Data-adaptive gene and pathway-based tests forrare-variant associations with survival outcomes

Abstract

Statistical methods for testing aggregate rare-variant genetic associations are typically based on either burden or dispersion tests (or a combination of the two). These methods lack statistical power in the presence of diverse genetic architectures. Moreover, few aggregate rare-variant association methods have been developed specifically for survival data. To address these issues, we propose data-adaptive gene- and pathway-based association tests based on Schoenfeld residuals in Cox proportional hazards models for association studies between an aggregate of rare-variants and survival outcomes. Our methods improve statistical power while maintaining flexibility across various genetic effect sizes and directions. We develop an efficient R package that enables fast computation and supports data simulation as well as gene- and pathway-level testing. Applying our approach to late bladder toxicity following radiotherapy for non-metastatic prostate cancer, we identify biologically relevant genes and pathways, replicate known signals, and capture additional associations. Our method provides a powerful, adaptive framework for survival-based genetic association studies of rare-variants. Keywords: aSPU, time-to-event outcomes, rare-variant associations, Cox regression, Schoenfeld residuals

Figures (6)

• Figure 1: Illustration of $\bm Z^s$ permutation
• Figure 2: The framework of aSPUS. The top-left stacked tables represent permutations of the variant matrix $\bm Z^s$; the shaded one is observed. The weight matrix $\Omega$ is calculated based on observed times and status for each subject in $\bm Z^m$, remaining constant throughout. For each pair of $(\gamma, \gamma_G)$, combining $\Omega$ with each permutation of $\bm Z^s$ in the score function yields the test statistic $u$. Aggregating $u$ from all permutations produces the corresponding $p$-value. Enumerating all pairs of $(\gamma, \gamma_G)$ finds the minimal $p$-value $p_{min}$ for each permutation. Finally, we compare $p_{min}$ values from permutations and the observation, treating $p_{aSPUS}$ as the final $p$-value for the gene or pathway of interest.
• Figure 3: Simulation results for gene- and pathway-based tests comparing three methods. a Power versus effect size $\beta\in(0,0.6]$ for combinations of 1, 3, and 5 causal SNPs in genes containing 10, 50, or 100 SNPs under a correlated SNP structure. Both aSPUS and the Burden test gain power as effect size increases, while aSPUS performs slightly better when genes contain more SNPs. b Power versus effect size $\beta\in(0,0.6]$ for pathways consisting of 20 genes with 5, 10, or 15 causal genes and 10 or 50 SNPs per gene. In both settings, aSPUS demonstrates a consistent advantage over the other methods. c--d Computational benchmarks based on 10 simulation replicates with 2,000 subjects and 80 SNPs in gene-based tests. Although aSPUS requires more CPU time, its memory usage is comparable to the other parametric methods.
• Figure 4: Summary of Data Analysis. a and b show QQ plots at gene- and pathway-levels. In gene-level tests, all methods except CopulaFLM have observed and expected $p$-values well-aligned along the diagonal. c shows detected gene-call overlap among the five methods, with FDR < 0.1. d Decomposes genes with per-variant adjusted effects for detected genes. e Summarizes the top 5 genes detected by at least one of the five methods with $\alpha = 0.05$. The central heatmap is colored by correlation between variant dosage. Genes are grouped and colored with their variants, ordered by $Z$ score from fitting Cox regression with the variant alone. Left panel columns indicate whether each method calls this gene.
• Figure 5: Pathway-call overlap among aSPUS, Burden, and SKAT, with the filter raw $p$-value < 0.05