Disentangling genetic feature selection and aggregation in transcriptome-wide association studies

Abstract

The success of transcriptome-wide association studies (TWAS) has led to substantial research towards improving the predictive accuracy of its core component of Genetically Regulated eXpression (GReX). GReX links expression information with genotype and phenotype by playing two roles simultaneously: it acts as both the outcome of the genotype-based predictive models (for predicting expressions) and the linear combination of genotypes (as the predicted expressions) for association tests. From the perspective of machine learning (considering SNPs as features), these are actually two separable steps—feature selection and feature aggregation—which can be independently conducted. In this work, we show that the single approach of GReX limits the adaptability of TWAS methodology and practice. By conducting simulations and real data analysis, we demonstrate that disentangled protocols adapting straightforward approaches for feature selection (e.g., simple marker test) and aggregation (e.g., kernel machines) outperform the standard TWAS protocols that rely on GReX. Our development provides more powerful novel tools for conducting TWAS. More importantly, our characterization of the exact nature of TWAS suggests that, instead of questionably binding two distinct steps into the same statistical form (GReX), methodological research focusing on optimal combinations of feature selection and aggregation approaches will bring higher power to TWAS protocols.

IBAS: Interaction-bridged association studies discovering novel genes underlying complex traits

“…Here, the key insight is that we used the interaction patterns to methodologically mediate the discovery, despite the real biological causality relationship is unknown. The same issue applies to many existing works leveraging multi-scale omics to identify association between genetics and phenotype (Gamazon et al , 2015; Zeng et al , 2017; Xie et al , 2017, 2016; Cao et al , 2021b, 2022; Brandes et al , 2020; Okada et al , 2016; Xu et al , 2017).…”

“…This test combines the effect of multiple variants in a kernel-based score test to test phenotypic association of the aggregate: where y is the vector of phenotypes and K is a kernel function based on the centralized genotype matrix of the GWAS dataset, G . While multiple kernels may be defined for the use in the score test (Wu et al , 2010), we use the modified kernel previously used in our work (Cao et al , 2021b, 2022): where W =diag( α 1 ,…, α i ,…, α p ). α i here, denotes the coefficients of association derived from Equation 4 for the reference data.…”

“…The above unique insight into TWAS has opened a door of conducting association mapping mediated by any “data-bridge” by replacing the gene expression and its predicted value with another entity for feature selection and a form of feature aggregation (Cao et al , 2022). Indeed, another development in our group has provided a useful tool leveraging brain images (He et al , 2023) in a similar manner.…”

“…More specifically, starting from a gene expression matrix of all genes in a prespecified pathway, we extract representative statistics such as PCA (Hotelling, 1933), t-SNE (der Maaten & Hinton, 2008) or UMAP (McInnes et al , 2018) to form the objective function for feature selection. Then, following our previous work (Cao et al , 2021b, 2022; He et al , 2022), a kernel machine (Wu et al , 2010) is employed for feature aggregation aimed at association test. The outcomes are single genes whose genetic variants alter the variation of gene-gene interactions in the focal pathway, which also impact the phenotypic changes.…”

“…Transcriptome-Wide Association Study (TWAS) is a popular model to conduct association mapping utilizing transcriptomes (Gamazon et al, 2015;Gusev et al, 2016). Although it only focuses on single genes, the in-depth characterization of TWAS by our group (Cao et al, 2021a(Cao et al, , 2022 offered a unique insight of developing models incorporating interactions. The mainstream format of TWAS is a two-step protocol: first, one can form an expression prediction model using (usually cis) genotypes in a reference dataset (e.g.…”