PCA outperforms popular hidden variable inference methods for molecular QTL mapping

Abstract

Background Estimating and accounting for hidden variables is widely practiced as an important step in molecular quantitative trait locus (molecular QTL, henceforth “QTL”) analysis for improving the power of QTL identification. However, few benchmark studies have been performed to evaluate the efficacy of the various methods developed for this purpose. Results Here we benchmark popular hidden variable inference methods including surrogate variable analysis (SVA), probabilistic estimation of expression residuals (PEER), and hidden covariates with prior (HCP) against principal component analysis (PCA)—a well-established dimension reduction and factor discovery method—via 362 synthetic and 110 real data sets. We show that PCA not only underlies the statistical methodology behind the popular methods but is also orders of magnitude faster, better-performing, and much easier to interpret and use. Conclusions To help researchers use PCA in their QTL analysis, we provide an R package along with a detailed guide, both of which are freely available at https://github.com/heatherjzhou/PCAForQTL. We believe that using PCA rather than SVA, PEER, or HCP will substantially improve and simplify hidden variable inference in QTL mapping as well as increase the transparency and reproducibility of QTL research.

Pitfalls and opportunities for applying latent variables in single-cell eQTL analyses

“…To balance the discovery power and potential false positives, there are different methods to determine the optimal number of factors. Two commonly used methods are the automatic elbow detection method and Buja and Eyuboglu (BE) algorithm, which has been comprehensively evaluated for PCA in bulk RNA data [ 16 ]. We ran these two methods for PCs inferred from the single-cell data for each QC option (see the “ Methods ” section).…”

“…These findings suggest that either PCs or PFs for single-cell eQTL mapping can be used to improve the number of eGene discoveries. However, the computational burden and flexibility are different between these two methods [ 13 , 16 ].…”

“…The PFs and PCs are generated with HVG2000 using QC option #12 as an example when testing the algorithm. We also implemented the automatic elbow detection method and Buja and Eyuboglu (BE) algorithm [ 16 ] to identify the optimal number of PCs for all the QC options as a comparison. The parameters used in BE algorithm are as default (B = 20, alpha = 0.05).…”

“…Some studies have run sensitivity tests [ 7 , 8 , 10 ], but such optimization has not been systematically evaluated for single-cell data at the population-scale level. A recent study [ 16 ] has evaluated the automatic elbow detection method and Buja and Eyuboglu (BE) algorithm for PCA in bulk eQTL analysis. However, such an investigation has not been conducted for single-cell pseudo-bulk data.…”

Disentangling genetic effects on transcriptional and post-transcriptional gene regulation through integrating exon and intron expression QTLs

“…We used the genotypes' first three principal components (PCs) as covariates, and the gene expression's PCs to account for technical and other co-variabilities in the RNA-Seq data, as suggested before 36,37 . We optimised the number of PCs used as covariates by maximising the number of genes with significant (FDR <5%) cis-QTLs, i.e., the first 70 PCs for exQTLs, the first 50 PCs for inQTLs, and the first 40 PCs for ex-inQTLs.…”