFC1 as a clinically meaningful and previously underappreciated contributor to IPN and demonstrate that scalable short-read WGS workflows can achieve diagnostic sensitivity comparable to targeted PCR assays. To address the broader absence of benchmarking infrastructure for somatic variant detection outside of cancer contexts, six HapMap cell lines were mixed at defined proportions to generate a synthetic sample spanning a designed variant allele fraction range of 0.25–16.5%, sequenced to 500x short-read and 100x long-read coverage. A pangenome graph alignment framework using Minigraph-Cactus was used to construct a benchmarking variant set comprising over 6 million SNVs, 1.8 million indels, 51,000 structural variants, and 10,000 mobile element insertions. Systematic evaluation of six somatic variant callers revealed substantial performance variability across caller, variant type, and genomic context, with marked degradation at low allele fractions and in repeat-rich regions. Per-gene sequencing coverage recommendations for SNV detection at 1% VAF are provided for 4,612 clinically relevant genes. Taken together, this work demonstrates that closing the gap between what WGS can detect in principle and what current tools detect in practice requires both targeted deployment of specialized variant detection methods and the deliberate construction of shared benchmarking resources. The pipelines and benchmarking datasets generated here are publicly available and have been adopted by multiple downstream studies within the SMaHT Network and beyond.

Committee Chair

Sheng Chih Jin

Committee Members

Ahmet Hoke; Danny E. Miller; Dennis Goldfarb; Tim Schedl; Ting Wang

Degree

Doctor of Philosophy (PhD)

Author's Department

Biology and Biomedical Sciences

Author's School

Graduate School of Arts and Sciences

Document Type

Dissertation

Date of Award

8-11-2026

Language

English (en)

Download

Share

COinS